fix ctrl shell

README.md

@@ -1,12 +0,0 @@
-iot
-=====
-
-An OTP application
-
-## erlang client sdk
-    https://github.com/emqx/emqtt
-
-Build
------
-
-    $ rebar3 compile

apps/endpoint/include/endpoint.hrl

@@ -10,6 +10,7 @@
 
 -record(http_endpoint, {
     url = <<>> :: binary(),
+    token = <<>> :: binary(),
     pool_size = 10 :: integer()
 }).
 
@@ -33,7 +34,7 @@
 -record(endpoint, {
     id :: integer(),
     %% 全局唯一，在路由规则中通过名称来指定
-    name :: binary(),
+    matcher :: binary(),
     %% 标题描述
     title = <<>> :: binary(),
     %% 配置项, 格式: #{<<"protocol">> => <<"http|https|ws|kafka|mqtt">>, <<"args">> => #{}}

apps/endpoint/src/endpoint.app.src

@@ -0,0 +1,24 @@
+{application, endpoint,
+ [{description, "Endpoint OTP application"},
+  {vsn, "0.1.0"},
+  {registered, [
+    endpoint_sup,
+    endpoint_adapter_sup,
+    endpoint_subscription,
+    endpoint_log
+  ]},
+  {mod, {endpoint_app, []}},
+  {applications, [
+    emqtt,
+    brod,
+    hackney,
+    gproc,
+    crypto,
+    kernel,
+    stdlib
+  ]},
+  {env, []},
+  {modules, []},
+  {licenses, ["Apache 2.0"]},
+  {links, []}
+ ]}.

apps/endpoint/src/endpoint.erl

@@ -0,0 +1,359 @@
+
+%%%-------------------------------------------------------------------
+%%% @author aresei
+%%% @copyright (C) 2023, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 06. 7月 2023 12:02
+%%%-------------------------------------------------------------------
+-module(endpoint).
+-include("endpoint.hrl").
+
+%% API
+-export([start_link/1]).
+-export([get_name/1, get_pid/1, forward/2, reload/2, clean_up/1]).
+-export([get_alias_pid/1, is_support/1, get_protocol/1]).
+-export([endpoint_record/1, parse_config/2]).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec start_link(Endpoint :: #endpoint{}) -> {'ok', pid()} | 'ignore' | {'error', term()}.
+start_link(Endpoint = #endpoint{id = Id, config = #http_endpoint{}}) ->
+    LocalName = get_name(Id),
+    endpoint_http:start_link(LocalName, Endpoint);
+start_link(Endpoint = #endpoint{id = Id, config = #mqtt_endpoint{}}) ->
+    LocalName = get_name(Id),
+    endpoint_mqtt:start_link(LocalName, Endpoint);
+start_link(Endpoint = #endpoint{id = Id, config = #kafka_endpoint{}}) ->
+    LocalName = get_name(Id),
+    endpoint_kafka:start_link(LocalName, Endpoint).
+
+-spec get_name(Id :: integer()) -> term().
+get_name(Id) when is_integer(Id) ->
+    {endpoint, Id}.
+
+-spec get_pid(Id :: integer()) -> undefined | pid().
+get_pid(Id) when is_integer(Id) ->
+    gproc:whereis_name({n, l, get_name(Id)}).
+
+-spec get_alias_name(Name :: binary()) -> term().
+get_alias_name(Name) when is_binary(Name) ->
+    {endpoint_alias, Name}.
+
+-spec get_alias_pid(Name :: binary()) -> undefined | pid().
+get_alias_pid(Name) when is_binary(Name) ->
+    gproc:whereis_name({n, l, get_alias_name(Name)}).
+
+-spec forward(Pid :: pid(), Metric :: binary()) -> ok.
+forward(Pid, Metric) when is_pid(Pid), is_binary(Metric) ->
+    gen_server:cast(Pid, {forward, Metric}).
+
+reload(Pid, NEndpoint = #endpoint{}) when is_pid(Pid) ->
+    gen_server:cast(Pid, {reload, NEndpoint}).
+
+-spec clean_up(Pid :: pid()) -> ok.
+clean_up(Pid) when is_pid(Pid) ->
+    gen_server:cast(Pid, cleanup).
+
+-spec get_protocol(Endpoint :: #endpoint{}) -> atom().
+get_protocol(#endpoint{config = #http_endpoint{}}) ->
+    http;
+get_protocol(#endpoint{config = #mqtt_endpoint{}}) ->
+    mqtt;
+get_protocol(#endpoint{config = #kafka_endpoint{}}) ->
+    kafka.
+
+-spec is_support(Protocol :: atom()) -> boolean().
+is_support(Protocol) when is_atom(Protocol) ->
+    {ok, Props} = application:get_env(endpoint, endpoints),
+    SupportProtocols = proplists:get_value(support_protocols, Props, []),
+    lists:member(Protocol, SupportProtocols).
+
+-spec endpoint_record(EndpointInfo :: #{}) -> error | {ok, Endpoint :: #endpoint{}}.
+endpoint_record(#{<<"id">> := Id, <<"matcher">> := Matcher, <<"title">> := Title, <<"type">> := Type, <<"config">> := ConfigJson,
+    <<"status">> := Status, <<"updated_at">> := UpdatedAt, <<"created_at">> := CreatedAt}) ->
+    case parse_config(Type, ConfigJson) of
+        {ok, Config} ->
+            {ok, #endpoint {
+                id = Id,
+                matcher = Matcher,
+                title = Title,
+                config = Config,
+                status = Status,
+                updated_at = UpdatedAt,
+                created_at = CreatedAt
+            }};
+        {error, _Reason} ->
+            error
+    end.
+
+-spec parse_config(Protocol :: binary(), Config :: map()) -> {ok, #mqtt_endpoint{} | #kafka_endpoint{} | #http_endpoint{}} | {error, Errors :: [Error :: binary()]}.
+parse_config(<<"mqtt">>, #{<<"host">> := Host, <<"port">> := Port0, <<"client_id">> := ClientId, <<"username">> := Username, <<"password">> := Password, <<"topic">> := Topic, <<"qos">> := Qos}) ->
+    {Port, PortErrors} = case parse_integer(Port0) of
+        {ok, ParsedPort} ->
+            {ParsedPort, []};
+        error ->
+            {undefined, [<<"port invalid">>]}
+    end,
+
+    CheckTerms = case Port of
+        undefined ->
+            [{host, Host}, {username, Username}, {password, Password}, {topic, Topic}, {qos, Qos}];
+        _ ->
+            [{host, Host}, {port, Port}, {username, Username}, {password, Password}, {topic, Topic}, {qos, Qos}]
+    end,
+    Errors = PortErrors ++ lists:filtermap(fun(Term) ->
+        case check_mqtt_argument(Term) of
+            ok ->
+                false;
+            {error, Error} ->
+                {true, Error}
+        end
+                             end, CheckTerms),
+    case Errors =:= [] of
+        true ->
+            {ok, #mqtt_endpoint{
+                host = Host,
+                port = Port,
+                client_id = ClientId,
+                username = Username,
+                password = Password,
+                topic = Topic,
+                qos = Qos
+            }};
+        false ->
+            {error, Errors}
+    end;
+parse_config(<<"http">>, C = #{<<"url">> := Url, <<"pool_size">> := PoolSize}) ->
+    Token = maps:get(<<"token">>, C, <<>>),
+    Errors = lists:filtermap(fun(Term) ->
+        case check_http_argument(Term) of
+            ok ->
+                false;
+            {error, Error} ->
+                {true, Error}
+        end
+                             end, [{url, Url}, {token, Token}, {pool_size, PoolSize}]),
+    case Errors =:= [] of
+        true ->
+            {ok, #http_endpoint{
+                url = Url,
+                token = Token,
+                pool_size = PoolSize
+            }};
+        false ->
+            {error, Errors}
+    end;
+parse_config(<<"kafka">>, #{<<"sasl_config">> := #{<<"username">> := Username, <<"password">> := Password, <<"mechanism">> := Mechanism0}, <<"bootstrap_servers">> := BootstrapServers, <<"topic">> := Topic}) ->
+    Errors = lists:filtermap(fun(Term) ->
+        case check_kafka_argument(Term) of
+            ok ->
+                false;
+            {error, Error} ->
+                {true, Error}
+        end
+                             end, [{username, Username}, {password, Password}, {topic, Topic}, {mechanism, Mechanism0}, {bootstrap_servers, BootstrapServers}]),
+    case Errors =:= [] of
+        true ->
+            Mechanism = parse_kafka_mechanism(Mechanism0),
+            {ok, #kafka_endpoint{
+                sasl_config = {Mechanism, Username, Password},
+                bootstrap_servers = parse_kafka_bootstrap_servers(BootstrapServers),
+                topic = Topic
+            }};
+        false ->
+            {error, Errors}
+    end;
+parse_config(<<"kafka">>, #{<<"bootstrap_servers">> := BootstrapServers, <<"topic">> := Topic}) ->
+    Errors = lists:filtermap(fun(Term) ->
+        case check_kafka_argument(Term) of
+            ok ->
+                false;
+            {error, Error} ->
+                {true, Error}
+        end
+                             end, [{topic, Topic}, {bootstrap_servers, BootstrapServers}]),
+    case Errors =:= [] of
+        true ->
+            {ok, #kafka_endpoint{
+                sasl_config = undefined,
+                bootstrap_servers = parse_kafka_bootstrap_servers(BootstrapServers),
+                topic = Topic
+            }};
+        false ->
+            {error, Errors}
+    end;
+parse_config(_, _) ->
+    {error, invalid_config}.
+
+-spec parse_kafka_bootstrap_servers(BootstrapServers :: [binary()]) -> Servers :: [{Host :: string(), Port :: integer()}].
+parse_kafka_bootstrap_servers(BootstrapServers) when is_list(BootstrapServers) ->
+    lists:map(fun(S) ->
+        {ok, Host, Port} = parse_kafka_bootstrap_server(S),
+        {Host, Port}
+              end, BootstrapServers).
+
+-spec parse_kafka_mechanism(Mechanism0 :: binary()) -> atom().
+parse_kafka_mechanism(Mechanism0) when is_binary(Mechanism0) ->
+    case Mechanism0 of
+        <<"sha_256">> ->
+            scram_sha_256;
+        <<"sha_512">> ->
+            scram_sha_512;
+        <<"plain">> ->
+            plain;
+        _ ->
+            plain
+    end.
+
+-spec check_kafka_argument(tuple()) -> ok | {error, Reason :: binary()}.
+check_kafka_argument({mechanism, Mechanism}) ->
+    case lists:member(Mechanism, [<<"sha_256">>, <<"sha_512">>, <<"plain">>]) of
+        true ->
+            ok;
+        false ->
+            {error, <<"mechanism invalid">>}
+    end;
+check_kafka_argument({username, Username}) ->
+    case Username /= <<>> of
+        true ->
+            ok;
+        false ->
+            {error, <<"username is empty">>}
+    end;
+check_kafka_argument({password, Password}) ->
+    case Password /= <<>> of
+        true ->
+            ok;
+        false ->
+            {error, <<"password is empty">>}
+    end;
+check_kafka_argument({topic, Topic}) ->
+    case Topic /= <<>> of
+        true ->
+            ok;
+        false ->
+            {error, <<"topic is empty">>}
+    end;
+check_kafka_argument({bootstrap_servers, BootstrapServers}) ->
+    case is_list(BootstrapServers) andalso length(BootstrapServers) > 0 of
+        true ->
+            InvalidServers = lists:filtermap(fun(S) ->
+                case parse_kafka_bootstrap_server(S) of
+                    {ok, _Host, _Port} ->
+                        false;
+                    error ->
+                        {true, S}
+                end
+                                             end, BootstrapServers),
+
+            case InvalidServers =:= [] of
+                true ->
+                    ok;
+                false ->
+                    {error, iolist_to_binary([<<"bootstrap_servers: ">>, lists:join(<<",">>, InvalidServers), <<" format is error">>])}
+            end;
+        false ->
+            {error, <<"bootstrap_servers is empty">>}
+    end.
+
+-spec check_http_argument(tuple()) -> ok | {error, Reason :: binary()}.
+check_http_argument({url, Url}) ->
+    case is_binary(Url) andalso Url /= <<>> of
+        true ->
+            ok;
+        false ->
+            {error, <<"url is empty">>}
+    end;
+check_http_argument({token, Token}) ->
+    case is_binary(Token) of
+        true ->
+            ok;
+        false ->
+            {error, <<"token invalid">>}
+    end;
+check_http_argument({pool_size, PoolSize}) ->
+    case is_integer(PoolSize) andalso PoolSize > 0 of
+        true ->
+            ok;
+        false ->
+            {error, <<"pool_size invalid">>}
+    end.
+
+-spec check_mqtt_argument(tuple()) -> ok | {error, Reason :: binary()}.
+check_mqtt_argument({host, Host}) ->
+    case Host /= <<>> of
+        true ->
+            ok;
+        false ->
+            {error, <<"host is empty">>}
+    end;
+check_mqtt_argument({port, Port}) ->
+    case is_integer(Port) andalso Port > 0 of
+        true ->
+            ok;
+        false ->
+            {error, <<"port invalid">>}
+    end;
+check_mqtt_argument({username, Username}) ->
+    case Username /= <<>> of
+        true ->
+            ok;
+        false ->
+            {error, <<"username is empty">>}
+    end;
+check_mqtt_argument({password, Password}) ->
+    case Password /= <<>> of
+        true ->
+            ok;
+        false ->
+            {error, <<"password is empty">>}
+    end;
+check_mqtt_argument({topic, Topic}) ->
+    case Topic /= <<>> of
+        true ->
+            ok;
+        false ->
+            {error, <<"topic is empty">>}
+    end;
+check_mqtt_argument({qos, Qos}) ->
+    case is_integer(Qos) andalso lists:member(Qos, [0, 1, 2]) of
+        true ->
+            ok;
+        false ->
+            {error, <<"qos invalid">>}
+    end.
+
+-spec parse_integer(term()) -> {ok, integer()} | error.
+parse_integer(Value) when is_integer(Value) ->
+    {ok, Value};
+parse_integer(Value) when is_binary(Value) ->
+    try binary_to_integer(Value) of
+        Int ->
+            {ok, Int}
+    catch
+        _:_ ->
+            error
+    end;
+parse_integer(_) ->
+    error.
+
+-spec parse_kafka_bootstrap_server(term()) -> {ok, string(), integer()} | error.
+parse_kafka_bootstrap_server(Server) when is_binary(Server) ->
+    case binary:split(Server, <<":">>) of
+        [Host0, Port0] ->
+            Host = binary_to_list(Host0),
+            case parse_integer(Port0) of
+                {ok, Port} when Host =/= [], Port > 0 ->
+                    {ok, Host, Port};
+                _ ->
+                    error
+            end;
+        _ ->
+            error
+    end;
+parse_kafka_bootstrap_server(_) ->
+    error.

apps/endpoint/src/endpoint_adapter_sup.erl

@@ -0,0 +1,83 @@
+%%%-------------------------------------------------------------------
+%% @doc endpoint top level supervisor.
+%% @end
+%%%-------------------------------------------------------------------
+
+-module(endpoint_adapter_sup).
+
+-behaviour(supervisor).
+-include("endpoint.hrl").
+
+-export([start_link/0]).
+-export([ensured_endpoint_started/1, delete_endpoint/1]).
+-export([start_endpoints/1]).
+
+-export([init/1]).
+
+-define(SERVER, ?MODULE).
+
+start_link() ->
+    supervisor:start_link({local, ?SERVER}, ?MODULE, []).
+
+%% sup_flags() = #{strategy => strategy(),         % optional
+%%                 intensity => non_neg_integer(), % optional
+%%                 period => pos_integer()}        % optional
+%% child_spec() = #{id => child_id(),       % mandatory
+%%                  start => mfargs(),      % mandatory
+%%                  restart => restart(),   % optional
+%%                  shutdown => shutdown(), % optional
+%%                  type => worker(),       % optional
+%%                  modules => modules()}   % optional
+init([]) ->
+    SupFlags = #{strategy => one_for_one, intensity => 1000, period => 3600},
+    {ok, {SupFlags, []}}.
+
+-spec start_endpoints(Endpoints :: list()) -> ok.
+start_endpoints(Endpoints) when is_list(Endpoints) ->
+    lists:foreach(fun(Endpoint) ->
+        Spec = child_spec(Endpoint),
+        {ok, _} = supervisor:start_child(?MODULE, Spec)
+                  end, Endpoints),
+    ok.
+
+-spec ensured_endpoint_started(Endpoint :: #endpoint{}) -> {ok, Pid :: pid()} | {error, Reason :: any()}.
+ensured_endpoint_started(Endpoint = #endpoint{}) ->
+    case supervisor:start_child(?MODULE, child_spec(Endpoint)) of
+        {ok, Pid} when is_pid(Pid) ->
+            {ok, Pid};
+        {error, {'already_started', Pid}} when is_pid(Pid) ->
+            {ok, Pid};
+        {error, Error} ->
+            {error, Error}
+    end.
+
+-spec delete_endpoint(Id :: integer()) -> ok | {error, Reason :: any()}.
+delete_endpoint(Id) when is_integer(Id) ->
+    Name = endpoint:get_name(Id),
+    case supervisor:terminate_child(?MODULE, Name) of
+        ok ->
+            delete_endpoint_child(Name);
+        {error, not_found} ->
+            delete_endpoint_child(Name);
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+delete_endpoint_child(Name) ->
+    case supervisor:delete_child(?MODULE, Name) of
+        ok ->
+            ok;
+        {error, not_found} ->
+            ok;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+child_spec(Endpoint = #endpoint{id = Id}) ->
+    Name = endpoint:get_name(Id),
+    #{id => Name,
+        start => {endpoint, start_link, [Endpoint]},
+        restart => permanent,
+        shutdown => 2000,
+        type => worker,
+        modules => ['endpoint']}.

apps/endpoint/src/endpoint_app.erl

@@ -0,0 +1,15 @@
+%%%-------------------------------------------------------------------
+%% @doc endpoint application entrypoint.
+%% @end
+%%%-------------------------------------------------------------------
+-module(endpoint_app).
+
+-behaviour(application).
+
+-export([start/2, stop/1]).
+
+start(_StartType, _StartArgs) ->
+    endpoint_sup:start_link().
+
+stop(_State) ->
+    ok.

apps/endpoint/src/endpoint_buffer.erl

@@ -0,0 +1,304 @@
+
+%%%-------------------------------------------------------------------
+%%% @author aresei
+%%% @copyright (C) 2023, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 06. 7月 2023 12:02
+%%%-------------------------------------------------------------------
+-module(endpoint_buffer).
+
+-include("endpoint.hrl").
+
+%% 消息重发间隔
+-define(RETRY_INTERVAL, 5000).
+%% 最大重试次数，不包含首次发送
+-define(MAX_RETRY_TIMES, 3).
+%% 与 endpoint_outbox 的单条记录限制保持一致。
+-define(MAX_PAYLOAD_BYTES, 32 * 1024).
+
+-export([new/2, append/2, append_only/2, trigger_next/1, trigger_n/1, handle_timeout/4, cleanup/1, recover_inflight/1, ack/2, stat/1, resize/2]).
+-export_type([buffer/0]).
+
+-type flight_source() :: outbox | memory.
+-type timer_entry() :: {reference(), binary(), non_neg_integer(), flight_source()}.
+
+-record(buffer, {
+    endpoint :: #endpoint{},
+    outbox :: endpoint_outbox:outbox(),
+    %% 当前待 ack 的数据及其重试定时器 #{Id => {TimerRef, Payload, RetryTimes, Source}}
+    timer_map = #{} :: #{integer() => timer_entry()},
+    %% 窗口大小，允许最大的未确认消息数
+    window_size = 10,
+    %% 未确认的消息数
+    flight_num = 0,
+    %% 记录成功处理的消息数
+    acc_num = 0
+}).
+
+-type buffer() :: #buffer{}.
+
+-spec new(Endpoint :: #endpoint{}, WindowSize :: integer()) -> Buffer :: #buffer{}.
+new(Endpoint = #endpoint{id = Id}, WindowSize) when is_integer(WindowSize), WindowSize > 0 ->
+    %% 读取配置文件
+    {ok, Endpoints} = application:get_env(endpoint, endpoints),
+    RootDir = proplists:get_value(root_dir, Endpoints),
+    OutboxDir = filename:join(RootDir, integer_to_list(Id)),
+
+    {ok, Outbox} = endpoint_outbox:open(OutboxDir, #{}),
+    #buffer{outbox = Outbox, endpoint = Endpoint, window_size = WindowSize}.
+
+-spec append(Payload :: binary(), Buffer :: #buffer{}) -> NBuffer :: #buffer{}.
+append(Payload, Buffer = #buffer{}) when is_binary(Payload) ->
+    case validate_payload_size(Payload, Buffer) of
+        ok ->
+            %% Persist before dispatch so a VM crash does not lose in-flight data.
+            case append_to_outbox(Payload, Buffer) of
+                {ok, NBuffer} ->
+                    trigger_next(NBuffer);
+                {dropped, NBuffer} ->
+                    trigger_next(NBuffer);
+                {error, NBuffer} ->
+                    NBuffer
+            end;
+        error ->
+            Buffer
+    end.
+
+-spec append_only(Payload :: binary(), Buffer :: #buffer{}) -> NBuffer :: #buffer{}.
+append_only(Payload, Buffer = #buffer{}) when is_binary(Payload) ->
+    case validate_payload_size(Payload, Buffer) of
+        ok ->
+            case append_to_outbox(Payload, Buffer) of
+                {ok, NBuffer} ->
+                    NBuffer;
+                {dropped, NBuffer} ->
+                    NBuffer;
+                {error, NBuffer} ->
+                    NBuffer
+            end;
+        error ->
+            Buffer
+    end.
+
+-spec trigger_n(Buffer :: #buffer{}) -> NBuffer :: #buffer{}.
+trigger_n(Buffer = #buffer{window_size = WindowSize}) ->
+    %% 最多允许window_size
+    lists:foldl(fun(_, Buffer0) -> trigger_next(Buffer0) end, Buffer, lists:seq(1, WindowSize)).
+
+%% 触发读取下一条数据
+-spec trigger_next(Buffer :: #buffer{}) -> NBuffer :: #buffer{}.
+trigger_next(Buffer = #buffer{outbox = Outbox, flight_num = FlightNum, window_size = WindowSize}) ->
+    case FlightNum < WindowSize of
+        false ->
+            Buffer;
+        true ->
+            case endpoint_outbox:next(Outbox) of
+                eof ->
+                    Buffer;
+                {ok, Id, Payload, NOutbox} ->
+                    ReceiverPid = self(),
+                    ReceiverPid ! {next_data, Id, Payload},
+                    schedule_retry(Id, Payload, 0, outbox, Buffer#buffer{outbox = NOutbox, flight_num = FlightNum + 1});
+                {error, Reason} ->
+                    logger:warning("[endpoint_buffer] read next outbox failed, endpoint_id: ~p, reason: ~p",
+                        [buffer_endpoint_id(Buffer), Reason]),
+                    Buffer
+            end
+    end.
+
+-spec handle_timeout(reference(), integer(), binary(), #buffer{}) -> #buffer{}.
+handle_timeout(TimerRef, Id, _Payload, Buffer = #buffer{timer_map = TimerMap})
+        when is_reference(TimerRef), is_integer(Id) ->
+    case maps:take(Id, TimerMap) of
+        {{TimerRef, Payload, RetryTimes, Source}, NTimerMap} when RetryTimes < ?MAX_RETRY_TIMES ->
+            logger:warning("[endpoint_buffer] retry message, endpoint_id: ~p, id: ~p, retry: ~p/~p",
+                [buffer_endpoint_id(Buffer), Id, RetryTimes + 1, ?MAX_RETRY_TIMES]),
+            ReceiverPid = self(),
+            ReceiverPid ! {next_data, Id, Payload},
+            schedule_retry(Id, Payload, RetryTimes + 1, Source, Buffer#buffer{timer_map = NTimerMap});
+        {{TimerRef, Payload, RetryTimes, outbox}, NTimerMap} ->
+            logger:warning("[endpoint_buffer] drop message after retries exhausted, endpoint_id: ~p, id: ~p, retries: ~p",
+                [buffer_endpoint_id(Buffer), Id, RetryTimes]),
+            drop_outbox_message(Id, Payload, RetryTimes, Buffer#buffer{timer_map = NTimerMap});
+        {{TimerRef, _, RetryTimes, memory}, NTimerMap} ->
+            logger:warning("[endpoint_buffer] drop memory message after retries exhausted, endpoint_id: ~p, id: ~p, retries: ~p",
+                [buffer_endpoint_id(Buffer), Id, RetryTimes]),
+            drop_memory_message(Buffer#buffer{timer_map = NTimerMap});
+        {{_OtherTimerRef, _OtherPayload, _RetryTimes, _Source}, _NTimerMap} ->
+            Buffer;
+        error ->
+            Buffer
+    end.
+
+-spec ack(Id :: integer(), Buffer :: #buffer{}) -> NBuffer :: #buffer{}.
+ack(Id, Buffer = #buffer{timer_map = TimerMap, outbox = Outbox, acc_num = AccNum, flight_num = FlightNum}) when is_integer(Id) ->
+    case maps:take(Id, TimerMap) of
+        {{TimerRef, _Payload, _RetryTimes, memory}, NTimerMap} ->
+            _ = erlang:cancel_timer(TimerRef),
+            NBuffer = Buffer#buffer{
+                timer_map = NTimerMap,
+                acc_num = AccNum + 1,
+                flight_num = max(FlightNum - 1, 0)
+            },
+            trigger_next(NBuffer);
+        {{TimerRef, Payload, RetryTimes, outbox}, NTimerMap} ->
+            _ = erlang:cancel_timer(TimerRef),
+            case endpoint_outbox:ack(Id, Outbox) of
+                {ok, NOutbox} ->
+                    NBuffer = Buffer#buffer{
+                        timer_map = NTimerMap,
+                        outbox = NOutbox,
+                        acc_num = AccNum + 1,
+                        flight_num = max(FlightNum - 1, 0)
+                    },
+                    trigger_next(NBuffer);
+                {error, Reason} ->
+                    logger:warning("[endpoint_buffer] ack outbox failed, endpoint_id: ~p, id: ~p, reason: ~p",
+                        [buffer_endpoint_id(Buffer), Id, Reason]),
+                    schedule_retry(Id, Payload, RetryTimes, outbox, Buffer#buffer{timer_map = NTimerMap})
+            end;
+        error ->
+            Buffer
+    end.
+
+%% 获取当前统计信息
+-spec stat(Buffer :: #buffer{}) -> map().
+stat(#buffer{acc_num = AccNum, outbox = Outbox, flight_num = FlightNum, timer_map = TimerMap}) ->
+    OutboxStat = endpoint_outbox:stat(Outbox),
+    WriteSeq = maps:get(write_seq, OutboxStat, 0),
+    AckedSeq = maps:get(acked_seq, OutboxStat, 0),
+    OutboxFlightNum = count_inflight(outbox, TimerMap),
+    MemoryFlightNum = count_inflight(memory, TimerMap),
+    QueueNum = max(WriteSeq - AckedSeq - OutboxFlightNum, 0),
+    OutboxStat#{
+        <<"acc_num">> => AccNum,
+        <<"queue_num">> => QueueNum,
+        <<"inflight_num">> => FlightNum,
+        <<"outbox_inflight_num">> => OutboxFlightNum,
+        <<"memory_inflight_num">> => MemoryFlightNum
+    }.
+
+-spec cleanup(Buffer :: #buffer{}) -> #buffer{}.
+cleanup(Buffer = #buffer{timer_map = TimerMap}) ->
+    cancel_timers(TimerMap),
+    NBuffer0 = persist_memory_inflight(Buffer),
+    reset_reader_after_recover(NBuffer0#buffer{timer_map = #{}, flight_num = 0}).
+
+-spec recover_inflight(Buffer :: #buffer{}) -> #buffer{}.
+recover_inflight(Buffer = #buffer{timer_map = TimerMap}) ->
+    cancel_timers(TimerMap),
+    NBuffer0 = persist_memory_inflight(Buffer),
+    case endpoint_outbox:reset_reader(NBuffer0#buffer.outbox) of
+        {ok, NOutbox} ->
+            NBuffer0#buffer{outbox = NOutbox, timer_map = #{}, flight_num = 0}
+    end.
+
+-spec resize(Buffer :: #buffer{}, WindowSize :: integer()) -> #buffer{}.
+resize(Buffer = #buffer{}, WindowSize) when is_integer(WindowSize), WindowSize > 0 ->
+    trigger_n(Buffer#buffer{window_size = WindowSize}).
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec buffer_endpoint_id(buffer()) -> integer().
+buffer_endpoint_id(#buffer{endpoint = #endpoint{id = Id}}) ->
+    Id.
+
+-spec validate_payload_size(binary(), buffer()) -> ok | error.
+validate_payload_size(Payload, Buffer) ->
+    case byte_size(Payload) =< ?MAX_PAYLOAD_BYTES of
+        true ->
+            ok;
+        false ->
+            logger:warning("[endpoint_buffer] payload too large, endpoint_id: ~p, size: ~p, max_size: ~p",
+                [buffer_endpoint_id(Buffer), byte_size(Payload), ?MAX_PAYLOAD_BYTES]),
+            error
+    end.
+
+-spec append_to_outbox(binary(), buffer()) -> {ok | dropped | error, buffer()}.
+append_to_outbox(Payload, Buffer = #buffer{outbox = Outbox}) ->
+    case endpoint_outbox:append(Payload, Outbox) of
+        {ok, _Seq, NOutbox} ->
+            {ok, Buffer#buffer{outbox = NOutbox}};
+        {dropped, capacity_reached, NOutbox} ->
+            logger:warning("[endpoint_buffer] outbox capacity reached, endpoint_id: ~p", [buffer_endpoint_id(Buffer)]),
+            {dropped, Buffer#buffer{outbox = NOutbox}};
+        {error, Reason} ->
+            logger:warning("[endpoint_buffer] append outbox failed, endpoint_id: ~p, reason: ~p",
+                [buffer_endpoint_id(Buffer), Reason]),
+            {error, Buffer}
+    end.
+
+-spec schedule_retry(integer(), binary(), non_neg_integer(), flight_source(), buffer()) -> buffer().
+schedule_retry(Id, Payload, RetryTimes, Source, Buffer = #buffer{timer_map = TimerMap})
+        when is_integer(Id), is_binary(Payload), is_integer(RetryTimes), RetryTimes >= 0 ->
+    TimerRef = erlang:start_timer(?RETRY_INTERVAL, self(), {endpoint_buffer_retry, Id, Payload}),
+    Buffer#buffer{timer_map = maps:put(Id, {TimerRef, Payload, RetryTimes, Source}, TimerMap)}.
+
+-spec cancel_timers(#{integer() => timer_entry()}) -> ok.
+cancel_timers(TimerMap) ->
+    lists:foreach(fun({_Id, {TimerRef, _Payload, _RetryTimes, _Source}}) ->
+        _ = erlang:cancel_timer(TimerRef),
+        ok
+    end, maps:to_list(TimerMap)),
+    ok.
+
+-spec drop_outbox_message(integer(), binary(), non_neg_integer(), buffer()) -> buffer().
+drop_outbox_message(Id, Payload, RetryTimes, Buffer = #buffer{outbox = Outbox, acc_num = AccNum, flight_num = FlightNum}) ->
+    case endpoint_outbox:ack(Id, Outbox) of
+        {ok, NOutbox} ->
+            trigger_next(Buffer#buffer{
+                outbox = NOutbox,
+                acc_num = AccNum + 1,
+                flight_num = max(FlightNum - 1, 0)
+            });
+        {error, Reason} ->
+            logger:warning("[endpoint_buffer] ack dropped message failed, endpoint_id: ~p, id: ~p, reason: ~p",
+                [buffer_endpoint_id(Buffer), Id, Reason]),
+            schedule_retry(Id, Payload, RetryTimes, outbox, Buffer)
+    end.
+
+-spec drop_memory_message(buffer()) -> buffer().
+drop_memory_message(Buffer = #buffer{acc_num = AccNum, flight_num = FlightNum}) ->
+    trigger_next(Buffer#buffer{
+        acc_num = AccNum + 1,
+        flight_num = max(FlightNum - 1, 0)
+    }).
+
+-spec persist_memory_inflight(buffer()) -> buffer().
+persist_memory_inflight(Buffer = #buffer{timer_map = TimerMap}) ->
+    MemoryInflight = lists:sort(
+        fun({IdA, _PayloadA}, {IdB, _PayloadB}) -> IdA > IdB end,
+        [{Id, Payload} || {Id, {_TimerRef, Payload, _RetryTimes, memory}} <- maps:to_list(TimerMap)]
+    ),
+    lists:foldl(fun({_Id, Payload}, AccBuffer) ->
+        case append_to_outbox(Payload, AccBuffer) of
+            {ok, NBuffer} ->
+                NBuffer;
+            {dropped, NBuffer} ->
+                NBuffer;
+            {error, NBuffer} ->
+                NBuffer
+        end
+    end, Buffer, MemoryInflight).
+
+-spec reset_reader_after_recover(buffer()) -> buffer().
+reset_reader_after_recover(Buffer = #buffer{outbox = Outbox}) ->
+    case endpoint_outbox:reset_reader(Outbox) of
+        {ok, NOutbox} ->
+            Buffer#buffer{outbox = NOutbox}
+    end.
+
+-spec count_inflight(flight_source(), #{integer() => timer_entry()}) -> non_neg_integer().
+count_inflight(Source, TimerMap) ->
+    maps:fold(fun(_Id, {_TimerRef, _Payload, _RetryTimes, EntrySource}, Acc) ->
+        case EntrySource =:= Source of
+            true ->
+                Acc + 1;
+            false ->
+                Acc
+        end
+    end, 0, TimerMap).

apps/endpoint/src/endpoint_http.erl

@@ -13,7 +13,7 @@
 -behaviour(gen_server).
 
 %% API
--export([start_link/3]).
+-export([start_link/2]).
 
 %% gen_server callbacks
 -export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
@@ -30,10 +30,10 @@
 %%%===================================================================
 
 %% @doc Spawns the server and registers the local name (unique)
--spec(start_link(LocalName :: atom(), AliasName :: atom(), Endpoint :: #endpoint{}) ->
+-spec(start_link(LocalName :: term(), Endpoint :: #endpoint{}) ->
     {ok, Pid :: pid()} | ignore | {error, Reason :: term()}).
-start_link(LocalName, AliasName, Endpoint = #endpoint{config = #http_endpoint{}}) when is_atom(LocalName), is_atom(AliasName) ->
-    gen_server:start_link({local, LocalName}, ?MODULE, [AliasName, Endpoint], []).
+start_link(LocalName, Endpoint = #endpoint{config = #http_endpoint{}}) ->
+    gen_server:start_link({via, gproc, {n, l, LocalName}}, ?MODULE, [Endpoint], []).
 
 %%%===================================================================
 %%% gen_server callbacks
@@ -44,9 +44,10 @@ start_link(LocalName, AliasName, Endpoint = #endpoint{config = #http_endpoint{}}
 -spec(init(Args :: term()) ->
     {ok, State :: #state{}} | {ok, State :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term()} | ignore).
-init([AliasName, Endpoint]) ->
-    Buffer = endpoint_buffer:new(Endpoint, 10),
-    true = gproc:reg({n, l, AliasName}),
+init([Endpoint = #endpoint{matcher = Matcher, config = #http_endpoint{pool_size = PoolSize}}]) ->
+    ok = endpoint_util:set_metadata(),
+    endpoint_subscription:subscribe(Matcher, self()),
+    Buffer = endpoint_buffer:new(Endpoint, PoolSize),
     {ok, #state{endpoint = Endpoint, buffer = Buffer}}.
 
 %% @private
@@ -69,13 +70,17 @@ handle_call(get_stat, _From, State = #state{buffer = Buffer}) ->
     {noreply, NewState :: #state{}} |
     {noreply, NewState :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term(), NewState :: #state{}}).
-handle_cast({forward, ServiceId, Metric}, State = #state{buffer = Buffer}) ->
-    NBuffer = endpoint_buffer:append({ServiceId, Metric}, Buffer),
+handle_cast({forward, Metric}, State = #state{buffer = Buffer}) ->
+    NBuffer = endpoint_buffer:append(Metric, Buffer),
     {noreply, State#state{buffer = NBuffer}};
+handle_cast({reload, NEndpoint = #endpoint{matcher = NMatcher, config = #http_endpoint{pool_size = PoolSize}}}, State = #state{endpoint = #endpoint{matcher = Matcher}, buffer = Buffer}) ->
+    ensure_subscription(Matcher, NMatcher),
+    NBuffer = endpoint_buffer:resize(Buffer, PoolSize),
+    {noreply, State#state{endpoint = NEndpoint, buffer = NBuffer}};
 
 handle_cast(cleanup, State = #state{buffer = Buffer}) ->
-    endpoint_buffer:cleanup(Buffer),
-    {noreply, State}.
+    NBuffer = endpoint_buffer:cleanup(Buffer),
+    {noreply, State#state{buffer = NBuffer}}.
 
 %% @private
 %% @doc Handling all non call/cast messages
@@ -83,28 +88,25 @@ handle_cast(cleanup, State = #state{buffer = Buffer}) ->
     {noreply, NewState :: #state{}} |
     {noreply, NewState :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term(), NewState :: #state{}}).
-handle_info({next_data, Id, {ServiceId, Metric}}, State = #state{buffer = Buffer, endpoint = #endpoint{config = #http_endpoint{url = Url}}}) ->
-    Headers = [
-        {<<"Content-Type">>, <<"application/octet-stream">>},
-        {<<"Service-Id">>, ServiceId}
-    ],
+handle_info({next_data, Id, Metric}, State = #state{buffer = Buffer, endpoint = #endpoint{config = #http_endpoint{url = Url, token = Token}}}) ->
+    Headers = [{<<"Content-Type">>, <<"application/json">>}] ++ patch_headers(Metric, Token),
     case hackney:request(post, Url, Headers, Metric) of
-        {ok, 200, _, ClientRef} ->
-            {ok, RespBody} = hackney:body(ClientRef),
-            hackney:close(ClientRef),
-            lager:debug("[endpoint_http] url: ~p, response is: ~p", [Url, RespBody]),
+        {ok, HttpCode, _, ClientRef} when HttpCode >= 200, HttpCode < 300 ->
+            RespBody = read_response_body(ClientRef),
+            logger:debug("[endpoint_http] url: ~p, response is: ~p", [Url, RespBody]),
             NBuffer = endpoint_buffer:ack(Id, Buffer),
-
             {noreply, State#state{buffer = NBuffer}};
         {ok, HttpCode, _, ClientRef} ->
-            {ok, RespBody} = hackney:body(ClientRef),
-            hackney:close(ClientRef),
-            lager:debug("[endpoint_http] url: ~p, http_code: ~p, response is: ~p", [Url, HttpCode, RespBody]),
+            RespBody = read_response_body(ClientRef),
+            logger:warning("[endpoint_http] url: ~p, http_code: ~p, response is: ~p", [Url, HttpCode, RespBody]),
             {noreply, State};
         {error, Reason} ->
-            lager:warning("[endpoint_http] url: ~p, get error: ~p", [Url, Reason]),
+            logger:warning("[endpoint_http] url: ~p, get error: ~p", [Url, Reason]),
             {noreply, State}
-    end.
+    end;
+handle_info({timeout, TimerRef, {endpoint_buffer_retry, Id, Payload}}, State = #state{buffer = Buffer}) ->
+    NBuffer = endpoint_buffer:handle_timeout(TimerRef, Id, Payload, Buffer),
+    {noreply, State#state{buffer = NBuffer}}.
 
 %% @private
 %% @doc This function is called by a gen_server when it is about to
@@ -113,7 +115,8 @@ handle_info({next_data, Id, {ServiceId, Metric}}, State = #state{buffer = Buffer
 %% with Reason. The return value is ignored.
 -spec(terminate(Reason :: (normal | shutdown | {shutdown, term()} | term()),
     State :: #state{}) -> term()).
-terminate(_Reason, _State = #state{}) ->
+terminate(_Reason, #state{buffer = Buffer}) ->
+    _ = endpoint_buffer:cleanup(Buffer),
     ok.
 
 %% @private
@@ -127,3 +130,28 @@ code_change(_OldVsn, State = #state{}, _Extra) ->
 %%%===================================================================
 %%% Internal functions
 %%%===================================================================
+
+-spec ensure_subscription(binary(), binary()) -> ok.
+ensure_subscription(Matcher, Matcher) ->
+    ok;
+ensure_subscription(Matcher, NMatcher) ->
+    ok = endpoint_subscription:unsubscribe(Matcher, self()),
+    endpoint_subscription:subscribe(NMatcher, self()).
+
+-spec read_response_body(reference()) -> binary() | term().
+read_response_body(ClientRef) ->
+    case hackney:body(ClientRef) of
+        {ok, RespBody} ->
+            hackney:close(ClientRef),
+            RespBody;
+        {error, Reason} ->
+            hackney:close(ClientRef),
+            Reason
+    end.
+
+-spec patch_headers(Metric :: binary(), any()) -> list().
+patch_headers(Metric, Token) when is_binary(Token), Token /= <<>> ->
+    Sign = endpoint_util:sha256(erlang:iolist_to_binary([Token, Metric, Token])),
+    [{<<"X-Signature">>, Sign}];
+patch_headers(_, _) ->
+    [].

apps/endpoint/src/endpoint_kafka.erl

@@ -0,0 +1,261 @@
+%%%-------------------------------------------------------------------
+%%% @author aresei
+%%% @copyright (C) 2023, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 06. 7月 2023 12:02
+%%%-------------------------------------------------------------------
+-module(endpoint_kafka).
+
+-include("endpoint.hrl").
+-behaviour(gen_statem).
+
+%% API
+-export([start_link/2]).
+
+%% gen_statem callbacks
+-export([callback_mode/0, init/1, terminate/3, code_change/4]).
+-export([disconnected/3, connected/3]).
+
+%% 消息重发间隔
+-define(RETRY_INTERVAL, 5000).
+
+-record(state, {
+    endpoint :: #endpoint{},
+    buffer :: endpoint_buffer:buffer(),
+    client_id :: atom(),
+    client_pid :: undefined | pid()
+}).
+
+-type kafka_state() :: disconnected | connected.
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec start_link(LocalName :: term(), Endpoint :: #endpoint{}) ->
+    {ok, pid()} | ignore | {error, term()}.
+start_link(LocalName, Endpoint = #endpoint{}) ->
+    gen_statem:start_link({via, gproc, {n, l, LocalName}}, ?MODULE, [Endpoint], []).
+
+%%%===================================================================
+%%% gen_statem callbacks
+%%%===================================================================
+
+-spec callback_mode() -> state_functions.
+callback_mode() ->
+    state_functions.
+
+-spec init(term()) -> gen_statem:init_result(kafka_state(), #state{}).
+init([Endpoint = #endpoint{id = Id, matcher = Matcher}]) ->
+    ok = endpoint_util:set_metadata(),
+    erlang:process_flag(trap_exit, true),
+    ok = endpoint_subscription:subscribe(Matcher, self()),
+    Buffer = endpoint_buffer:new(Endpoint, 10),
+    ClientId = list_to_atom("brod_client:" ++ integer_to_list(Id)),
+    {ok, disconnected, #state{endpoint = Endpoint, buffer = Buffer, client_id = ClientId},
+     [{state_timeout, 0, connect}]}.
+
+-spec disconnected(gen_statem:event_type(), term(), #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+disconnected({call, From}, get_stat, State = #state{buffer = Buffer}) ->
+    reply_stat(From, Buffer, State);
+disconnected(cast, {forward, Metric}, State = #state{buffer = Buffer}) ->
+    store_metric(Metric, Buffer, State);
+disconnected(cast, cleanup, State = #state{buffer = Buffer}) ->
+    cleanup_buffer(Buffer, State);
+disconnected(cast, {reload, NEndpoint = #endpoint{matcher = NMatcher}},
+             State = #state{endpoint = #endpoint{matcher = Matcher}, client_id = ClientId}) ->
+    reload_endpoint(Matcher, NMatcher, ClientId, NEndpoint, State);
+disconnected(state_timeout, connect, State) ->
+    try_connect(State);
+disconnected(info, {timeout, TimerRef, {endpoint_buffer_retry, Id, Payload}}, State = #state{buffer = Buffer}) ->
+    NBuffer = endpoint_buffer:handle_timeout(TimerRef, Id, Payload, Buffer),
+    {keep_state, State#state{buffer = NBuffer}};
+disconnected(info, {next_data, _Id, _Tuple}, State) ->
+    {keep_state, State};
+disconnected(info, {ack, Id}, State = #state{buffer = Buffer}) ->
+    ack_buffer(Id, Buffer, State);
+disconnected(info, Info, State) ->
+    unknown_info(Info, disconnected, State);
+disconnected(EventType, EventContent, State) ->
+    unknown_event(EventType, EventContent, disconnected, State).
+
+-spec connected(gen_statem:event_type(), term(), #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+connected({call, From}, get_stat, State = #state{buffer = Buffer}) ->
+    reply_stat(From, Buffer, State);
+connected(cast, {forward, Metric}, State = #state{buffer = Buffer}) ->
+    forward_metric(Metric, Buffer, State);
+connected(cast, cleanup, State = #state{buffer = Buffer}) ->
+    cleanup_buffer(Buffer, State);
+connected(cast, {reload, NEndpoint = #endpoint{matcher = NMatcher}},
+          State = #state{endpoint = #endpoint{matcher = Matcher}, client_id = ClientId}) ->
+    reload_endpoint(Matcher, NMatcher, ClientId, NEndpoint, State);
+connected(info, {next_data, Id, Metric},
+          State = #state{
+              buffer = Buffer,
+              client_pid = ClientPid,
+              client_id = ClientId,
+              endpoint = #endpoint{config = #kafka_endpoint{topic = Topic}}
+          }) ->
+    ReceiverPid = self(),
+    AckCb = fun(Partition, BaseOffset) ->
+        logger:debug("[endpoint_kafka] ack partion: ~p, offset: ~p", [Partition, BaseOffset]),
+        ReceiverPid ! {ack, Id}
+    end,
+    case catch brod:produce_cb(ClientPid, Topic, random, <<>>, Metric, AckCb) of
+        {ok, _CallRef} ->
+            {keep_state, State};
+        {ok, _CallRef, _ProducerPid} ->
+            {keep_state, State};
+        {error, Reason} ->
+            logger:warning("[endpoint_kafka] produce topic: ~p, get error: ~p", [Topic, Reason]),
+            stop_kafka_client(ClientId),
+            NBuffer = endpoint_buffer:recover_inflight(Buffer),
+            {next_state, disconnected, State#state{client_pid = undefined, buffer = NBuffer},
+             [{state_timeout, ?RETRY_INTERVAL, connect}]};
+        {'EXIT', Reason} ->
+            logger:warning("[endpoint_kafka] produce topic: ~p, exit with reason: ~p", [Topic, Reason]),
+            stop_kafka_client(ClientId),
+            NBuffer = endpoint_buffer:recover_inflight(Buffer),
+            {next_state, disconnected, State#state{client_pid = undefined, buffer = NBuffer},
+             [{state_timeout, ?RETRY_INTERVAL, connect}]}
+    end;
+connected(info, {timeout, TimerRef, {endpoint_buffer_retry, Id, Payload}}, State = #state{buffer = Buffer}) ->
+    NBuffer = endpoint_buffer:handle_timeout(TimerRef, Id, Payload, Buffer),
+    {keep_state, State#state{buffer = NBuffer}};
+connected(info, {ack, Id}, State = #state{buffer = Buffer}) ->
+    ack_buffer(Id, Buffer, State);
+connected(info, {'EXIT', ClientPid, Reason},
+          State = #state{client_pid = ClientPid, endpoint = #endpoint{title = Title}, buffer = Buffer}) ->
+    logger:warning("[endpoint_kafka] endpoint: ~p, conn pid exit with reason: ~p", [Title, Reason]),
+    NBuffer = endpoint_buffer:recover_inflight(Buffer),
+    {next_state, disconnected, State#state{client_pid = undefined, buffer = NBuffer},
+     [{state_timeout, ?RETRY_INTERVAL, connect}]};
+connected(info, Info, State) ->
+    unknown_info(Info, connected, State);
+connected(EventType, EventContent, State) ->
+    unknown_event(EventType, EventContent, connected, State).
+
+-spec terminate(term(), kafka_state(), #state{}) -> term().
+terminate(Reason, _StateName, #state{endpoint = #endpoint{title = Title}, buffer = Buffer, client_id = ClientId}) ->
+    logger:debug("[endpoint_kafka] endpoint: ~p, terminate with reason: ~p", [Title, Reason]),
+    stop_kafka_client(ClientId),
+    endpoint_buffer:cleanup(Buffer),
+    ok.
+
+-spec code_change(term() | {down, term()}, kafka_state(), #state{}, term()) ->
+    {ok, kafka_state(), #state{}} | {error, term()}.
+code_change(_OldVsn, StateName, State = #state{}, _Extra) ->
+    {ok, StateName, State}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec reply_stat(gen_statem:from(), endpoint_buffer:buffer(), #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+reply_stat(From, Buffer, State) ->
+    Stat = endpoint_buffer:stat(Buffer),
+    {keep_state, State, [{reply, From, {ok, Stat}}]}.
+
+-spec forward_metric(binary(), endpoint_buffer:buffer(), #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+forward_metric(Metric, Buffer, State) ->
+    NBuffer = endpoint_buffer:append(Metric, Buffer),
+    {keep_state, State#state{buffer = NBuffer}}.
+
+-spec store_metric(binary(), endpoint_buffer:buffer(), #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+store_metric(Metric, Buffer, State) ->
+    NBuffer = endpoint_buffer:append_only(Metric, Buffer),
+    {keep_state, State#state{buffer = NBuffer}}.
+
+-spec cleanup_buffer(endpoint_buffer:buffer(), #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+cleanup_buffer(Buffer, _State) ->
+    NBuffer = endpoint_buffer:cleanup(Buffer),
+    {keep_state, _State#state{buffer = NBuffer}}.
+
+-spec ack_buffer(integer(), endpoint_buffer:buffer(), #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+ack_buffer(Id, Buffer, State) ->
+    NBuffer = endpoint_buffer:ack(Id, Buffer),
+    {keep_state, State#state{buffer = NBuffer}}.
+
+-spec reload_endpoint(binary(), binary(), atom(), #endpoint{}, #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+reload_endpoint(Matcher, NMatcher, ClientId, NEndpoint, State = #state{}) ->
+    ensure_subscription(Matcher, NMatcher),
+    stop_kafka_client(ClientId),
+    NBuffer = endpoint_buffer:recover_inflight(State#state.buffer),
+    {next_state, disconnected, State#state{endpoint = NEndpoint, client_pid = undefined, buffer = NBuffer},
+     [{state_timeout, 0, connect}]}.
+
+-spec try_connect(#state{}) -> gen_statem:event_handler_result(kafka_state(), #state{}).
+try_connect(State = #state{
+                buffer = Buffer,
+                client_id = ClientId,
+                endpoint = #endpoint{
+                    title = Title,
+                    config = #kafka_endpoint{
+                        sasl_config = SaslConfig,
+                        bootstrap_servers = BootstrapServers,
+                        topic = Topic
+                    }
+                }
+            }) ->
+    logger:debug("[endpoint_kafka] endpoint: ~p, create postman", [Title]),
+    BaseConfig = [
+        {reconnect_cool_down_seconds, 5},
+        {socket_options, [{keepalive, true}]}
+    ],
+    ClientConfig = case SaslConfig of
+                       {Mechanism, Username, Password} ->
+                           [{sasl, {Mechanism, Username, Password}} | BaseConfig];
+                       undefined ->
+                           BaseConfig
+                   end,
+    case catch brod:start_link_client(BootstrapServers, ClientId, ClientConfig) of
+        {ok, ClientPid} ->
+            case brod:start_producer(ClientId, Topic, []) of
+                ok ->
+                    NBuffer = endpoint_buffer:trigger_n(Buffer),
+                    {next_state, connected, State#state{buffer = NBuffer, client_pid = ClientPid}};
+                {error, Reason} ->
+                    logger:debug("[endpoint_kafka] start_producer: ~p, get error: ~p", [ClientId, Reason]),
+                    stop_kafka_client(ClientId),
+                    {keep_state, State#state{client_pid = undefined},
+                     [{state_timeout, ?RETRY_INTERVAL, connect}]}
+            end;
+        Error ->
+            logger:debug("[endpoint_kafka] start_client: ~p, get error: ~p", [ClientId, Error]),
+            {keep_state, State#state{client_pid = undefined},
+             [{state_timeout, ?RETRY_INTERVAL, connect}]}
+    end.
+
+-spec unknown_info(term(), kafka_state(), #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+unknown_info(Info, StateName, State) ->
+    logger:warning("[endpoint_kafka] unknown message: ~p, status: ~p", [Info, StateName]),
+    {keep_state, State}.
+
+-spec unknown_event(gen_statem:event_type(), term(), kafka_state(), #state{}) ->
+    gen_statem:event_handler_result(kafka_state(), #state{}).
+unknown_event(EventType, EventContent, StateName, State) ->
+    logger:warning("[endpoint_kafka] unknown event: ~p, content: ~p, status: ~p", [EventType, EventContent, StateName]),
+    {keep_state, State}.
+
+-spec ensure_subscription(binary(), binary()) -> ok.
+ensure_subscription(Matcher, Matcher) ->
+    ok;
+ensure_subscription(Matcher, NMatcher) ->
+    ok = endpoint_subscription:unsubscribe(Matcher, self()),
+    endpoint_subscription:subscribe(NMatcher, self()).
+
+-spec stop_kafka_client(atom()) -> ok.
+stop_kafka_client(ClientId) when is_atom(ClientId) ->
+    _ = catch brod:stop_client(ClientId),
+    ok.

apps/endpoint/src/endpoint_log.erl

@@ -0,0 +1,191 @@
+%%%-------------------------------------------------------------------
+%%% @doc Endpoint diagnostic log.
+%%%
+%%% 当前用于记录 publish 时没有匹配到任何订阅者的数据，避免这类信息进入
+%%% endpoint_subscription 的热路径同步磁盘写入。
+%%%-------------------------------------------------------------------
+-module(endpoint_log).
+
+-behaviour(gen_server).
+
+%% API
+-export([start_link/0, unmatched_publish/2]).
+
+%% gen_server callbacks
+-export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
+
+-define(SERVER, ?MODULE).
+-define(LOG_NAME, endpoint_unmatched_publish_log).
+-define(LOG_FILE_NAME, "unmatched_publish.log").
+-define(DEFAULT_MAX_BYTES, 10485760).
+-define(DEFAULT_MAX_FILES, 10).
+-define(MAX_LOG_CONTENT_BYTES, 32 * 1024).
+
+-record(state, {
+    enabled = false :: boolean(),
+    log_name = ?LOG_NAME :: atom()
+}).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec start_link() -> {ok, pid()} | ignore | {error, term()}.
+start_link() ->
+    gen_server:start_link({local, ?SERVER}, ?MODULE, [], []).
+
+-spec unmatched_publish(RouteKey :: binary(), Content :: binary()) -> ok.
+unmatched_publish(RouteKey, Content) when is_binary(RouteKey), is_binary(Content) ->
+    case erlang:whereis(?SERVER) of
+        undefined ->
+            ok;
+        Pid when is_pid(Pid) ->
+            gen_server:cast(Pid, {unmatched_publish, RouteKey, Content}),
+            ok
+    end.
+
+%%%===================================================================
+%%% gen_server callbacks
+%%%===================================================================
+
+-spec init(term()) -> {ok, #state{}}.
+init([]) ->
+    ok = endpoint_util:set_metadata(),
+    case open_log() of
+        ok ->
+            {ok, #state{enabled = true}};
+        {error, Reason} ->
+            logger:warning("[endpoint_log] open disk_log failed, reason: ~p", [Reason]),
+            {ok, #state{enabled = false}}
+    end.
+
+-spec handle_call(term(), gen_server:from(), #state{}) ->
+    {reply, ok, #state{}}.
+handle_call(_Request, _From, State) ->
+    {reply, ok, State}.
+
+-spec handle_cast(term(), #state{}) -> {noreply, #state{}}.
+handle_cast({unmatched_publish, RouteKey, Content}, State = #state{enabled = true, log_name = LogName}) ->
+    Entry = encode_unmatched_publish(RouteKey, Content),
+    case disk_log:blog(LogName, Entry) of
+        ok ->
+            {noreply, State};
+        {error, Reason} ->
+            logger:warning("[endpoint_log] write unmatched publish log failed, reason: ~p", [Reason]),
+            {noreply, State#state{enabled = false}}
+    end;
+handle_cast({unmatched_publish, _RouteKey, _Content}, State) ->
+    {noreply, State};
+handle_cast(_Request, State) ->
+    {noreply, State}.
+
+-spec handle_info(term(), #state{}) -> {noreply, #state{}}.
+handle_info(Info, State) ->
+    logger:debug("[endpoint_log] unknown info: ~p", [Info]),
+    {noreply, State}.
+
+-spec terminate(term(), #state{}) -> ok.
+terminate(_Reason, #state{enabled = true, log_name = LogName}) ->
+    _ = disk_log:close(LogName),
+    ok;
+terminate(_Reason, _State) ->
+    ok.
+
+-spec code_change(term() | {down, term()}, #state{}, term()) ->
+    {ok, #state{}} | {error, term()}.
+code_change(_OldVsn, State = #state{}, _Extra) ->
+    {ok, State}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec open_log() -> ok | {error, term()}.
+open_log() ->
+    Config = log_config(),
+    case log_dir(Config) of
+        {ok, Path} ->
+            File = filename:join(Path, ?LOG_FILE_NAME),
+            ok = filelib:ensure_dir(File),
+            case disk_log:open([
+                {name, ?LOG_NAME},
+                {file, File},
+                {type, wrap},
+                {format, external},
+                {size, {config_pos_integer(max_bytes, Config, ?DEFAULT_MAX_BYTES),
+                        config_pos_integer(max_files, Config, ?DEFAULT_MAX_FILES)}},
+                {linkto, self()},
+                {repair, true}
+            ]) of
+                {ok, ?LOG_NAME} ->
+                    ok;
+                {repaired, ?LOG_NAME, _Recovered, _BadBytes} ->
+                    ok;
+                {error, Reason} ->
+                    {error, Reason}
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec log_config() -> proplists:proplist().
+log_config() ->
+    case application:get_env(endpoint, endpoint_log) of
+        {ok, Config} when is_list(Config) ->
+            Config;
+        _ ->
+            []
+    end.
+
+-spec log_dir(proplists:proplist()) -> {ok, file:filename_all()} | {error, term()}.
+log_dir(Config) ->
+    case proplists:get_value(path, Config) of
+        Path when is_list(Path); is_binary(Path) ->
+            {ok, Path};
+        _ ->
+            {error, {missing_config, endpoint_log, path}}
+    end.
+
+-spec config_pos_integer(atom(), proplists:proplist(), pos_integer()) -> pos_integer().
+config_pos_integer(Key, Config, Default) ->
+    case proplists:get_value(Key, Config, Default) of
+        Value when is_integer(Value), Value > 0 ->
+            Value;
+        _ ->
+            Default
+    end.
+
+-spec encode_unmatched_publish(binary(), binary()) -> binary().
+encode_unmatched_publish(RouteKey, Content) ->
+    {LoggedContent, Truncated} = maybe_truncate(Content),
+    iolist_to_binary([
+        format_local_time(), <<"\t">>,
+        <<"unmatched_publish">>, <<"\t">>,
+        RouteKey, <<"\t">>,
+        integer_to_binary(byte_size(Content)), <<"\t">>,
+        boolean_to_binary(Truncated), <<"\t">>,
+        base64:encode(LoggedContent), <<"\n">>
+    ]).
+
+-spec boolean_to_binary(boolean()) -> binary().
+boolean_to_binary(true) ->
+    <<"true">>;
+boolean_to_binary(false) ->
+    <<"false">>.
+
+-spec format_local_time() -> binary().
+format_local_time() ->
+    MilliSeconds = erlang:system_time(millisecond) rem 1000,
+    {{Year, Month, Day}, {Hour, Minute, Second}} = calendar:local_time(),
+    iolist_to_binary(io_lib:format(
+        "~4..0B-~2..0B-~2..0B ~2..0B:~2..0B:~2..0B.~3..0B",
+        [Year, Month, Day, Hour, Minute, Second, MilliSeconds]
+    )).
+
+-spec maybe_truncate(binary()) -> {binary(), boolean()}.
+maybe_truncate(Content) when byte_size(Content) =< ?MAX_LOG_CONTENT_BYTES ->
+    {Content, false};
+maybe_truncate(Content) ->
+    MaxBytes = ?MAX_LOG_CONTENT_BYTES,
+    <<LoggedContent:MaxBytes/binary, _/binary>> = Content,
+    {LoggedContent, true}.

apps/endpoint/src/endpoint_mqtt.erl

@@ -0,0 +1,285 @@
+%%%-------------------------------------------------------------------
+%%% @author aresei
+%%% @copyright (C) 2023, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 06. 7月 2023 12:02
+%%%-------------------------------------------------------------------
+-module(endpoint_mqtt).
+
+-include("endpoint.hrl").
+-behaviour(gen_statem).
+
+%% API
+-export([start_link/2]).
+
+%% gen_statem callbacks
+-export([callback_mode/0, init/1, terminate/3, code_change/4]).
+-export([disconnected/3, connected/3]).
+
+%% 消息重发间隔
+-define(RETRY_INTERVAL, 15000).
+
+-record(state, {
+    endpoint :: #endpoint{},
+    buffer :: endpoint_buffer:buffer(),
+    conn_pid :: undefined | pid(),
+    %% 待确认的数据, #{PacketId :: integer() => Id :: integer()}
+    inflight = #{}
+}).
+
+-type mqtt_state() :: disconnected | connected.
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec start_link(LocalName :: term(), Endpoint :: #endpoint{}) ->
+    {ok, pid()} | ignore | {error, term()}.
+start_link(LocalName, Endpoint = #endpoint{}) ->
+    gen_statem:start_link({via, gproc, {n, l, LocalName}}, ?MODULE, [Endpoint], []).
+
+%%%===================================================================
+%%% gen_statem callbacks
+%%%===================================================================
+
+-spec callback_mode() -> state_functions.
+callback_mode() ->
+    state_functions.
+
+-spec init(term()) -> gen_statem:init_result(mqtt_state(), #state{}).
+init([Endpoint = #endpoint{matcher = Matcher}]) ->
+    ok = endpoint_util:set_metadata(),
+    erlang:process_flag(trap_exit, true),
+    ok = endpoint_subscription:subscribe(Matcher, self()),
+    Buffer = endpoint_buffer:new(Endpoint, 10),
+    {ok, disconnected, #state{endpoint = Endpoint, buffer = Buffer}, [{next_event, internal, connect}]}.
+
+-spec disconnected(gen_statem:event_type(), term(), #state{}) ->
+    gen_statem:event_handler_result(mqtt_state(), #state{}).
+disconnected({call, From}, get_stat, State = #state{buffer = Buffer}) ->
+    reply_stat(From, Buffer, State);
+disconnected(cast, {forward, Metric}, State = #state{buffer = Buffer}) ->
+    store_metric(Metric, Buffer, State);
+disconnected(cast, cleanup, State = #state{buffer = Buffer}) ->
+    cleanup_buffer(Buffer, State);
+disconnected(cast, {reload, NEndpoint = #endpoint{matcher = NMatcher}},
+             State = #state{endpoint = #endpoint{matcher = Matcher}, conn_pid = ConnPid}) ->
+    reload_endpoint(Matcher, NMatcher, ConnPid, NEndpoint, State);
+disconnected(internal, connect, State) ->
+    {keep_state, State, [{next_event, internal, do_connect}]};
+disconnected(state_timeout, connect, State) ->
+    {keep_state, State, [{next_event, internal, do_connect}]};
+disconnected(internal, do_connect,
+             State = #state{
+                 buffer = Buffer,
+                 endpoint = #endpoint{
+                     title = Title,
+                     config = #mqtt_endpoint{
+                         host = Host,
+                         port = Port,
+                         username = Username,
+                         password = Password,
+                         client_id = ClientId
+                     }
+                 }
+             }) ->
+    logger:debug("[endpoint_mqtt] endpoint: ~ts, create postman", [Title]),
+    Opts = [
+        {owner, self()},
+        {clientid, ClientId},
+        {host, binary_to_list(Host)},
+        {port, Port},
+        {tcp_opts, []},
+        {username, binary_to_list(Username)},
+        {password, binary_to_list(Password)},
+        {keepalive, 86400},
+        {auto_ack, true},
+        {connect_timeout, 5000},
+        {proto_ver, v5},
+        {retry_interval, 5000}
+    ],
+    case connect_mqtt(Opts) of
+        {ok, ConnPid} ->
+            logger:debug("[endpoint_mqtt] connect success, pid: ~p", [ConnPid]),
+            NBuffer = endpoint_buffer:trigger_n(Buffer),
+            {next_state, connected, State#state{conn_pid = ConnPid, buffer = NBuffer}};
+        {error, Reason} ->
+            logger:warning("[endpoint_mqtt] connect get error: ~p", [Reason]),
+            {keep_state, State#state{conn_pid = undefined, inflight = #{}},
+             [{state_timeout, ?RETRY_INTERVAL, connect}]}
+    end;
+disconnected(info, {timeout, TimerRef, {endpoint_buffer_retry, Id, Payload}}, State = #state{buffer = Buffer}) ->
+    NBuffer = endpoint_buffer:handle_timeout(TimerRef, Id, Payload, Buffer),
+    {keep_state, State#state{buffer = NBuffer}};
+disconnected(info, {next_data, _Id, _Tuple}, State) ->
+    {keep_state, State};
+disconnected(info, {'EXIT', ConnPid, Reason},
+             State = #state{endpoint = #endpoint{title = Title}, conn_pid = ConnPid}) ->
+    logger:warning("[endpoint_mqtt] endpoint: ~p, conn pid exit with reason: ~p", [Title, Reason]),
+    {keep_state, State#state{conn_pid = undefined, inflight = #{}},
+     [{state_timeout, ?RETRY_INTERVAL, connect}]};
+disconnected(info, Info, State) ->
+    unknown_info(Info, disconnected, State);
+disconnected(EventType, EventContent, State) ->
+    unknown_event(EventType, EventContent, disconnected, State).
+
+-spec connected(gen_statem:event_type(), term(), #state{}) ->
+    gen_statem:event_handler_result(mqtt_state(), #state{}).
+connected({call, From}, get_stat, State = #state{buffer = Buffer}) ->
+    reply_stat(From, Buffer, State);
+connected(cast, {forward, Metric}, State = #state{buffer = Buffer}) ->
+    forward_metric(Metric, Buffer, State);
+connected(cast, cleanup, State = #state{buffer = Buffer}) ->
+    cleanup_buffer(Buffer, State);
+connected(cast, {reload, NEndpoint = #endpoint{matcher = NMatcher}},
+          State = #state{endpoint = #endpoint{matcher = Matcher}, conn_pid = ConnPid}) ->
+    reload_endpoint(Matcher, NMatcher, ConnPid, NEndpoint, State);
+connected(internal, connect, State) ->
+    {keep_state, State};
+connected(info, {next_data, Id, Metric},
+          State = #state{
+              conn_pid = ConnPid,
+              buffer = Buffer,
+              inflight = InFlight,
+              endpoint = #endpoint{config = #mqtt_endpoint{topic = Topic, qos = Qos}}
+          }) ->
+    logger:debug("[endpoint_mqtt] will publish topic: ~p, metric: ~p, qos: ~p", [Topic, Metric, Qos]),
+    case emqtt:publish(ConnPid, Topic, #{}, Metric, [{qos, Qos}, {retain, true}]) of
+        ok ->
+            NBuffer = endpoint_buffer:ack(Id, Buffer),
+            {keep_state, State#state{buffer = NBuffer}};
+        {ok, PacketId} ->
+            {keep_state, State#state{inflight = maps:put(PacketId, Id, InFlight)}};
+        {error, Reason} ->
+            logger:warning("[endpoint_mqtt] send message to topic: ~p, get error: ~p", [Topic, Reason]),
+            stop_mqtt_conn(ConnPid),
+            NBuffer = endpoint_buffer:recover_inflight(Buffer),
+            {next_state, disconnected, State#state{conn_pid = undefined, inflight = #{}, buffer = NBuffer},
+             [{state_timeout, ?RETRY_INTERVAL, connect}]}
+    end;
+connected(info, {timeout, TimerRef, {endpoint_buffer_retry, Id, Payload}}, State = #state{buffer = Buffer}) ->
+    NBuffer = endpoint_buffer:handle_timeout(TimerRef, Id, Payload, Buffer),
+    {keep_state, State#state{buffer = NBuffer}};
+connected(info, {disconnected, ReasonCode, Properties},
+          State = #state{conn_pid = ConnPid, buffer = Buffer}) ->
+    logger:debug("[endpoint_mqtt] Recv a DISONNECT packet - ReasonCode: ~p, Properties: ~p", [ReasonCode, Properties]),
+    stop_mqtt_conn(ConnPid),
+    NBuffer = endpoint_buffer:recover_inflight(Buffer),
+    {next_state, disconnected, State#state{conn_pid = undefined, inflight = #{}, buffer = NBuffer},
+     [{state_timeout, ?RETRY_INTERVAL, connect}]};
+connected(info, {publish, Message = #{packet_id := _PacketId, payload := Payload}}, State) ->
+    logger:debug("[endpoint_mqtt] Recv a publish packet: ~p, payload: ~p", [Message, Payload]),
+    {keep_state, State};
+connected(info, {puback, #{packet_id := PacketId}},
+          State = #state{inflight = Inflight, buffer = Buffer}) ->
+    case maps:take(PacketId, Inflight) of
+        {Id, RestInflight} ->
+            NBuffer = endpoint_buffer:ack(Id, Buffer),
+            {keep_state, State#state{buffer = NBuffer, inflight = RestInflight}};
+        error ->
+            {keep_state, State}
+    end;
+connected(info, {'EXIT', ConnPid, Reason},
+          State = #state{endpoint = #endpoint{title = Title}, conn_pid = ConnPid, buffer = Buffer}) ->
+    logger:warning("[endpoint_mqtt] endpoint: ~p, conn pid exit with reason: ~p", [Title, Reason]),
+    NBuffer = endpoint_buffer:recover_inflight(Buffer),
+    {next_state, disconnected, State#state{conn_pid = undefined, inflight = #{}, buffer = NBuffer},
+     [{state_timeout, ?RETRY_INTERVAL, connect}]};
+connected(info, Info, State) ->
+    unknown_info(Info, connected, State);
+connected(EventType, EventContent, State) ->
+    unknown_event(EventType, EventContent, connected, State).
+
+-spec terminate(term(), mqtt_state(), #state{}) -> term().
+terminate(Reason, _StateName, #state{endpoint = #endpoint{title = Title}, buffer = Buffer, conn_pid = ConnPid}) ->
+    logger:debug("[endpoint_mqtt] endpoint: ~p, terminate with reason: ~p", [Title, Reason]),
+    stop_mqtt_conn(ConnPid),
+    endpoint_buffer:cleanup(Buffer),
+    ok.
+
+-spec code_change(term() | {down, term()}, mqtt_state(), #state{}, term()) ->
+    {ok, mqtt_state(), #state{}} | {error, term()}.
+code_change(_OldVsn, StateName, State = #state{}, _Extra) ->
+    {ok, StateName, State}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec reply_stat(gen_statem:from(), endpoint_buffer:buffer(), #state{}) ->
+    gen_statem:event_handler_result(mqtt_state(), #state{}).
+reply_stat(From, Buffer, State) ->
+    Stat = endpoint_buffer:stat(Buffer),
+    {keep_state, State, [{reply, From, {ok, Stat}}]}.
+
+-spec forward_metric(binary(), endpoint_buffer:buffer(), #state{}) ->
+    gen_statem:event_handler_result(mqtt_state(), #state{}).
+forward_metric(Metric, Buffer, State) ->
+    NBuffer = endpoint_buffer:append(Metric, Buffer),
+    {keep_state, State#state{buffer = NBuffer}}.
+
+-spec store_metric(binary(), endpoint_buffer:buffer(), #state{}) ->
+    gen_statem:event_handler_result(mqtt_state(), #state{}).
+store_metric(Metric, Buffer, State) ->
+    NBuffer = endpoint_buffer:append_only(Metric, Buffer),
+    {keep_state, State#state{buffer = NBuffer}}.
+
+-spec cleanup_buffer(endpoint_buffer:buffer(), #state{}) ->
+    gen_statem:event_handler_result(mqtt_state(), #state{}).
+cleanup_buffer(Buffer, State) ->
+    NBuffer = endpoint_buffer:cleanup(Buffer),
+    {keep_state, State#state{buffer = NBuffer}}.
+
+-spec reload_endpoint(binary(), binary(), undefined | pid(), #endpoint{}, #state{}) ->
+    gen_statem:event_handler_result(mqtt_state(), #state{}).
+reload_endpoint(Matcher, NMatcher, ConnPid, NEndpoint, State = #state{}) ->
+    ensure_subscription(Matcher, NMatcher),
+    stop_mqtt_conn(ConnPid),
+    NBuffer = endpoint_buffer:recover_inflight(State#state.buffer),
+    {next_state, disconnected, State#state{endpoint = NEndpoint, conn_pid = undefined, inflight = #{}, buffer = NBuffer},
+     [{next_event, internal, do_connect}]}.
+
+-spec unknown_info(term(), mqtt_state(), #state{}) ->
+    gen_statem:event_handler_result(mqtt_state(), #state{}).
+unknown_info(Info, StateName, State) ->
+    logger:warning("[endpoint_mqtt] unknown message: ~p, status: ~p", [Info, StateName]),
+    {keep_state, State}.
+
+-spec unknown_event(gen_statem:event_type(), term(), mqtt_state(), #state{}) ->
+    gen_statem:event_handler_result(mqtt_state(), #state{}).
+unknown_event(EventType, EventContent, StateName, State) ->
+    logger:warning("[endpoint_mqtt] unknown event: ~p, content: ~p, status: ~p", [EventType, EventContent, StateName]),
+    {keep_state, State}.
+
+-spec ensure_subscription(binary(), binary()) -> ok.
+ensure_subscription(Matcher, Matcher) ->
+    ok;
+ensure_subscription(Matcher, NMatcher) ->
+    ok = endpoint_subscription:unsubscribe(Matcher, self()),
+    endpoint_subscription:subscribe(NMatcher, self()).
+
+-spec connect_mqtt(list()) -> {ok, pid()} | {error, term()}.
+connect_mqtt(Opts) ->
+    try
+        {ok, ConnPid} = emqtt:start_link(Opts),
+        logger:debug("[endpoint_mqtt] start connect, options: ~p", [Opts]),
+        case emqtt:connect(ConnPid, 5000) of
+            {ok, _} ->
+                {ok, ConnPid};
+            {error, Reason} ->
+                stop_mqtt_conn(ConnPid),
+                {error, Reason}
+        end
+    catch
+        _:Error ->
+            {error, Error}
+    end.
+
+-spec stop_mqtt_conn(undefined | pid()) -> ok.
+stop_mqtt_conn(undefined) ->
+    ok;
+stop_mqtt_conn(ConnPid) when is_pid(ConnPid) ->
+    _ = catch emqtt:stop(ConnPid),
+    ok.

apps/endpoint/src/endpoint_outbox.erl

@@ -0,0 +1,953 @@
+%%%-------------------------------------------------------------------
+%%% @author Codex
+%%% @doc
+%%% endpoint 持久化 outbox 的第一阶段实现：
+%%% - 一个 endpoint 一个目录
+%%% - segment 按递增编号命名
+%%% - metadata 只记录连续 ack 指针
+%%% - inflight 不落盘，崩溃后从 acked_seq + 1 重放
+%%%-------------------------------------------------------------------
+-module(endpoint_outbox).
+
+-export([open/2, close/1, append/2, ack/2, next/1, reset_reader/1, stat/1]).
+-export_type([outbox/0]).
+
+-record(segment, {
+    id :: pos_integer(),
+    path :: file:filename_all(),
+    start_seq = 0 :: non_neg_integer(),
+    end_seq = 0 :: non_neg_integer(),
+    records = 0 :: non_neg_integer(),
+    bytes = 0 :: non_neg_integer()
+}).
+
+-record(outbox_writer, {
+    segment = 1 :: pos_integer(),
+    next_seq = 1 :: pos_integer(),
+    fd = undefined :: undefined | file:fd()
+}).
+
+-record(outbox_reader, {
+    segment :: undefined | pos_integer(),
+    offset = 0 :: non_neg_integer(),
+    fd = undefined :: undefined | file:fd()
+}).
+
+-record(outbox_checkpoint, {
+    acked_segment = 1 :: pos_integer(),
+    acked_seq = 0 :: non_neg_integer(),
+    pending_acks = [] :: [pos_integer()]
+}).
+
+-record(outbox, {
+    dir :: file:filename_all(),
+    metadata_path :: file:filename_all(),
+    max_records = 500000 :: pos_integer(),
+    max_bytes = 134217728 :: pos_integer(),
+    max_segments = 10 :: pos_integer(),
+    segments = [] :: [#segment{}],
+    writer = #outbox_writer{} :: #outbox_writer{},
+    reader = #outbox_reader{} :: #outbox_reader{},
+    checkpoint = #outbox_checkpoint{} :: #outbox_checkpoint{}
+}).
+
+-type outbox() :: #outbox{}.
+
+-define(METADATA_FILE, "metadata.json").
+-define(SEGMENT_EXT, ".log").
+-define(MAX_PAYLOAD_BYTES, 32 * 1024).
+-define(RECORD_MAGIC, <<"EPOBOX01">>).
+-define(RECORD_VERSION, 1).
+-define(RECORD_HEADER_BYTES, 21).
+-define(RECORD_FOOTER_BYTES, 4).
+-define(RECORD_OVERHEAD_BYTES, ?RECORD_HEADER_BYTES + ?RECORD_FOOTER_BYTES).
+-define(RESYNC_READ_BYTES, 4096).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+%% Opts 支持的参数：
+%% - max_records => pos_integer()
+%%     单个 segment 允许的最大记录数，默认 500000
+%% - max_bytes => pos_integer()
+%%     单个 segment 允许的最大字节数，默认 134217728
+%% - max_segments => pos_integer()
+%%     未确认 segment 的最大数量，默认 10
+-spec open(file:filename_all(), map()) -> {ok, outbox()} | {error, term()}.
+open(Dir, Opts) when is_map(Opts) ->
+    ok = ensure_dir(Dir),
+    MetadataPath = filename:join(Dir, ?METADATA_FILE),
+    Segments = load_segments(Dir),
+    Meta = read_metadata(MetadataPath),
+    LastSegmentId = last_segment_id(Segments, maps:get(write_segment, Meta, 1)),
+    LastSeq = last_seq(Segments, maps:get(write_seq, Meta, 0)),
+    AckedSeq = min(maps:get(acked_seq, Meta, 0), LastSeq),
+    AckedSegment = segment_id_for_seq(Segments, AckedSeq, maps:get(acked_segment, Meta, 1)),
+    Outbox0 = #outbox{
+        dir = Dir,
+        metadata_path = MetadataPath,
+        max_records = maps:get(max_records, Opts, 500000),
+        max_bytes = maps:get(max_bytes, Opts, 134217728),
+        max_segments = maps:get(max_segments, Opts, 10),
+        writer = #outbox_writer{
+            segment = LastSegmentId,
+            next_seq = LastSeq + 1
+        },
+        checkpoint = #outbox_checkpoint{
+            acked_segment = AckedSegment,
+            acked_seq = AckedSeq
+        },
+        segments = Segments
+    },
+    {ReadSegment, ReadOffset} = locate_reader(Outbox0),
+    Outbox1 = Outbox0#outbox{
+        reader = #outbox_reader{
+            segment = ReadSegment,
+            offset = ReadOffset
+        }
+    },
+    maybe
+        {ok, Outbox} ?= open_fds(Outbox1),
+        ok ?= persist_metadata(Outbox),
+        {ok, Outbox}
+    else
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec close(outbox()) -> ok.
+close(Outbox = #outbox{}) ->
+    _ = close_open_fds(Outbox),
+    ok.
+
+-spec append(binary(), outbox()) ->
+    {ok, Seq :: pos_integer(), outbox()} | {dropped, capacity_reached, outbox()} | {error, term()}.
+append(Payload, Outbox = #outbox{}) when is_binary(Payload) ->
+    case validate_payload_size(Payload) of
+        ok ->
+            RecordBytes = encoded_record_size(Payload),
+            case ensure_writable_segment(RecordBytes, Outbox) of
+                {drop, NOutbox} ->
+                    {dropped, capacity_reached, NOutbox};
+                {ok, SegmentId, NOutbox0} ->
+                    maybe
+                        {ok, NOutbox1} ?= ensure_write_fd(SegmentId, NOutbox0),
+                        Writer1 = NOutbox1#outbox.writer,
+                        Seq = Writer1#outbox_writer.next_seq,
+                        SegmentPath = segment_path(NOutbox1#outbox.dir, SegmentId),
+                        ok ?= append_record(Writer1#outbox_writer.fd, Seq, Payload),
+                        Segments = update_written_segment(
+                            NOutbox1#outbox.segments, SegmentId, SegmentPath, Seq, RecordBytes),
+                        NOutbox2 = NOutbox1#outbox{
+                            writer = Writer1#outbox_writer{
+                                segment = SegmentId,
+                                next_seq = Seq + 1
+                            },
+                            segments = Segments
+                        },
+                        {ok, NOutbox} ?= maybe_reset_reader_after_append(Outbox, NOutbox2),
+                        ok ?= persist_metadata(NOutbox),
+                        {ok, Seq, NOutbox}
+                    else
+                        {error, Reason} ->
+                            {error, Reason}
+                    end
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec ack(pos_integer(), outbox()) -> {ok, outbox()} | {error, term()}.
+ack(Seq, Outbox = #outbox{checkpoint = Checkpoint}) when is_integer(Seq), Seq > 0 ->
+    AckedSeq = Checkpoint#outbox_checkpoint.acked_seq,
+    case Seq =< AckedSeq of
+        true ->
+            {ok, Outbox};
+        false ->
+            PendingAcks0 = ordsets:add_element(Seq, Checkpoint#outbox_checkpoint.pending_acks),
+            {NackedSeq, PendingAcks} = advance_acked_seq(AckedSeq, PendingAcks0),
+            NackedSegment = segment_id_for_seq(
+                Outbox#outbox.segments,
+                NackedSeq,
+                Checkpoint#outbox_checkpoint.acked_segment),
+            Outbox0 = Outbox#outbox{
+                checkpoint = Checkpoint#outbox_checkpoint{
+                    acked_seq = NackedSeq,
+                    acked_segment = NackedSegment,
+                    pending_acks = PendingAcks
+                }
+            },
+            maybe
+                {ok, Outbox1} ?= prune_acked_segments(Outbox0),
+                ok ?= maybe_persist_metadata(NackedSeq =/= AckedSeq, Outbox1),
+                {ok, Outbox1}
+            else
+                {error, Reason} ->
+                    {error, Reason}
+            end
+    end.
+
+-spec next(outbox()) ->
+    eof | {ok, Seq :: pos_integer(), Payload :: binary(), outbox()} | {error, term()}.
+next(#outbox{reader = #outbox_reader{segment = undefined}}) ->
+    eof;
+next(Outbox = #outbox{
+        reader = Reader,
+        segments = Segments
+    }) ->
+    SegmentId = Reader#outbox_reader.segment,
+    Offset = Reader#outbox_reader.offset,
+    case ensure_read_fd(Outbox) of
+        {ok, NOutbox0} ->
+            Reader0 = NOutbox0#outbox.reader,
+            case read_next_record(Reader0#outbox_reader.fd, Offset) of
+                eof ->
+                    case next_segment_id(Segments, SegmentId) of
+                        undefined ->
+                            eof;
+                        NextSegmentId ->
+                            case switch_read_segment(NextSegmentId, 0, NOutbox0) of
+                                {ok, NOutbox1} ->
+                                    next(NOutbox1);
+                                {error, Reason} ->
+                                    {error, Reason}
+                            end
+                    end;
+                {ok, _RecordOffset, Seq, Payload, NextOffset} ->
+                    {ok, Seq, Payload, NOutbox0#outbox{
+                        reader = Reader0#outbox_reader{offset = NextOffset}
+                    }};
+                {error, Reason} ->
+                    {error, Reason}
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec reset_reader(outbox()) -> {ok, outbox()}.
+reset_reader(Outbox = #outbox{}) ->
+    {ReadSegment, ReadOffset} = locate_reader(Outbox),
+    switch_read_segment(ReadSegment, ReadOffset, Outbox).
+
+-spec stat(outbox()) -> map().
+stat(#outbox{
+        dir = Dir,
+        writer = #outbox_writer{
+            segment = WriteSegment,
+            next_seq = NextSeq
+        },
+        checkpoint = #outbox_checkpoint{
+            acked_segment = AckedSegment,
+            acked_seq = AckedSeq
+        },
+        segments = Segments
+    }) ->
+    #{
+        dir => Dir,
+        write_segment => WriteSegment,
+        write_seq => max(NextSeq - 1, 0),
+        acked_segment => AckedSegment,
+        acked_seq => AckedSeq,
+        segment_num => length(Segments),
+        pending_segment_num => pending_segment_count(Segments, AckedSeq)
+    }.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec ensure_dir(file:filename_all()) -> ok.
+ensure_dir(Dir) ->
+    ok = filelib:ensure_dir(filename:join(Dir, "dummy")),
+    ok.
+
+-spec read_metadata(file:filename_all()) -> map().
+read_metadata(Path) ->
+    case file:read_file(Path) of
+        {ok, Bin} ->
+            case catch json:decode(Bin) of
+                #{<<"write_segment">> := WriteSegment,
+                  <<"write_seq">> := WriteSeq,
+                  <<"acked_segment">> := AckedSegment,
+                  <<"acked_seq">> := AckedSeq}
+                        when is_integer(WriteSegment), is_integer(WriteSeq),
+                             is_integer(AckedSegment), is_integer(AckedSeq) ->
+                    #{
+                        write_segment => WriteSegment,
+                        write_seq => WriteSeq,
+                        acked_segment => AckedSegment,
+                        acked_seq => AckedSeq
+                    };
+                _ ->
+                    default_metadata()
+            end;
+        {error, enoent} ->
+            default_metadata();
+        {error, _} ->
+            default_metadata()
+    end.
+
+-spec default_metadata() -> map().
+default_metadata() ->
+    #{
+        write_segment => 1,
+        write_seq => 0,
+        acked_segment => 1,
+        acked_seq => 0
+    }.
+
+-spec persist_metadata(outbox()) -> ok | {error, term()}.
+persist_metadata(#outbox{
+        metadata_path = MetadataPath,
+        writer = #outbox_writer{
+            segment = WriteSegment,
+            next_seq = NextSeq
+        },
+        checkpoint = #outbox_checkpoint{
+            acked_segment = AckedSegment,
+            acked_seq = AckedSeq
+        }
+    }) ->
+    TmpPath = MetadataPath ++ ".tmp",
+    Metadata = #{
+        <<"write_segment">> => WriteSegment,
+        <<"write_seq">> => max(NextSeq - 1, 0),
+        <<"acked_segment">> => AckedSegment,
+        <<"acked_seq">> => AckedSeq
+    },
+    Bin = iolist_to_binary(json:encode(Metadata)),
+    ok = file:write_file(TmpPath, Bin),
+    ok = file:rename(TmpPath, MetadataPath),
+    ok.
+
+-spec load_segments(file:filename_all()) -> [#segment{}].
+load_segments(Dir) ->
+    Pattern = filename:join(Dir, "*" ++ ?SEGMENT_EXT),
+    SegmentFiles = lists:sort(filelib:wildcard(Pattern)),
+    lists:filtermap(fun(Path) ->
+        case parse_segment_id(filename:basename(Path)) of
+            {ok, Id} ->
+                case scan_segment(Path, Id) of
+                    {ok, Segment = #segment{records = Records}} when Records > 0 ->
+                        {true, Segment};
+                    _ ->
+                        false
+                end;
+            error ->
+                false
+        end
+    end, SegmentFiles).
+
+-spec parse_segment_id(string()) -> {ok, pos_integer()} | error.
+parse_segment_id(Filename) ->
+    case filename:extension(Filename) of
+        ?SEGMENT_EXT ->
+            Base = filename:rootname(Filename, ?SEGMENT_EXT),
+            try
+                Id = list_to_integer(Base),
+                case Id > 0 of
+                    true ->
+                        {ok, Id};
+                    false ->
+                        error
+                end
+            catch
+                _:_ ->
+                    error
+            end;
+        _ ->
+            error
+    end.
+
+-spec scan_segment(file:filename_all(), pos_integer()) -> {ok, #segment{}} | {error, term()}.
+scan_segment(Path, Id) ->
+    case file:open(Path, [read, raw, binary]) of
+        {ok, Fd} ->
+            try
+                scan_segment_records(Fd, 0, #segment{id = Id, path = Path})
+            after
+                ok = file:close(Fd)
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec scan_segment_records(file:fd(), non_neg_integer(), #segment{}) -> {ok, #segment{}} | {error, term()}.
+scan_segment_records(Fd, Offset, Segment = #segment{}) ->
+    case read_next_record(Fd, Offset) of
+        eof ->
+            {ok, Segment};
+        {ok, RecordOffset, Seq, _Payload, NextOffset} ->
+            RecordBytes = NextOffset - RecordOffset,
+            NextSegment = Segment#segment{
+                start_seq = case Segment#segment.records of
+                    0 -> Seq;
+                    _ -> Segment#segment.start_seq
+                end,
+                end_seq = Seq,
+                records = Segment#segment.records + 1,
+                bytes = Segment#segment.bytes + RecordBytes
+            },
+            scan_segment_records(Fd, NextOffset, NextSegment);
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec encoded_record_size(binary()) -> pos_integer().
+encoded_record_size(Payload) ->
+    ?RECORD_OVERHEAD_BYTES + byte_size(Payload).
+
+-spec validate_payload_size(binary()) -> ok | {error, payload_too_large}.
+validate_payload_size(Payload) when byte_size(Payload) =< ?MAX_PAYLOAD_BYTES ->
+    ok;
+validate_payload_size(_Payload) ->
+    {error, payload_too_large}.
+
+-spec ensure_writable_segment(pos_integer(), outbox()) ->
+    {ok, pos_integer(), outbox()} | {drop, outbox()}.
+ensure_writable_segment(RecordBytes, Outbox = #outbox{
+        max_records = MaxRecords,
+        max_bytes = MaxBytes,
+        max_segments = MaxSegments,
+        segments = Segments,
+        writer = #outbox_writer{segment = WriteSegment},
+        checkpoint = #outbox_checkpoint{acked_seq = AckedSeq}
+    }) ->
+    case find_segment(Segments, WriteSegment) of
+        false ->
+            {ok, WriteSegment, Outbox};
+        #segment{records = Records, bytes = Bytes} ->
+            NeedRoll = Records > 0 andalso (Records >= MaxRecords orelse Bytes + RecordBytes > MaxBytes),
+            case NeedRoll of
+                false ->
+                    {ok, WriteSegment, Outbox};
+                true ->
+                    case pending_segment_count(Segments, AckedSeq) >= MaxSegments of
+                        true ->
+                            {drop, Outbox};
+                        false ->
+                            {ok, WriteSegment + 1, Outbox#outbox{
+                                writer = ((Outbox#outbox.writer)#outbox_writer{
+                                    segment = WriteSegment + 1
+                                })
+                            }}
+                    end
+            end
+    end.
+
+-spec append_record(file:fd(), pos_integer(), binary()) -> ok | {error, term()}.
+append_record(Fd, Seq, Payload) when Fd =/= undefined ->
+    PayloadSize = byte_size(Payload),
+    Crc = record_crc(?RECORD_VERSION, Seq, PayloadSize, Payload),
+    Bin = <<
+        ?RECORD_MAGIC/binary,
+        ?RECORD_VERSION:8,
+        Seq:64/unsigned-big-integer,
+        PayloadSize:32/unsigned-big-integer,
+        Payload/binary,
+        Crc:32/unsigned-big-integer
+    >>,
+    case file:write(Fd, Bin) of
+        ok ->
+            ok;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec update_written_segment([#segment{}], pos_integer(), file:filename_all(), pos_integer(), pos_integer()) -> [#segment{}].
+update_written_segment(Segments, SegmentId, Path, Seq, RecordBytes) ->
+    case take_segment(SegmentId, Segments) of
+        {undefined, Rest} ->
+            lists:sort(fun compare_segment/2, [
+                #segment{
+                    id = SegmentId,
+                    path = Path,
+                    start_seq = Seq,
+                    end_seq = Seq,
+                    records = 1,
+                    bytes = RecordBytes
+                } | Rest
+            ]);
+        {Segment, Rest} ->
+            lists:sort(fun compare_segment/2, [
+                Segment#segment{
+                    end_seq = Seq,
+                    records = Segment#segment.records + 1,
+                    bytes = Segment#segment.bytes + RecordBytes
+                } | Rest
+            ])
+    end.
+
+-spec compare_segment(#segment{}, #segment{}) -> boolean().
+compare_segment(#segment{id = Id0}, #segment{id = Id1}) ->
+    Id0 =< Id1.
+
+-spec find_segment([#segment{}], pos_integer()) -> false | #segment{}.
+find_segment(Segments, SegmentId) ->
+    lists:keyfind(SegmentId, #segment.id, Segments).
+
+-spec take_segment(pos_integer(), [#segment{}]) -> {undefined | #segment{}, [#segment{}]}.
+take_segment(SegmentId, Segments) ->
+    case lists:partition(fun(#segment{id = Id}) -> Id =:= SegmentId end, Segments) of
+        {[Segment], Rest} ->
+            {Segment, Rest};
+        {[], Rest} ->
+            {undefined, Rest}
+    end.
+
+-spec last_segment_id([#segment{}], pos_integer()) -> pos_integer().
+last_segment_id([], Default) ->
+    Default;
+last_segment_id(Segments, _Default) ->
+    (lists:last(Segments))#segment.id.
+
+-spec last_seq([#segment{}], non_neg_integer()) -> non_neg_integer().
+last_seq([], Default) ->
+    Default;
+last_seq(Segments, _Default) ->
+    (lists:last(Segments))#segment.end_seq.
+
+-spec segment_id_for_seq([#segment{}], non_neg_integer(), pos_integer()) -> pos_integer().
+segment_id_for_seq(_Segments, 0, Default) ->
+    Default;
+segment_id_for_seq(Segments, Seq, Default) ->
+    case lists:dropwhile(fun(#segment{end_seq = EndSeq}) -> EndSeq < Seq end, Segments) of
+        [#segment{id = SegmentId, start_seq = StartSeq, end_seq = EndSeq} | _]
+                when Seq >= StartSeq, Seq =< EndSeq ->
+            SegmentId;
+        _ ->
+            Default
+    end.
+
+-spec pending_segment_count([#segment{}], non_neg_integer()) -> non_neg_integer().
+pending_segment_count(Segments, AckedSeq) ->
+    length([Segment || Segment = #segment{end_seq = EndSeq} <- Segments, EndSeq > AckedSeq]).
+
+-spec advance_acked_seq(non_neg_integer(), [pos_integer()]) -> {non_neg_integer(), [pos_integer()]}.
+advance_acked_seq(AckedSeq, PendingAcks) ->
+    NextSeq = AckedSeq + 1,
+    case ordsets:is_element(NextSeq, PendingAcks) of
+        true ->
+            advance_acked_seq(NextSeq, ordsets:del_element(NextSeq, PendingAcks));
+        false ->
+            {AckedSeq, PendingAcks}
+    end.
+
+-spec prune_acked_segments(outbox()) -> {ok, outbox()} | {error, term()}.
+prune_acked_segments(Outbox = #outbox{
+        segments = Segments,
+        checkpoint = #outbox_checkpoint{acked_seq = AckedSeq}
+    }) ->
+    {DeleteSegments, KeepSegments} = lists:partition(fun(#segment{end_seq = EndSeq}) ->
+        EndSeq =< AckedSeq
+    end, Segments),
+    Outbox0 = detach_deleted_fds(DeleteSegments, Outbox#outbox{segments = KeepSegments}),
+    ok = lists:foldl(fun(#segment{path = Path}, ok) ->
+        case file:delete(Path) of
+            ok ->
+                ok;
+            {error, enoent} ->
+                ok;
+            {error, Reason} ->
+                throw({delete_segment_failed, Path, Reason})
+        end
+    end, ok, DeleteSegments),
+    {ok, reset_reader_after_prune(Outbox0)}.
+
+-spec reset_reader_after_prune(outbox()) -> outbox().
+reset_reader_after_prune(Outbox = #outbox{
+        reader = #outbox_reader{segment = ReadSegment},
+        segments = Segments
+    }) ->
+    case ReadSegment =:= undefined orelse lists:keymember(ReadSegment, #segment.id, Segments) of
+        true ->
+            Outbox;
+        false ->
+            case reset_reader(Outbox) of
+                {ok, NOutbox} ->
+                    NOutbox
+            end
+    end.
+
+-spec locate_reader(outbox()) -> {undefined | pos_integer(), non_neg_integer()}.
+locate_reader(#outbox{
+        segments = Segments,
+        checkpoint = #outbox_checkpoint{acked_seq = AckedSeq}
+    }) ->
+    NextSeq = AckedSeq + 1,
+    case lists:dropwhile(fun(#segment{end_seq = EndSeq}) -> EndSeq < NextSeq end, Segments) of
+        [] ->
+            {undefined, 0};
+        [#segment{id = SegmentId, start_seq = StartSeq} = Segment | _] ->
+            case NextSeq =< StartSeq of
+                true ->
+                    {SegmentId, 0};
+                false ->
+                    case locate_offset_for_seq(Segment, NextSeq) of
+                        {ok, Offset} ->
+                            {SegmentId, Offset};
+                        {error, _} ->
+                            {SegmentId, 0}
+                    end
+            end
+    end.
+
+-spec locate_offset_for_seq(#segment{}, pos_integer()) -> {ok, non_neg_integer()} | {error, term()}.
+locate_offset_for_seq(#segment{path = Path}, TargetSeq) ->
+    case file:open(Path, [read, raw, binary]) of
+        {ok, Fd} ->
+            try
+                locate_offset_loop(Fd, TargetSeq, 0)
+            after
+                ok = file:close(Fd)
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec locate_offset_loop(file:fd(), pos_integer(), non_neg_integer()) ->
+    {ok, non_neg_integer()} | {error, term()}.
+locate_offset_loop(Fd, TargetSeq, Offset) ->
+    case read_next_record(Fd, Offset) of
+        eof ->
+            {ok, Offset};
+        {ok, RecordOffset, Seq, _Payload, _NextOffset} when Seq >= TargetSeq ->
+            {ok, RecordOffset};
+        {ok, _RecordOffset, _Seq, _Payload, NextOffset} ->
+            locate_offset_loop(Fd, TargetSeq, NextOffset);
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec read_next_record(file:fd(), non_neg_integer()) ->
+    eof | {ok, non_neg_integer(), pos_integer(), binary(), non_neg_integer()} | {error, term()}.
+read_next_record(Fd, Offset) ->
+    case read_record_at(Fd, Offset) of
+        eof ->
+            eof;
+        {ok, Seq, Payload, NextOffset} ->
+            {ok, Offset, Seq, Payload, NextOffset};
+        {error, _Reason} ->
+            resync_next_record(Fd, Offset + 1)
+    end.
+
+-spec read_record_at(file:fd(), non_neg_integer()) ->
+    eof | {ok, pos_integer(), binary(), non_neg_integer()} | {error, term()}.
+read_record_at(Fd, Offset) ->
+    maybe
+        {ok, _} ?= file:position(Fd, Offset),
+        read_record(Fd, Offset)
+    else
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec read_record(file:fd(), non_neg_integer()) ->
+    eof | {ok, pos_integer(), binary(), non_neg_integer()} | {error, term()}.
+read_record(Fd, Offset) ->
+    case file:read(Fd, ?RECORD_HEADER_BYTES) of
+        eof ->
+            eof;
+        {ok, <<
+            Magic:8/binary,
+            Version:8,
+            Seq:64/unsigned-big-integer,
+            PayloadSize:32/unsigned-big-integer
+        >>} when Magic =:= ?RECORD_MAGIC, Version =:= ?RECORD_VERSION ->
+            maybe
+                ok ?= validate_read_payload_size(PayloadSize),
+                read_record_body(Fd, Offset, Seq, PayloadSize)
+            else
+                {error, Reason} ->
+                    {error, Reason}
+            end;
+        {ok, Header} when byte_size(Header) < ?RECORD_HEADER_BYTES ->
+            {error, truncated_header};
+        {ok, _} ->
+            {error, invalid_record_header};
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec validate_read_payload_size(non_neg_integer()) -> ok | {error, payload_too_large}.
+validate_read_payload_size(PayloadSize) when PayloadSize =< ?MAX_PAYLOAD_BYTES ->
+    ok;
+validate_read_payload_size(_PayloadSize) ->
+    {error, payload_too_large}.
+
+-spec read_record_body(file:fd(), non_neg_integer(), pos_integer(), non_neg_integer()) ->
+    {ok, pos_integer(), binary(), non_neg_integer()} | {error, term()}.
+read_record_body(Fd, Offset, Seq, PayloadSize) ->
+    case file:read(Fd, PayloadSize + ?RECORD_FOOTER_BYTES) of
+        {ok, <<Payload:PayloadSize/binary, StoredCrc:32/unsigned-big-integer>>} ->
+            ExpectedCrc = record_crc(?RECORD_VERSION, Seq, PayloadSize, Payload),
+            case StoredCrc =:= ExpectedCrc of
+                true ->
+                    {ok, Seq, Payload, Offset + ?RECORD_OVERHEAD_BYTES + PayloadSize};
+                false ->
+                    {error, invalid_record_crc}
+            end;
+        eof ->
+            {error, truncated_payload};
+        {ok, _} ->
+            {error, truncated_payload};
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec record_crc(byte(), pos_integer(), non_neg_integer(), binary()) -> non_neg_integer().
+record_crc(Version, Seq, PayloadSize, Payload) ->
+    erlang:crc32(<<
+        Version:8,
+        Seq:64/unsigned-big-integer,
+        PayloadSize:32/unsigned-big-integer,
+        Payload/binary
+    >>).
+
+-spec resync_next_record(file:fd(), non_neg_integer()) ->
+    eof | {ok, non_neg_integer(), pos_integer(), binary(), non_neg_integer()} | {error, term()}.
+resync_next_record(Fd, SearchOffset) ->
+    case find_next_magic(Fd, SearchOffset) of
+        eof ->
+            eof;
+        {ok, RecordOffset} ->
+            case read_record_at(Fd, RecordOffset) of
+                eof ->
+                    eof;
+                {ok, Seq, Payload, NextOffset} ->
+                    {ok, RecordOffset, Seq, Payload, NextOffset};
+                {error, _Reason} ->
+                    resync_next_record(Fd, RecordOffset + 1)
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec find_next_magic(file:fd(), non_neg_integer()) ->
+    eof | {ok, non_neg_integer()} | {error, term()}.
+find_next_magic(Fd, Offset) ->
+    maybe
+        {ok, _} ?= file:position(Fd, Offset),
+        find_next_magic_loop(Fd, Offset, <<>>)
+    else
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec find_next_magic_loop(file:fd(), non_neg_integer(), binary()) ->
+    eof | {ok, non_neg_integer()} | {error, term()}.
+find_next_magic_loop(Fd, Offset, Tail) ->
+    case file:read(Fd, ?RESYNC_READ_BYTES) of
+        eof ->
+            eof;
+        {ok, Bin} ->
+            SearchBin = <<Tail/binary, Bin/binary>>,
+            BaseOffset = Offset - byte_size(Tail),
+            case binary:match(SearchBin, ?RECORD_MAGIC) of
+                {Pos, _Len} ->
+                    {ok, BaseOffset + Pos};
+                nomatch ->
+                    NextTail = trailing_bytes(SearchBin, byte_size(?RECORD_MAGIC) - 1),
+                    find_next_magic_loop(Fd, Offset + byte_size(Bin), NextTail)
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec trailing_bytes(binary(), non_neg_integer()) -> binary().
+trailing_bytes(Bin, KeepBytes) ->
+    Size = byte_size(Bin),
+    case Size =< KeepBytes of
+        true ->
+            Bin;
+        false ->
+            binary:part(Bin, Size - KeepBytes, KeepBytes)
+    end.
+
+-spec next_segment_id([#segment{}], pos_integer()) -> undefined | pos_integer().
+next_segment_id(Segments, SegmentId) ->
+    case lists:dropwhile(fun(#segment{id = Id}) -> Id =< SegmentId end, Segments) of
+        [#segment{id = NextSegmentId} | _] ->
+            NextSegmentId;
+        [] ->
+            undefined
+    end.
+
+-spec segment_path(file:filename_all(), pos_integer()) -> file:filename_all().
+segment_path(Dir, SegmentId) ->
+    filename:join(Dir, io_lib:format("~6..0B~s", [SegmentId, ?SEGMENT_EXT])).
+
+-spec open_fds(outbox()) -> {ok, outbox()} | {error, term()}.
+open_fds(Outbox) ->
+    case maybe_open_write_fd(Outbox) of
+        {ok, Outbox0} ->
+            maybe
+                {ok, NOutbox} ?= maybe_open_read_fd(Outbox0),
+                {ok, NOutbox}
+            else
+                {error, Reason} ->
+                    _ = close_open_fds(Outbox0),
+                    {error, Reason}
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec maybe_open_write_fd(outbox()) -> {ok, outbox()} | {error, term()}.
+maybe_open_write_fd(Outbox = #outbox{
+        segments = Segments,
+        writer = #outbox_writer{
+            segment = WriteSegment,
+            fd = undefined
+        }
+    }) ->
+    case find_segment(Segments, WriteSegment) of
+        false ->
+            {ok, Outbox};
+        _ ->
+            ensure_write_fd(WriteSegment, Outbox)
+    end.
+
+-spec maybe_open_read_fd(outbox()) -> {ok, outbox()} | {error, term()}.
+maybe_open_read_fd(Outbox = #outbox{reader = #outbox_reader{segment = undefined}}) ->
+    {ok, Outbox};
+maybe_open_read_fd(Outbox = #outbox{reader = #outbox_reader{fd = undefined}}) ->
+    ensure_read_fd(Outbox);
+maybe_open_read_fd(Outbox = #outbox{}) ->
+    {ok, Outbox}.
+
+-spec ensure_write_fd(pos_integer(), outbox()) -> {ok, outbox()} | {error, term()}.
+ensure_write_fd(SegmentId, Outbox = #outbox{
+        writer = #outbox_writer{
+            segment = SegmentId,
+            fd = Fd
+        }
+    })
+        when Fd =/= undefined ->
+    {ok, Outbox};
+ensure_write_fd(SegmentId, Outbox = #outbox{dir = Dir}) ->
+    Writer0 = (Outbox#outbox.writer)#outbox_writer{segment = SegmentId},
+    Outbox0 = close_write_fd(Outbox#outbox{writer = Writer0}),
+    Path = segment_path(Dir, SegmentId),
+    case file:open(Path, [append, raw, binary]) of
+        {ok, Fd} ->
+            Writer1 = (Outbox0#outbox.writer)#outbox_writer{fd = Fd},
+            {ok, Outbox0#outbox{writer = Writer1}};
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec ensure_read_fd(outbox()) -> {ok, outbox()} | {error, term()}.
+ensure_read_fd(Outbox = #outbox{reader = #outbox_reader{segment = undefined}}) ->
+    {ok, Outbox};
+ensure_read_fd(Outbox = #outbox{reader = #outbox_reader{fd = Fd}}) when Fd =/= undefined ->
+    {ok, Outbox};
+ensure_read_fd(Outbox = #outbox{
+        reader = Reader,
+        dir = Dir
+    }) ->
+    SegmentId = Reader#outbox_reader.segment,
+    Offset = Reader#outbox_reader.offset,
+    Path = segment_path(Dir, SegmentId),
+    case open_read_fd(Path, Offset) of
+        {ok, Fd} ->
+            {ok, Outbox#outbox{
+                reader = Reader#outbox_reader{fd = Fd}
+            }};
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec switch_read_segment(undefined | pos_integer(), non_neg_integer(), outbox()) ->
+    {ok, outbox()} | {error, term()}.
+switch_read_segment(undefined, _Offset, Outbox) ->
+    Reader0 = #outbox_reader{segment = undefined, offset = 0},
+    {ok, close_read_fd(Outbox#outbox{reader = Reader0})};
+switch_read_segment(SegmentId, Offset, Outbox = #outbox{reader = Reader}) ->
+    ensure_read_fd(close_read_fd(Outbox#outbox{
+        reader = Reader#outbox_reader{
+            segment = SegmentId,
+            offset = Offset
+        }
+    })).
+
+-spec maybe_reset_reader_after_append(outbox(), outbox()) -> {ok, outbox()} | {error, term()}.
+maybe_reset_reader_after_append(#outbox{reader = #outbox_reader{segment = undefined}}, Outbox) ->
+    reset_reader(Outbox);
+maybe_reset_reader_after_append(_PrevOutbox, Outbox) ->
+    {ok, Outbox}.
+
+-spec detach_deleted_fds([#segment{}], outbox()) -> outbox().
+detach_deleted_fds(DeleteSegments, Outbox = #outbox{
+        reader = #outbox_reader{segment = ReadSegment},
+        writer = #outbox_writer{segment = WriteSegment}
+    }) ->
+    DeleteIds = [SegmentId || #segment{id = SegmentId} <- DeleteSegments],
+    Outbox0 = case lists:member(ReadSegment, DeleteIds) of
+        true ->
+            close_read_fd(Outbox);
+        false ->
+            Outbox
+    end,
+    case lists:member(WriteSegment, DeleteIds) of
+        true ->
+            close_write_fd(Outbox0);
+        false ->
+            Outbox0
+    end.
+
+-spec close_open_fds(outbox()) -> outbox().
+close_open_fds(Outbox) ->
+    close_write_fd(close_read_fd(Outbox)).
+
+-spec close_read_fd(outbox()) -> outbox().
+close_read_fd(Outbox = #outbox{reader = #outbox_reader{fd = undefined}}) ->
+    Outbox;
+close_read_fd(Outbox = #outbox{reader = Reader}) ->
+    Fd = Reader#outbox_reader.fd,
+    _ = maybe_close_fd(Fd),
+    Outbox#outbox{reader = Reader#outbox_reader{fd = undefined}}.
+
+-spec close_write_fd(outbox()) -> outbox().
+close_write_fd(Outbox = #outbox{writer = #outbox_writer{fd = undefined}}) ->
+    Outbox;
+close_write_fd(Outbox = #outbox{writer = Writer}) ->
+    Fd = Writer#outbox_writer.fd,
+    _ = maybe_close_fd(Fd),
+    Outbox#outbox{writer = Writer#outbox_writer{fd = undefined}}.
+
+-spec maybe_close_fd(file:fd()) -> ok.
+maybe_close_fd(Fd) ->
+    case file:close(Fd) of
+        ok ->
+            ok;
+        {error, badarg} ->
+            ok;
+        {error, terminated} ->
+            ok;
+        {error, _Reason} ->
+            ok
+    end.
+
+-spec maybe_persist_metadata(boolean(), outbox()) -> ok | {error, term()}.
+maybe_persist_metadata(true, Outbox) ->
+    persist_metadata(Outbox);
+maybe_persist_metadata(false, _Outbox) ->
+    ok.
+
+-spec open_read_fd(file:filename_all(), non_neg_integer()) -> {ok, file:fd()} | {error, term()}.
+open_read_fd(Path, Offset) ->
+    case file:open(Path, [read, raw, binary]) of
+        {ok, Fd} ->
+            maybe
+                {ok, _} ?= file:position(Fd, Offset),
+                {ok, Fd}
+            else
+                {error, Reason} ->
+                    ok = file:close(Fd),
+                    {error, Reason}
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.

apps/endpoint/src/endpoint_subscription.erl

@@ -0,0 +1,457 @@
+%%%-------------------------------------------------------------------
+%%% @author anlicheng
+%%% @copyright (C) 2025, <COMPANY>
+%%% @doc
+%%% Endpoint 数据分发订阅索引。
+%%%
+%%% 当前实现面向较大订阅量场景，publish 热路径不再全量遍历订阅表，而是把
+%%% 订阅拆成 exact 和 wildcard 两类：
+%%%
+%%% <ul>
+%%%   <li>精确订阅写入 endpoint_subscription_exact，publish 时按 RouteKey
+%%%       直接 ets:lookup/2。</li>
+%%%   <li>包含通配符的订阅写入 ETS trie。trie 边存放在
+%%%       endpoint_subscription_trie_edge，节点订阅存放在
+%%%       endpoint_subscription_trie_sub。</li>
+%%%   <li>endpoint_subscription_reverse 保存 SubscriberPid 到订阅对象的反向索引，
+%%%       用于 unsubscribe 和进程 DOWN 清理。</li>
+%%%   <li>endpoint_subscription_pid 记录每个 SubscriberPid 的 monitor 和订阅计数，
+%%%       同一个 pid 只 monitor 一次。</li>
+%%% </ul>
+%%%
+%%% 所有 matcher 共用同一个 root node，root node id 固定为 0。新 trie 节点使用
+%%% 单调递增的 integer node id。subscribe/unsubscribe 通过本 gen_server 串行写 ETS；
+%%% publish/2 直接并发读取 ETS，避免中心 gen_server 成为数据分发瓶颈。
+%%%
+%%% 匹配规则：
+%%%
+%%% <ul>
+%%%   <li>普通 segment 必须完全相等，例如 <<"device/a/temp">> 只精确匹配
+%%%       同名订阅。</li>
+%%%   <li><<"*">> 是单级通配，只匹配一个 segment。例如
+%%%       <<"device/*/temp">> 匹配 <<"device/a/temp">>，不匹配
+%%%       <<"device/a/b/temp">>。</li>
+%%%   <li><<"+">> 是末尾多级通配，只允许出现在 matcher 最后一段，并且至少
+%%%       匹配一个剩余 segment。例如 <<"device/+">> 匹配 <<"device/a">>
+%%%       和 <<"device/a/temp">>，不匹配 <<"device">>。</li>
+%%% </ul>
+%%%
+%%% publish/2 的匹配流程：
+%%%
+%%% <ol>
+%%%   <li>从 exact 表直接查找 RouteKey 对应的订阅。</li>
+%%%   <li>把 RouteKey 按 "/" 拆分后，从 trie root 开始逐层匹配。每层最多查找
+%%%       当前 segment 分支和 <<"*">> 分支。</li>
+%%%   <li>每消费一层前，收集当前候选节点上的 plus 订阅；全部 segment 消费完后，
+%%%       收集候选节点上的 exact 订阅。</li>
+%%%   <li>合并 exact 和 trie 命中结果，按原有优先级排序，再按 SubscriberPid 去重。
+%%%       同一个 pid 同时被多个 matcher 命中时，保留优先级最高的订阅。</li>
+%%%   <li>对最终订阅者执行 endpoint:forward/2。</li>
+%%% </ol>
+%%% @end
+%%% Created : 07. 11月 2025 16:27
+%%%-------------------------------------------------------------------
+-module(endpoint_subscription).
+-author("anlicheng").
+
+-behaviour(gen_server).
+
+%% API
+-export([start_link/0]).
+-export([subscribe/2, unsubscribe/2, publish/2, get_subscribers/0]).
+-export([is_valid_components/1, of_components/1]).
+
+%% gen_server callbacks
+-export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
+
+-define(SERVER, ?MODULE).
+-define(EXACT_TAB, endpoint_subscription_exact).
+-define(TRIE_EDGE_TAB, endpoint_subscription_trie_edge).
+-define(TRIE_SUB_TAB, endpoint_subscription_trie_sub).
+-define(REVERSE_TAB, endpoint_subscription_reverse).
+-define(PID_TAB, endpoint_subscription_pid).
+-define(ROOT_NODE, 0).
+
+%% 定义订阅者
+-record(subscriber, {
+    topic :: binary(),
+    subscriber_pid :: pid(),
+    components = [],
+    monitor_ref :: undefined | reference(),
+    %% 优先级
+    %% 1. 完全匹配的topic优先级别最高
+    %% 2. 带 * 的订阅
+    %% 3. 带 + 的订阅
+    order :: integer()
+}).
+
+-record(state, {
+    exact_tid :: ets:tid(),
+    edge_tid :: ets:tid(),
+    trie_sub_tid :: ets:tid(),
+    reverse_tid :: ets:tid(),
+    pid_tid :: ets:tid(),
+    next_node_id = 1 :: pos_integer()
+}).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec subscribe(Topic :: binary(), SubscriberPid :: pid()) -> ok | {error, Reason :: binary()}.
+subscribe(Topic, SubscriberPid) when is_binary(Topic), is_pid(SubscriberPid) ->
+    gen_server:call(?SERVER, {subscribe, Topic, SubscriberPid}).
+
+-spec unsubscribe(Topic :: binary(), SubscriberPid :: pid()) -> ok.
+unsubscribe(Topic, SubscriberPid) when is_binary(Topic), is_pid(SubscriberPid) ->
+    gen_server:call(?SERVER, {unsubscribe, Topic, SubscriberPid}).
+
+-spec get_subscribers() -> {ok, Subscribers :: list()}.
+get_subscribers() ->
+    gen_server:call(?SERVER, get_subscribers).
+
+-spec publish(RouteKey :: binary(), Content :: binary()) -> ok.
+publish(RouteKey, Content) when is_binary(RouteKey), is_binary(Content) ->
+    case ets:info(?EXACT_TAB) of
+        undefined ->
+            ok;
+        _ ->
+            MatchedSubscribers = match_route_key(RouteKey),
+            lists:foreach(fun(#subscriber{subscriber_pid = SubscriberPid}) ->
+                endpoint:forward(SubscriberPid, Content)
+            end, MatchedSubscribers),
+            maybe_log_unmatched_publish(RouteKey, Content, MatchedSubscribers),
+            logger:debug("[endpoint_subscription] route_key: ~p, match_count: ~p", [RouteKey, length(MatchedSubscribers)]),
+            ok
+    end.
+
+%% @doc Spawns the server and registers the local name (unique)
+-spec(start_link() ->
+    {ok, Pid :: pid()} | ignore | {error, Reason :: term()}).
+start_link() ->
+    gen_server:start_link({local, ?SERVER}, ?MODULE, [], []).
+
+%%%===================================================================
+%%% gen_server callbacks
+%%%===================================================================
+
+%% @private
+%% @doc Initializes the server
+-spec(init(Args :: term()) ->
+    {ok, State :: #state{}} | {ok, State :: #state{}, timeout() | hibernate} |
+    {stop, Reason :: term()} | ignore).
+init([]) ->
+    ok = endpoint_util:set_metadata(),
+    ExactTid = ets:new(?EXACT_TAB, [named_table, protected, bag, {read_concurrency, true}]),
+    EdgeTid = ets:new(?TRIE_EDGE_TAB, [named_table, protected, set, {read_concurrency, true}]),
+    TrieSubTid = ets:new(?TRIE_SUB_TAB, [named_table, protected, bag, {read_concurrency, true}]),
+    ReverseTid = ets:new(?REVERSE_TAB, [named_table, protected, bag]),
+    PidTid = ets:new(?PID_TAB, [named_table, protected, set]),
+    {ok, #state{
+        exact_tid = ExactTid,
+        edge_tid = EdgeTid,
+        trie_sub_tid = TrieSubTid,
+        reverse_tid = ReverseTid,
+        pid_tid = PidTid
+    }}.
+
+%% @private
+%% @doc Handling call messages
+-spec(handle_call(Request :: term(), From :: {pid(), Tag :: term()},
+    State :: #state{}) ->
+    {reply, Reply :: term(), NewState :: #state{}} |
+    {reply, Reply :: term(), NewState :: #state{}, timeout() | hibernate} |
+    {noreply, NewState :: #state{}} |
+    {noreply, NewState :: #state{}, timeout() | hibernate} |
+    {stop, Reason :: term(), Reply :: term(), NewState :: #state{}} |
+    {stop, Reason :: term(), NewState :: #state{}}).
+%%  同一个SubscriberPid只能订阅同一个topic一次
+handle_call(get_subscribers, _From, State = #state{exact_tid = ExactTid, trie_sub_tid = TrieSubTid}) ->
+    Subscribers = exact_subscribers(ExactTid) ++ trie_subscribers(TrieSubTid),
+    {reply, {ok, Subscribers}, State};
+
+handle_call({subscribe, Topic, SubscriberPid}, _From, State = #state{reverse_tid = ReverseTid}) ->
+    Components = of_components(Topic),
+    case is_valid_components(Components) of
+        true ->
+            case has_subscription(ReverseTid, Topic, SubscriberPid) of
+                true ->
+                    {reply, ok, State};
+                false ->
+                    {MonitorRef, State1} = ensure_pid_monitor(SubscriberPid, State),
+                    Sub = #subscriber{
+                        topic = Topic,
+                        subscriber_pid = SubscriberPid,
+                        components = Components,
+                        monitor_ref = MonitorRef,
+                        order = order_num(Components)
+                    },
+                    State2 = insert_subscription(Topic, Components, SubscriberPid, Sub, State1),
+                    {reply, ok, State2}
+            end;
+        false ->
+            {reply, {error, <<"invalid topic name">>}, State}
+    end;
+
+handle_call({unsubscribe, Topic, SubscriberPid}, _From, State = #state{reverse_tid = ReverseTid}) ->
+    Removed = [Reverse || Reverse = {SubscriberPid0, Topic0, _Kind, _NodeId, _Sub} <- ets:lookup(ReverseTid, SubscriberPid),
+        SubscriberPid =:= SubscriberPid0, Topic =:= Topic0],
+    lists:foreach(fun(Reverse) -> delete_subscription(Reverse, State) end, Removed),
+    State1 = release_pid_monitor(SubscriberPid, length(Removed), State),
+    {reply, ok, State1}.
+
+%% @private
+%% @doc Handling cast messages
+-spec(handle_cast(Request :: term(), State :: #state{}) ->
+    {noreply, NewState :: #state{}} |
+    {noreply, NewState :: #state{}, timeout() | hibernate} |
+    {stop, Reason :: term(), NewState :: #state{}}).
+handle_cast(_Request, State = #state{}) ->
+    {noreply, State}.
+
+%% @private
+%% @doc Handling all non call/cast messages
+-spec(handle_info(Info :: timeout() | term(), State :: #state{}) ->
+    {noreply, NewState :: #state{}} |
+    {noreply, NewState :: #state{}, timeout() | hibernate} |
+    {stop, Reason :: term(), NewState :: #state{}}).
+handle_info({'DOWN', MonitorRef, process, SubscriberPid, Reason}, State = #state{reverse_tid = ReverseTid, pid_tid = PidTid}) ->
+    logger:debug("[endpoint_subscription] subscriber: ~p, down with reason: ~p", [SubscriberPid, Reason]),
+    Removed = ets:lookup(ReverseTid, SubscriberPid),
+    lists:foreach(fun(Reverse) -> delete_subscription(Reverse, State) end, Removed),
+    case ets:lookup(PidTid, SubscriberPid) of
+        [{SubscriberPid, MonitorRef, _Count}] ->
+            ets:delete(PidTid, SubscriberPid);
+        _ ->
+            ok
+    end,
+    {noreply, State};
+
+handle_info(Info, State = #state{}) ->
+    logger:debug("[endpoint_subscription] get unknown info: ~p", [Info]),
+    {noreply, State}.
+
+%% @private
+%% @doc This function is called by a gen_server when it is about to
+%% terminate. It should be the opposite of Module:init/1 and do any
+%% necessary cleaning up. When it returns, the gen_server terminates
+%% with Reason. The return value is ignored.
+-spec(terminate(Reason :: (normal | shutdown | {shutdown, term()} | term()),
+    State :: #state{}) -> term()).
+terminate(_Reason, _State = #state{}) ->
+    ok.
+
+%% @private
+%% @doc Convert process state when code is changed
+-spec(code_change(OldVsn :: term() | {down, term()}, State :: #state{},
+    Extra :: term()) ->
+    {ok, NewState :: #state{}} | {error, Reason :: term()}).
+code_change(_OldVsn, State = #state{}, _Extra) ->
+    {ok, State}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec match_route_key(binary()) -> [#subscriber{}].
+match_route_key(RouteKey) when is_binary(RouteKey) ->
+    ExactSubs = [Sub || {_Topic, Sub} <- ets:lookup(?EXACT_TAB, RouteKey)],
+    TrieSubs = match_trie(of_components(RouteKey)),
+    Sorted = lists:sort(fun compare_subscriber/2, ExactSubs ++ TrieSubs),
+    dedupe_subscribers(Sorted).
+
+-spec maybe_log_unmatched_publish(binary(), binary(), [#subscriber{}]) -> ok.
+maybe_log_unmatched_publish(RouteKey, Content, []) ->
+    endpoint_log:unmatched_publish(RouteKey, Content);
+maybe_log_unmatched_publish(_RouteKey, _Content, _MatchedSubscribers) ->
+    ok.
+
+-spec of_components(Topic :: binary()) -> [binary()].
+of_components(Topic) when is_binary(Topic) ->
+    binary:split(Topic, <<$/>>, [global]).
+
+is_valid_components([]) ->
+    true;
+is_valid_components([<<$+>>|T]) ->
+    length(T) =:= 0;
+is_valid_components([<<$*>>|T]) ->
+    is_valid_components(T);
+is_valid_components([_|T]) ->
+    is_valid_components(T).
+
+-spec order_num(Components :: list()) -> integer().
+order_num([]) ->
+    1;
+order_num([<<$*>>|_]) ->
+    2;
+order_num([<<$+>>|_]) ->
+    3;
+order_num([_|Tail]) ->
+    order_num(Tail).
+
+-spec has_subscription(ets:tid(), binary(), pid()) -> boolean().
+has_subscription(ReverseTid, Topic, SubscriberPid) ->
+    Subscriptions = ets:lookup(ReverseTid, SubscriberPid),
+    lists:any(fun({SubscriberPid0, Topic0, _Kind, _NodeId, _Sub}) ->
+        Topic =:= Topic0 andalso SubscriberPid =:= SubscriberPid0
+    end, Subscriptions).
+
+-spec compare_subscriber(#subscriber{}, #subscriber{}) -> boolean().
+compare_subscriber(#subscriber{order = Order0, topic = Topic0}, #subscriber{order = Order1, topic = Topic1}) ->
+    case Order0 =:= Order1 of
+        true ->
+            Topic0 =< Topic1;
+        false ->
+            Order0 < Order1
+    end.
+
+-spec dedupe_subscribers([#subscriber{}]) -> [#subscriber{}].
+dedupe_subscribers(Subscribers) ->
+    {_, Result} = lists:foldl(fun(S = #subscriber{subscriber_pid = SubscriberPid}, {Seen, Acc}) ->
+        case sets:is_element(SubscriberPid, Seen) of
+            true ->
+                {Seen, Acc};
+            false ->
+                {sets:add_element(SubscriberPid, Seen), [S | Acc]}
+        end
+    end, {sets:new(), []}, Subscribers),
+    lists:reverse(Result).
+
+-spec insert_subscription(binary(), [binary()], pid(), #subscriber{}, #state{}) -> #state{}.
+insert_subscription(Topic, Components, SubscriberPid, Sub, State = #state{exact_tid = ExactTid, trie_sub_tid = TrieSubTid, reverse_tid = ReverseTid}) ->
+    case has_wildcard(Components) of
+        false ->
+            true = ets:insert(ExactTid, {Topic, Sub}),
+            true = ets:insert(ReverseTid, {SubscriberPid, Topic, exact, undefined, Sub}),
+            State;
+        true ->
+            {Kind, NodeId, State1} = insert_trie_subscription(Components, State),
+            true = ets:insert(TrieSubTid, {{NodeId, subscription_match_type(Kind)}, Sub}),
+            true = ets:insert(ReverseTid, {SubscriberPid, Topic, Kind, NodeId, Sub}),
+            State1
+    end.
+
+-spec insert_trie_subscription([binary()], #state{}) -> {trie_exact | trie_plus, non_neg_integer(), #state{}}.
+insert_trie_subscription(Components, State) ->
+    case Components of
+        [] ->
+            {trie_exact, ?ROOT_NODE, State};
+        _ ->
+            case lists:last(Components) of
+                <<"+">> ->
+                    Prefix = lists:sublist(Components, length(Components) - 1),
+                    {NodeId, State1} = ensure_path(Prefix, State),
+                    {trie_plus, NodeId, State1};
+                _ ->
+                    {NodeId, State1} = ensure_path(Components, State),
+                    {trie_exact, NodeId, State1}
+            end
+    end.
+
+-spec subscription_match_type(trie_exact | trie_plus) -> exact | plus.
+subscription_match_type(trie_exact) ->
+    exact;
+subscription_match_type(trie_plus) ->
+    plus.
+
+-spec ensure_path([binary()], #state{}) -> {non_neg_integer(), #state{}}.
+ensure_path(Components, State) ->
+    ensure_path(?ROOT_NODE, Components, State).
+
+-spec ensure_path(non_neg_integer(), [binary()], #state{}) -> {non_neg_integer(), #state{}}.
+ensure_path(NodeId, [], State) ->
+    {NodeId, State};
+ensure_path(NodeId, [Segment | Rest], State = #state{edge_tid = EdgeTid, next_node_id = NextNodeId}) ->
+    EdgeKey = {NodeId, Segment},
+    case ets:lookup(EdgeTid, EdgeKey) of
+        [{EdgeKey, ChildNodeId}] ->
+            ensure_path(ChildNodeId, Rest, State);
+        [] ->
+            true = ets:insert(EdgeTid, {EdgeKey, NextNodeId}),
+            ensure_path(NextNodeId, Rest, State#state{next_node_id = NextNodeId + 1})
+    end.
+
+-spec delete_subscription(tuple(), #state{}) -> ok.
+delete_subscription(Reverse = {_SubscriberPid, Topic, exact, undefined, Sub}, #state{exact_tid = ExactTid, reverse_tid = ReverseTid}) ->
+    true = ets:delete_object(ExactTid, {Topic, Sub}),
+    true = ets:delete_object(ReverseTid, Reverse),
+    ok;
+delete_subscription(Reverse = {_SubscriberPid, _Topic, Kind, NodeId, Sub}, #state{trie_sub_tid = TrieSubTid, reverse_tid = ReverseTid}) ->
+    true = ets:delete_object(TrieSubTid, {{NodeId, subscription_match_type(Kind)}, Sub}),
+    true = ets:delete_object(ReverseTid, Reverse),
+    ok.
+
+-spec ensure_pid_monitor(pid(), #state{}) -> {reference(), #state{}}.
+ensure_pid_monitor(SubscriberPid, State = #state{pid_tid = PidTid}) ->
+    case ets:lookup(PidTid, SubscriberPid) of
+        [{SubscriberPid, MonitorRef, Count}] ->
+            true = ets:insert(PidTid, {SubscriberPid, MonitorRef, Count + 1}),
+            {MonitorRef, State};
+        [] ->
+            MonitorRef = erlang:monitor(process, SubscriberPid),
+            true = ets:insert(PidTid, {SubscriberPid, MonitorRef, 1}),
+            {MonitorRef, State}
+    end.
+
+-spec release_pid_monitor(pid(), non_neg_integer(), #state{}) -> #state{}.
+release_pid_monitor(_SubscriberPid, 0, State) ->
+    State;
+release_pid_monitor(SubscriberPid, RemovedCount, State = #state{pid_tid = PidTid}) ->
+    case ets:lookup(PidTid, SubscriberPid) of
+        [{SubscriberPid, MonitorRef, Count}] when Count =< RemovedCount ->
+            erlang:demonitor(MonitorRef, [flush]),
+            ets:delete(PidTid, SubscriberPid),
+            State;
+        [{SubscriberPid, MonitorRef, Count}] ->
+            true = ets:insert(PidTid, {SubscriberPid, MonitorRef, Count - RemovedCount}),
+            State;
+        [] ->
+            State
+    end.
+
+-spec match_trie([binary()]) -> [#subscriber{}].
+match_trie(Components) ->
+    match_trie([?ROOT_NODE], Components, []).
+
+-spec match_trie([non_neg_integer()], [binary()], [#subscriber{}]) -> [#subscriber{}].
+match_trie(NodeIds, [], Acc) ->
+    collect_trie_subscribers(NodeIds, exact) ++ Acc;
+match_trie(NodeIds, [Segment | Rest], Acc) ->
+    PlusSubs = collect_trie_subscribers(NodeIds, plus),
+    NextNodeIds = next_trie_nodes(NodeIds, Segment),
+    match_trie(NextNodeIds, Rest, PlusSubs ++ Acc).
+
+-spec collect_trie_subscribers([non_neg_integer()], exact | plus) -> [#subscriber{}].
+collect_trie_subscribers(NodeIds, MatchType) ->
+    lists:flatmap(fun(NodeId) ->
+        [Sub || {{NodeId0, MatchType0}, Sub} <- ets:lookup(?TRIE_SUB_TAB, {NodeId, MatchType}),
+            NodeId0 =:= NodeId, MatchType0 =:= MatchType]
+    end, NodeIds).
+
+-spec next_trie_nodes([non_neg_integer()], binary()) -> [non_neg_integer()].
+next_trie_nodes(NodeIds, Segment) ->
+    lists:usort(lists:flatmap(fun(NodeId) ->
+        lookup_child(NodeId, Segment) ++ lookup_child(NodeId, <<"*">>)
+    end, NodeIds)).
+
+-spec lookup_child(non_neg_integer(), binary()) -> [non_neg_integer()].
+lookup_child(NodeId, Segment) ->
+    EdgeKey = {NodeId, Segment},
+    case ets:lookup(?TRIE_EDGE_TAB, EdgeKey) of
+        [{EdgeKey, ChildNodeId}] ->
+            [ChildNodeId];
+        [] ->
+            []
+    end.
+
+-spec has_wildcard([binary()]) -> boolean().
+has_wildcard(Components) ->
+    lists:any(fun(Component) -> Component =:= <<"*">> orelse Component =:= <<"+">> end, Components).
+
+-spec exact_subscribers(ets:tid()) -> [#subscriber{}].
+exact_subscribers(ExactTid) ->
+    [Sub || {_Topic, Sub} <- ets:tab2list(ExactTid)].
+
+-spec trie_subscribers(ets:tid()) -> [#subscriber{}].
+trie_subscribers(TrieSubTid) ->
+    [Sub || {_Key, Sub} <- ets:tab2list(TrieSubTid)].

apps/endpoint/src/endpoint_sup.erl

@@ -0,0 +1,58 @@
+%%%-------------------------------------------------------------------
+%% @doc endpoint top level supervisor.
+%% @end
+%%%-------------------------------------------------------------------
+
+-module(endpoint_sup).
+
+-behaviour(supervisor).
+-include("endpoint.hrl").
+
+-export([start_link/0]).
+-export([init/1]).
+
+-define(SERVER, ?MODULE).
+
+start_link() ->
+    supervisor:start_link({local, ?SERVER}, ?MODULE, []).
+
+%% sup_flags() = #{strategy => strategy(),         % optional
+%%                 intensity => non_neg_integer(), % optional
+%%                 period => pos_integer()}        % optional
+%% child_spec() = #{id => child_id(),       % mandatory
+%%                  start => mfargs(),      % mandatory
+%%                  restart => restart(),   % optional
+%%                  shutdown => shutdown(), % optional
+%%                  type => worker(),       % optional
+%%                  modules => modules()}   % optional
+init([]) ->
+    SupFlags = #{strategy => one_for_all, intensity => 1000, period => 3600},
+    ChildSpecs = [
+        #{
+            id => endpoint_log,
+            start => {'endpoint_log', start_link, []},
+            restart => permanent,
+            shutdown => 2000,
+            type => worker,
+            modules => ['endpoint_log']
+        },
+
+        #{
+            id => endpoint_subscription,
+            start => {'endpoint_subscription', start_link, []},
+            restart => permanent,
+            shutdown => 2000,
+            type => worker,
+            modules => ['endpoint_subscription']
+        },
+
+        #{
+            id => 'endpoint_adapter_sup',
+            start => {'endpoint_adapter_sup', start_link, []},
+            restart => permanent,
+            shutdown => 2000,
+            type => supervisor,
+            modules => ['endpoint_adapter_sup']
+        }
+    ],
+    {ok, {SupFlags, ChildSpecs}}.

apps/endpoint/src/endpoint_tester.erl

@@ -0,0 +1,93 @@
+%%%-------------------------------------------------------------------
+%%% @author anlicheng
+%%% @copyright (C) 2026, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 11. 5月 2026 20:16
+%%%-------------------------------------------------------------------
+-module(endpoint_tester).
+-author("anlicheng").
+-include("endpoint.hrl").
+
+%% API
+-export([test/1]).
+
+%% http接口的测试
+test(#http_endpoint{url = Url, pool_size = PoolSize}) when is_integer(PoolSize), PoolSize > 0 ->
+    Body = <<"">>,
+    ContentType = "application/json",
+    case httpc:request(post, {Url, [], ContentType, Body}, [], []) of
+        {ok, _} ->
+            ok;
+        {error, Reason} ->
+            {error, Reason}
+    end;
+
+%% 测试mqtt
+test(#mqtt_endpoint{host = Host, port = Port, username = Username, password = Password}) ->
+    %% 用户测试的client
+    ClientId = "mqtt_client_test:" ++ iot_util:rand_bytes(16),
+    Opts = [
+        {owner, self()},
+        {clientid, ClientId},
+        {host, binary_to_list(Host)},
+        {port, Port},
+        {tcp_opts, []},
+        {username, binary_to_list(Username)},
+        {password, binary_to_list(Password)},
+        {keepalive, 86400},
+        {auto_ack, true},
+        {connect_timeout, 5000},
+        {proto_ver, v5},
+        {retry_interval, 5000}
+    ],
+
+    case emqtt:start_link(Opts) of
+        {ok, ConnPid} ->
+            case catch emqtt:connect(ConnPid, 5000) of
+                {ok, _} ->
+                    emqtt:stop(ConnPid),
+                    ok;
+                {error, _Reason} ->
+                    {error, <<"connect mqtt server failed">>};
+                _Error ->
+                    emqtt:stop(ConnPid),
+                    {error, <<"connect mqtt server failed">>}
+            end;
+        Other ->
+            logger:warning("[endpint_handler] test connect mqtt with options: ~p, get error: ~p", [Opts, Other]),
+            {error, <<"connect mqtt server failed">>}
+    end;
+
+%% 测试kafka
+test(#kafka_endpoint{sasl_config = SaslConfig, bootstrap_servers = BootstrapServers, topic = Topic}) ->
+    BaseConfig = [
+        {reconnect_cool_down_seconds, 5},
+        {socket_options, [{keepalive, true}]}
+    ],
+
+    ClientConfig = case SaslConfig of
+                       {Mechanism, Username, Password} ->
+                           [{sasl, {Mechanism, Username, Password}}|BaseConfig];
+                       undefined ->
+                           BaseConfig
+                   end,
+    ClientId = brod_client_test,
+    _ = catch brod:stop_client(ClientId),
+
+    case catch brod:start_link_client(BootstrapServers, ClientId, ClientConfig) of
+        {ok, _ClientPid} ->
+            case brod:start_producer(ClientId, Topic, _ProducerConfig = []) of
+                ok ->
+                    ok = brod:stop_client(ClientId),
+                    ok;
+                {error, Reason} ->
+                    logger:debug("[endpint_handler] start_producer: ~p, get error: ~p", [ClientId, Reason]),
+                    _ = catch brod:stop_client(ClientId),
+                    {error, <<"config kafka server failed">>}
+            end;
+        Error ->
+            logger:debug("[endpint_handler] start_client: ~p, get error: ~p", [ClientId, Error]),
+            {error, <<"config kafka server failed">>}
+    end.

apps/endpoint/src/endpoint_util.erl

@@ -0,0 +1,18 @@
+%%%-------------------------------------------------------------------
+%%% @doc Endpoint-local utility functions.
+%%%-------------------------------------------------------------------
+-module(endpoint_util).
+
+-export([set_metadata/0, sha256/1]).
+
+-spec set_metadata() -> ok.
+set_metadata() ->
+    logger:set_process_metadata(#{domain => [endpoint]}).
+
+-spec sha256(string() | binary()) -> binary().
+sha256(Str) when is_list(Str) ->
+    sha256(unicode:characters_to_binary(Str));
+sha256(Bin) when is_binary(Bin) ->
+    HashBin = crypto:hash(sha256, Bin),
+    HexStr = lists:flatten([io_lib:format("~2.16.0B", [B]) || B <- binary:bin_to_list(HashBin)]),
+    list_to_binary(string:lowercase(HexStr)).

apps/iot/include/domain_model.hrl

@@ -0,0 +1,20 @@
+%%%-------------------------------------------------------------------
+%%% @author Codex
+%%% @doc
+%%% 领域模型定义
+%%% @end
+%%%-------------------------------------------------------------------
+
+-record(host_info, {
+    id :: integer(),
+    uuid :: binary(),
+    authorize_status :: integer(),
+    status :: integer()
+}).
+
+-record(device_info, {
+    id :: integer(),
+    host_id :: integer(),
+    device_uuid :: binary(),
+    status :: integer()
+}).

apps/iot/include/iot.hrl

@@ -13,6 +13,10 @@
 -define(HOST_ONLINE, 1).
 -define(HOST_NOT_JOINED, -1).
 
+%% 主机是否授权处理上报数据
+-define(HOST_DENIED, 0).
+-define(HOST_AUTHORIZED, 1).
+
 %% 设备是否在线状态
 -define(DEVICE_OFFLINE, 0).
 -define(DEVICE_ONLINE, 1).
@@ -22,93 +26,3 @@
 -define(TASK_STATUS_INIT,   -1).  %% 未接入
 -define(TASK_STATUS_FAILED, 0).    %% 离线
 -define(TASK_STATUS_OK,     1).     %% 在线
-
-%% efka主动发起的消息体类型, 消息大类
--define(PACKET_REQUEST, 16#01).
--define(PACKET_RESPONSE, 16#02).
-
-%% 服务器基于pub/sub的消息, 消息大类
--define(PACKET_PUB, 16#03).
-
-%% push调用不需要返回, 消息大类
--define(PACKET_COMMAND, 16#04).
-
-%% 服务器端推送消息
--define(PACKET_ASYNC_CALL, 16#05).
--define(PACKET_ASYNC_CALL_REPLY, 16#06).
-
-%% ping包，客户端主动发起
--define(PACKET_PING, 16#FF).
-
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-%%%% 二级分类定义
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-
-%% 主机端上报数据类型标识
--define(METHOD_AUTH, 16#01).
--define(METHOD_DATA, 16#02).
--define(METHOD_PING, 16#03).
--define(METHOD_INFORM, 16#04).
--define(METHOD_EVENT, 16#05).
--define(METHOD_PHASE, 16#06).
--define(METHOD_REQUEST_SERVICE_CONFIG, 16#07).
-
-%%%% 命令类型子分类, 不需要返回值
-
-%% 授权
--define(COMMAND_AUTH, 16#08).
-
-%%%% 主动推送的消息类型子分类, 需要返回值
-
--define(PUSH_DEPLOY, 16#01).
--define(PUSH_START_SERVICE, 16#02).
--define(PUSH_STOP_SERVICE, 16#03).
-
--define(PUSH_SERVICE_CONFIG, 16#04).
--define(PUSH_INVOKE, 16#05).
--define(PUSH_TASK_LOG, 16#06).
-
-%% 缓存数据库表
--record(kv, {
-    key :: binary(),
-    val :: binary() | list() | map() | sets:set(),
-    expire_at = 0 :: integer(),
-    type :: atom()
-}).
-
-%% id生成器
--record(id_generator, {
-    tab :: atom(),
-    increment_id = 0 :: integer()
-}).
-
-%% 统计项
--record(option, {
-    success_num = 0,
-    fail_num = 0
-}).
-
-%% 北向数据
--record(north_data, {
-    id = 0 :: integer(),
-    location_code :: binary(),
-    %% 数据库类型的endpoint， 可以返回list: [{K, V}, {K1, V1}]
-    fields :: [{K :: binary(), V :: any()}],
-    timestamp = 0 :: integer()
-}).
-
-%% 事件相关的数据
--record(event_data, {
-    id = 0 :: integer(),
-    location_code :: binary(),
-    event_type :: integer(),
-    params :: map()
-}).
-
-%% 发送数据
--record(post_data, {
-    id = 0 :: integer(),
-    location_code :: binary(),
-    %% 数据库类型的endpoint， 可以返回list: [{K, V}, {K1, V1}]
-    body :: binary() | list()
-}).

apps/iot/include/iot_tables.hrl

@@ -1,23 +1,17 @@
 %%%-------------------------------------------------------------------
 %%% @author anlicheng
-%%% @copyright (C) 2025, <COMPANY>
+%%% @copyright (C) 2026, <COMPANY>
 %%% @doc
 %%%
 %%% @end
-%%% Created : 08. 5月 2025 12:08
+%%% Created : 24. 4月 2026 16:54
 %%%-------------------------------------------------------------------
 -author("anlicheng").
 
-%% 用来保存微服务的配置
--record(service_config, {
-    service_id :: binary(),
-    host_uuid :: binary(),
-    config_json = <<>> :: binary(),
-    %% 保持上一个版本的内容，错误时回滚
-    last_config_json = <<>> :: binary(),
-    %% 最后一次修改的用户id
-    last_edit_user :: integer(),
-    %% 状态： 0: 停止, 1: 运行中
-    update_ts = 0,
-    create_ts = 0
+-record(efka_client, {
+    uuid :: binary(),
+    token_hash :: binary(),
+    salt :: binary(),
+    heartbeat_secret :: binary(),
+    timestamp = 0 :: integer()
 }).

apps/iot/include/message_pb.hrl

@@ -1,133 +0,0 @@
-%% -*- coding: utf-8 -*-
-%% Automatically generated, do not edit
-%% Generated by gpb_compile version 4.21.1
-
--ifndef(message_pb).
--define(message_pb, true).
-
--define(message_pb_gpb_version, "4.21.1").
-
-
--ifndef('AUTH_REQUEST_PB_H').
--define('AUTH_REQUEST_PB_H', true).
--record(auth_request,
-        {uuid = <<>>            :: unicode:chardata() | undefined, % = 1, optional
-         username = <<>>        :: unicode:chardata() | undefined, % = 2, optional
-         salt = <<>>            :: unicode:chardata() | undefined, % = 4, optional
-         token = <<>>           :: unicode:chardata() | undefined, % = 5, optional
-         timestamp = 0          :: non_neg_integer() | undefined % = 6, optional, 32 bits
-        }).
--endif.
-
--ifndef('AUTH_REPLY_PB_H').
--define('AUTH_REPLY_PB_H', true).
--record(auth_reply,
-        {code = 0               :: non_neg_integer() | undefined, % = 1, optional, 32 bits
-         message = <<>>         :: unicode:chardata() | undefined % = 2, optional
-        }).
--endif.
-
--ifndef('PUB_PB_H').
--define('PUB_PB_H', true).
--record(pub,
-        {topic = <<>>           :: unicode:chardata() | undefined, % = 1, optional
-         content = <<>>         :: iodata() | undefined % = 2, optional
-        }).
--endif.
-
--ifndef('COMMAND_PB_H').
--define('COMMAND_PB_H', true).
--record(command,
-        {command_type = <<>>    :: unicode:chardata() | undefined, % = 1, optional
-         command = <<>>         :: iodata() | undefined % = 2, optional
-        }).
--endif.
-
--ifndef('RPC_DEPLOY_PB_H').
--define('RPC_DEPLOY_PB_H', true).
--record(rpc_deploy,
-        {packet_id = 0          :: non_neg_integer() | undefined, % = 1, optional, 32 bits
-         task_id = 0            :: non_neg_integer() | undefined, % = 2, optional, 32 bits
-         config = <<>>          :: unicode:chardata() | undefined % = 3, optional
-        }).
--endif.
-
--ifndef('RPC_START_CONTAINER_PB_H').
--define('RPC_START_CONTAINER_PB_H', true).
--record(rpc_start_container,
-        {packet_id = 0          :: non_neg_integer() | undefined, % = 1, optional, 32 bits
-         container_name = <<>>  :: unicode:chardata() | undefined % = 2, optional
-        }).
--endif.
-
--ifndef('RPC_STOP_CONTAINER_PB_H').
--define('RPC_STOP_CONTAINER_PB_H', true).
--record(rpc_stop_container,
-        {packet_id = 0          :: non_neg_integer() | undefined, % = 1, optional, 32 bits
-         container_name = <<>>  :: unicode:chardata() | undefined % = 2, optional
-        }).
--endif.
-
--ifndef('RPC_CONFIG_CONTAINER_PB_H').
--define('RPC_CONFIG_CONTAINER_PB_H', true).
--record(rpc_config_container,
-        {packet_id = 0          :: non_neg_integer() | undefined, % = 1, optional, 32 bits
-         container_name = <<>>  :: unicode:chardata() | undefined, % = 2, optional
-         config = <<>>          :: iodata() | undefined % = 3, optional
-        }).
--endif.
-
--ifndef('FETCH_TASK_LOG_PB_H').
--define('FETCH_TASK_LOG_PB_H', true).
--record(fetch_task_log,
-        {task_id = 0            :: non_neg_integer() | undefined % = 1, optional, 32 bits
-        }).
--endif.
-
--ifndef('CONTAINER_CONFIG_PB_H').
--define('CONTAINER_CONFIG_PB_H', true).
--record(container_config,
-        {container_name = <<>>  :: unicode:chardata() | undefined, % = 1, optional
-         config = <<>>          :: iodata() | undefined % = 2, optional
-        }).
--endif.
-
--ifndef('DATA_PB_H').
--define('DATA_PB_H', true).
--record(data,
-        {service_id = <<>>      :: unicode:chardata() | undefined, % = 1, optional
-         device_uuid = <<>>     :: unicode:chardata() | undefined, % = 2, optional
-         route_key = <<>>       :: unicode:chardata() | undefined, % = 3, optional
-         metric = <<>>          :: iodata() | undefined % = 4, optional
-        }).
--endif.
-
--ifndef('EVENT_PB_H').
--define('EVENT_PB_H', true).
--record(event,
-        {service_id = <<>>      :: unicode:chardata() | undefined, % = 1, optional
-         event_type = 0         :: non_neg_integer() | undefined, % = 2, optional, 32 bits
-         params = <<>>          :: unicode:chardata() | undefined % = 3, optional
-        }).
--endif.
-
--ifndef('PING_PB_H').
--define('PING_PB_H', true).
--record(ping,
-        {adcode = <<>>          :: unicode:chardata() | undefined, % = 1, optional
-         boot_time = 0          :: non_neg_integer() | undefined, % = 2, optional, 32 bits
-         province = <<>>        :: unicode:chardata() | undefined, % = 3, optional
-         city = <<>>            :: unicode:chardata() | undefined, % = 4, optional
-         efka_version = <<>>    :: unicode:chardata() | undefined, % = 5, optional
-         kernel_arch = <<>>     :: unicode:chardata() | undefined, % = 6, optional
-         ips = []               :: [unicode:chardata()] | undefined, % = 7, repeated
-         cpu_core = 0           :: non_neg_integer() | undefined, % = 8, optional, 32 bits
-         cpu_load = 0           :: non_neg_integer() | undefined, % = 9, optional, 32 bits
-         cpu_temperature = 0.0  :: float() | integer() | infinity | '-infinity' | nan | undefined, % = 10, optional
-         disk = []              :: [integer()] | undefined, % = 11, repeated, 32 bits
-         memory = []            :: [integer()] | undefined, % = 12, repeated, 32 bits
-         interfaces = <<>>      :: unicode:chardata() | undefined % = 13, optional
-        }).
--endif.
-
--endif.

apps/iot/include/protocol.hrl

@@ -0,0 +1,16 @@
+%%%-------------------------------------------------------------------
+%%% @author anlicheng
+%%% @copyright (C) 2025, <COMPANY>
+%%% @doc
+%%% 传输层协议常量定义
+%%% @end
+%%%-------------------------------------------------------------------
+-author("anlicheng").
+
+%% 一级帧类型
+%% REQUEST: 需要响应的请求帧
+%% REPLY: 对 REQUEST 的响应帧
+%% CAST: 单向消息帧，不需要响应
+-define(FRAME_REQUEST, 16#01).
+-define(FRAME_REPLY, 16#02).
+-define(FRAME_CAST, 16#03).

apps/iot/priv/2024-01-18-log.txt

@@ -1,20 +0,0 @@
-{"device_uuid":"20118448129232486417014256831677", "key":"epi", "label":"kWh", "name":"正向总有功电能", "timestamp":1705576081, "type":"AI", "unit":5, "value":314498}
-{"device_uuid":"20118448129232486417014256831677", "key":"a_voltage", "label":"V", "name":"A相电压", "timestamp":1705576081, "type":"AI", "unit":0, "value":224.1}
-{"device_uuid":"20118448129232486417014256831677", "key":"a_current", "label":"A", "name":"A相电流", "timestamp":1705576081, "type":"AI", "unit":3, "value":0.394}
-{"device_uuid":"20118448129232486417014256831677", "key":"b_voltage", "label":"V", "name":"B相电压", "timestamp":1705576081, "type":"AI", "unit":0, "value":225.3}
-{"device_uuid":"20118448129232486417014256831677", "key":"b_current", "label":"A", "name":"B相电流", "timestamp":1705576081, "type":"AI", "unit":3, "value":0.323}
-{"device_uuid":"20118448129232486417014256831677", "key":"c_voltage", "label":"V", "name":"C相电压", "timestamp":1705576081, "type":"AI", "unit":0, "value":225}
-{"device_uuid":"20118448129232486417014256831677", "key":"c_current", "label":"A", "name":"C相电流", "timestamp":1705576081, "type":"AI", "unit":3, "value":0.269}
-{"device_uuid":"20118448129232486417014256831677", "key":"active_power", "label":"kW", "name":"瞬时总有功功率", "timestamp":1705576081, "type":"AI", "unit":23, "value":20.91}
-{"device_uuid":"20118448129232486417014256831677", "key":"power_factor", "label":"", "name":"总功率因数", "timestamp":1705576081, "type":"AI", "unit":16, "value":0.994}
-{"device_uuid":"20118448549921177617014256841701", "key":"epi", "label":"kWh", "name":"正向总有功电能", "timestamp":1705576038, "type":"AI", "unit":5, "value":299493.6}
-{"device_uuid":"20118448549921177617014256841701", "key":"a_voltage", "label":"V", "name":"A相电压", "timestamp":1705576038, "type":"AI", "unit":0, "value":225.9}
-{"device_uuid":"20118448549921177617014256841701", "key":"a_current", "label":"A", "name":"A相电流", "timestamp":1705576038, "type":"AI", "unit":3, "value":0.318}
-{"device_uuid":"20118448549921177617014256841701", "key":"b_voltage", "label":"V", "name":"B相电压", "timestamp":1705576038, "type":"AI", "unit":0, "value":225.7}
-{"device_uuid":"20118448549921177617014256841701", "key":"b_current", "label":"A", "name":"B相电流", "timestamp":1705576038, "type":"AI", "unit":3, "value":0.305}
-{"device_uuid":"20118448549921177617014256841701", "key":"c_voltage", "label":"V", "name":"C相电压", "timestamp":1705576038, "type":"AI", "unit":0, "value":225}
-{"device_uuid":"20118448549921177617014256841701", "key":"c_current", "label":"A", "name":"C相电流", "timestamp":1705576038, "type":"AI", "unit":3, "value":0.3}
-{"device_uuid":"20118448549921177617014256841701", "key":"active_power", "label":"kW", "name":"瞬时总有功功率", "timestamp":1705576038, "type":"AI", "unit":23, "value":15.056}
-{"device_uuid":"20118448549921177617014256841701", "key":"power_factor", "label":"", "name":"总功率因数", "timestamp":1705576038, "type":"AI", "unit":16, "value":1}
-{"device_uuid":"20118448970190438417014256851724", "key":"epi", "label":"kWh", "name":"正向总有功电能", "timestamp":1705576117, "type":"AI", "unit":5, "value":285011}
-{"device_uuid":"20118448970190438417014256851724", "key":"a_voltage", "label":"V", "name":"A相电压", "timestamp":1705576117, "type":"AI", "unit":0, "value":225.1}

apps/iot/priv/ssl/server.crt

@@ -0,0 +1,23 @@
+-----BEGIN CERTIFICATE-----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+-----END CERTIFICATE-----

apps/iot/priv/ssl/server.csr

@@ -0,0 +1,18 @@
+-----BEGIN CERTIFICATE REQUEST-----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+-----END CERTIFICATE REQUEST-----

apps/iot/priv/ssl/server.key

@@ -0,0 +1,28 @@
+-----BEGIN PRIVATE KEY-----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+-----END PRIVATE KEY-----

apps/iot/src/adapters/control_api/iot_api_client.erl

@@ -0,0 +1,259 @@
+%%%-------------------------------------------------------------------
+%%% @author anlicheng
+%%% @copyright (C) 2023, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 24. 12月 2023 15:42
+%%%-------------------------------------------------------------------
+-module(iot_api_client).
+-author("anlicheng").
+-include("domain_model.hrl").
+
+%% API
+-export([ai_event/1]).
+
+-define(API_TOKEN, <<"wv6fGyBhl*7@AsD9">>).
+
+-export([get_all_hosts/0, get_host_by_id/1, get_host_by_uuid/1, change_host_status/2]).
+-export([get_host_devices/1, get_device_by_uuid/1, change_device_status/2]).
+-export([get_all_endpoints/0, get_endpoint/1]).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec get_all_hosts() -> [HostUUID :: binary()].
+get_all_hosts() ->
+    case do_get("/get_all_hosts", []) of
+        {ok, Ids} ->
+            Ids;
+        _ ->
+            []
+    end.
+
+-spec get_host_by_uuid(UUID :: binary()) -> undefined | {ok, HostInfo :: #host_info{}}.
+get_host_by_uuid(UUID) when is_binary(UUID) ->
+    case do_get("/get_host_by_uuid", [{<<"uuid">>, UUID}]) of
+        {ok, HostInfo} ->
+            case host_info_record(HostInfo) of
+                {ok, Record} ->
+                    {ok, Record};
+                error ->
+                    undefined
+            end;
+        _ ->
+            undefined
+    end.
+
+-spec get_host_by_id(HostId :: integer()) -> undefined | {ok, HostInfo :: #host_info{}}.
+get_host_by_id(HostId) when is_integer(HostId) ->
+    case do_get("/get_host_by_id", [{<<"host_id">>, integer_to_binary(HostId)}]) of
+        {ok, HostInfo} ->
+            case host_info_record(HostInfo) of
+                {ok, Record} ->
+                    {ok, Record};
+                error ->
+                    undefined
+            end;
+        _ ->
+            undefined
+    end.
+
+%% 修改主机的状态
+-spec change_host_status(UUID :: binary(), Status :: integer()) -> {ok, Result :: any()} | {error, Reason :: any()}.
+change_host_status(UUID, NStatus) when is_binary(UUID), is_integer(NStatus) ->
+    do_post("/change_host_status", #{<<"uuid">> => UUID, <<"new_status">> => NStatus}).
+
+-spec get_host_devices(HostId :: integer()) -> {ok, Devices :: [#device_info{}]} | {error, Reason::any()}.
+get_host_devices(HostId) when is_integer(HostId) ->
+    case do_get("/get_host_devices", [{<<"host_id">>, integer_to_binary(HostId)}]) of
+        {ok, DeviceInfos} when is_list(DeviceInfos) ->
+            device_info_records(DeviceInfos);
+        {ok, _Other} ->
+            {error, invalid_device_infos};
+        Error ->
+            Error
+    end.
+
+-spec get_device_by_uuid(DeviceUUID :: binary()) -> {ok, DeviceInfo :: #device_info{}} | undefined.
+get_device_by_uuid(DeviceUUID) when is_binary(DeviceUUID) ->
+    case do_get("/get_device_by_uuid", [{<<"device_uuid">>, DeviceUUID}]) of
+        {ok, DeviceInfo} ->
+            case device_info_record(DeviceInfo) of
+                {ok, Record} ->
+                    {ok, Record};
+                error ->
+                    undefined
+            end;
+        _ ->
+            undefined
+    end.
+
+%% 修改主机的状态
+-spec change_device_status(DeviceUUID :: binary(), Status :: integer()) -> {ok, AffectedRows :: integer()} | {error, Reason :: any()}.
+change_device_status(DeviceUUID, NStatus) when is_binary(DeviceUUID), is_integer(NStatus) ->
+    do_post("/change_device_status", #{<<"device_uuid">> => DeviceUUID, <<"new_status">> => NStatus}).
+
+%%%-------------------------------------------------------------------
+%% endpoint相关的api
+%%%-------------------------------------------------------------------
+%% API
+
+-spec get_all_endpoints() -> [Endpoint :: map()].
+get_all_endpoints() ->
+    case do_get("/get_all_endpoints", []) of
+        {ok, Endpoints} ->
+            Endpoints;
+        _ ->
+            []
+    end.
+
+-spec get_endpoint(Id :: integer()) -> undefined | {ok, EndpointInfo :: map()}.
+get_endpoint(Id) when is_integer(Id) ->
+    case do_get("/get_endpoint", [{<<"id">>, integer_to_binary(Id)}]) of
+        {ok, EndpointInfo} when is_map(EndpointInfo) ->
+            {ok, EndpointInfo};
+        _ ->
+            undefined
+    end.
+
+ai_event(Id) when is_integer(Id) ->
+    Token = iot_util:md5(<<?API_TOKEN/binary, (integer_to_binary(Id))/binary, ?API_TOKEN/binary>>),
+    {ok, Url} = application:get_env(iot, api_url),
+
+    Headers = [
+        {<<"content-type">>, <<"application/json">>}
+    ],
+    ReqData = #{
+        <<"token">> => Token,
+        <<"id">> => Id
+    },
+    Body = iolist_to_binary(json:encode(ReqData)),
+    case hackney:request(post, Url, Headers, Body, [{pool, false}]) of
+        {ok, 200, _, ClientRef} ->
+            {ok, RespBody} = hackney:body(ClientRef),
+            logger:debug("[iot_api_client] send body: ~p, get error is: ~p", [Body, RespBody]),
+            hackney:close(ClientRef);
+        {ok, HttpCode, _, ClientRef} ->
+            {ok, RespBody} = hackney:body(ClientRef),
+            hackney:close(ClientRef),
+            logger:warning("[iot_api_client] send body: ~p, get error is: ~p", [Body, {HttpCode, RespBody}]);
+        {error, Reason} ->
+            logger:warning("[iot_api_client] send body: ~p, get error is: ~p", [Body, Reason])
+    end.
+
+%%%-------------------------------------------------------------------
+%% helper methods
+%%%-------------------------------------------------------------------
+
+-spec do_post(Path :: string(), Params :: map()) -> {ok, Resp :: any()} | {error, Reason :: any()}.
+do_post(Path, Params) when is_list(Path), is_map(Params) ->
+    {ok, BaseUrl} = application:get_env(iot, api_url),
+    Headers = [
+        {<<"content-type">>, <<"application/json">>},
+        {<<"Accept">>, <<"application/json">>}
+    ],
+    Url = BaseUrl ++ Path,
+    Body = iolist_to_binary(json:encode(Params)),
+    case hackney:request(post, Url, Headers, Body, [{pool, false}]) of
+        {ok, 200, _, ClientRef} ->
+            {ok, RespBody} = hackney:body(ClientRef),
+            logger:debug("[iot_api_client] request url: ~p, send body: ~p, get response is: ~p", [Url, Body, RespBody]),
+            hackney:close(ClientRef),
+            case catch json:decode(RespBody) of
+                #{<<"result">> := Result} ->
+                    {ok, Result};
+                #{<<"error">> := #{<<"code">> := Code, <<"message">> := Message}} ->
+                    {error, {Code, Message}};
+                {error, Reason} ->
+                    {error, Reason};
+                Other ->
+                    {error, Other}
+            end;
+        {ok, HttpCode, _, ClientRef} ->
+            {ok, RespBody} = hackney:body(ClientRef),
+            hackney:close(ClientRef),
+            logger:warning("[iot_api_client] request url: ~p, send body: ~p, get error is: ~p", [Url, Body, {HttpCode, RespBody}]),
+            {error, {HttpCode, RespBody}};
+        {error, Reason} ->
+            logger:warning("[iot_api_client] request url: ~p, send body: ~p, get error is: ~p", [Url, Body, Reason]),
+            {error, Reason}
+    end.
+
+-spec do_get(Path :: string(), Params :: [{Key :: binary(), Val :: binary()}]) -> {ok, Resp :: any()} | {error, Reason :: any()}.
+do_get(Path, Params) when is_list(Path), is_list(Params) ->
+    {ok, BaseUrl} = application:get_env(iot, api_url),
+    Headers = [
+        {<<"Accept">>, <<"application/json">>}
+    ],
+
+    Url = case length(Params) > 0 of
+              true ->
+                  QS = binary_to_list(uri_string:compose_query(Params)),
+                  BaseUrl ++ Path ++ "?" ++ QS;
+              false ->
+                  BaseUrl ++ Path
+          end,
+
+    case hackney:request(get, Url, Headers, <<>>, [{pool, false}]) of
+        {ok, 200, _, ClientRef} ->
+            {ok, RespBody} = hackney:body(ClientRef),
+            hackney:close(ClientRef),
+            logger:debug("[iot_api_client] url: ~p, get response is: ~p", [Url, RespBody]),
+            case catch json:decode(RespBody) of
+                #{<<"result">> := Result} ->
+                    {ok, Result};
+                #{<<"error">> := #{<<"code">> := Code, <<"message">> := Message}} ->
+                    {error, {Code, Message}};
+                {error, Reason} ->
+                    {error, Reason};
+                Other ->
+                    {error, Other}
+            end;
+        {ok, HttpCode, _, ClientRef} ->
+            {ok, RespBody} = hackney:body(ClientRef),
+            hackney:close(ClientRef),
+            logger:warning("[iot_api_client] request url: ~p, get error is: ~p", [Url, {HttpCode, RespBody}]),
+            {error, {HttpCode, RespBody}};
+        {error, Reason} ->
+            logger:warning("[iot_api_client] request url: ~p, get error is: ~p", [Url, Reason]),
+            {error, Reason}
+    end.
+
+-spec host_info_record(map()) -> {ok, #host_info{}} | error.
+host_info_record(#{<<"id">> := Id, <<"uuid">> := UUID, <<"authorize_status">> := AuthorizeStatus, <<"status">> := Status})
+        when is_integer(Id), is_binary(UUID), is_integer(AuthorizeStatus), is_integer(Status) ->
+    {ok, #host_info{
+        id = Id,
+        uuid = UUID,
+        authorize_status = AuthorizeStatus,
+        status = Status
+    }};
+host_info_record(_) ->
+    error.
+
+-spec device_info_record(map()) -> {ok, #device_info{}} | error.
+device_info_record(#{<<"id">> := Id, <<"host_id">> := HostId, <<"device_uuid">> := DeviceUUID, <<"status">> := Status})
+        when is_integer(Id), is_integer(HostId), is_binary(DeviceUUID), is_integer(Status) ->
+    {ok, #device_info{
+        id = Id,
+        host_id = HostId,
+        device_uuid = DeviceUUID,
+        status = Status
+    }};
+device_info_record(_) ->
+    error.
+
+-spec device_info_records([map()]) -> {ok, [#device_info{}]} | {error, invalid_device_info}.
+device_info_records(DeviceInfos) ->
+    lists:foldr(fun(DeviceInfo, Acc) ->
+        case {device_info_record(DeviceInfo), Acc} of
+            {{ok, Record}, {ok, Records}} ->
+                {ok, [Record | Records]};
+            {error, _} ->
+                {error, invalid_device_info};
+            {_, Error = {error, _}} ->
+                Error
+        end
+    end, {ok, []}, DeviceInfos).

apps/iot/src/adapters/influxdb/influx_client.erl

@@ -57,7 +57,7 @@ get_precision(Timestamp) when is_integer(Timestamp) ->
             <<"ms">>
     end.
 
--spec write_data(Measurement :: binary(), Tags :: map(), FieldsList :: list(), Timestamp :: integer()) -> no_return().
+-spec write_data(Measurement :: binary(), Tags :: map(), FieldsList :: list(), Timestamp :: integer()) -> term().
 write_data(Measurement, Tags, FieldsList, Timestamp) when is_binary(Measurement), is_map(Tags), is_list(FieldsList), is_integer(Timestamp) ->
     %% 过来掉没有key的选项
     NFieldsList = lists:filter(fun data_filter/1, FieldsList),
@@ -76,12 +76,12 @@ write_data(Measurement, Tags, FieldsList, Timestamp) when is_binary(Measurement)
             ok
     end.
 
--spec write(Pid :: pid(), Bucket :: binary(), Org :: binary(), Points :: list()) -> no_return().
+-spec write(Pid :: pid(), Bucket :: binary(), Org :: binary(), Points :: list()) -> ok.
 write(Pid, Bucket, Org, Points) when is_pid(Pid), is_binary(Bucket), is_binary(Org), is_list(Points) ->
     write(Pid, Bucket, Org, <<"ms">>, Points).
 
 %% Precision的值为: ms|ns|s; 表示时间的精度，默认为毫秒(ms)
--spec write(Pid :: pid(), Bucket :: binary(), Org :: binary(), Precision :: binary(), Points :: list()) -> no_return().
+-spec write(Pid :: pid(), Bucket :: binary(), Org :: binary(), Precision :: binary(), Points :: list()) -> ok.
 write(Pid, Bucket, Org, Precision, Points) when is_pid(Pid), is_binary(Bucket), is_binary(Org), is_binary(Precision), is_list(Points) ->
     gen_server:cast(Pid, {write, Bucket, Org, Precision, Points}).
 
@@ -101,6 +101,7 @@ start_link(Opts) when is_list(Opts) ->
     {ok, State :: #state{}} | {ok, State :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term()} | ignore).
 init([InfluxProps]) ->
+    ok = iot_log:set_metadata(),
     Token = proplists:get_value(token, InfluxProps),
     Host = proplists:get_value(host, InfluxProps),
     Port = proplists:get_value(port, InfluxProps),
@@ -151,18 +152,18 @@ handle_cast({write, Bucket, Org, Precision, Points}, State = #state{host = Host,
         query => Query
     }),
 
-    lager:debug("[influx_client] url is: ~p, headers: ~p, body: ~ts", [Url, Headers, Body]),
+    logger:debug("[influx_client] url is: ~p, headers: ~p, body: ~ts", [Url, Headers, Body]),
     case hackney:request(post, Url, Headers, GZipBody, [{pool, false}]) of
         {ok, StatusCode, _RespHeaders, ClientRef} ->
             case hackney:body(ClientRef) of
                 {ok, RespBody} ->
-                    lager:debug("[influx_client] status_code: ~p, response body is: ~p", [StatusCode, RespBody]);
+                    logger:debug("[influx_client] status_code: ~p, response body is: ~p", [StatusCode, RespBody]);
                 {error, Error} ->
-                    lager:warning("[influx_client] status_code: ~p, error is: ~p", [Error])
+                    logger:warning("[influx_client] status_code: ~p, error is: ~p", [Error])
             end,
             hackney:close(ClientRef);
         {error, Reason} ->
-            lager:warning("[influx_client] request result is: ~p", [Reason])
+            logger:warning("[influx_client] request result is: ~p", [Reason])
     end,
 
     {noreply, State}.

apps/iot/src/adapters/influxdb/influx_point.erl

@@ -45,7 +45,7 @@ field_val(V) when is_float(V) ->
 field_val(V) when is_binary(V) ->
     <<$", V/binary, $">>;
 field_val(V) when is_list(V); is_map(V) ->
-    S = base64:encode(iolist_to_binary(jiffy:encode(V, [force_utf8]))),
+    S = base64:encode(iolist_to_binary(json:encode(V))),
     <<$", "base64:", S/binary, $">>;
 field_val(true) ->
     <<"true">>;
@@ -56,4 +56,3 @@ as_list(L) when is_list(L) ->
     L;
 as_list(L) when is_map(L) ->
     maps:to_list(L).
-

apps/iot/src/ctrl/ctrl_acceptor.erl

@@ -0,0 +1,41 @@
+%%%-------------------------------------------------------------------
+%% @doc Control socket acceptor.
+%% @end
+%%%-------------------------------------------------------------------
+
+-module(ctrl_acceptor).
+
+-export([start_link/1, init/1]).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec start_link(gen_tcp:socket()) -> {ok, pid()} | {error, term()}.
+start_link(ListenSocket) ->
+    proc_lib:start_link(?MODULE, init, [ListenSocket]).
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec init(gen_tcp:socket()) -> ok.
+init(ListenSocket) ->
+    ok = iot_log:set_metadata(),
+    proc_lib:init_ack({ok, self()}),
+    accept_loop(ListenSocket).
+
+-spec accept_loop(gen_tcp:socket()) -> ok.
+accept_loop(ListenSocket) ->
+    case gen_tcp:accept(ListenSocket) of
+        {ok, Socket} ->
+            {ok, Pid} = ctrl_channel:start(),
+            ok = gen_tcp:controlling_process(Socket, Pid),
+            ok = ctrl_channel:set_socket(Pid, Socket),
+            accept_loop(ListenSocket);
+        {error, closed} ->
+            ok;
+        {error, Reason} ->
+            logger:error("[ctrl_acceptor] accept failed: ~p", [Reason]),
+            exit({accept_failed, Reason})
+    end.

apps/iot/src/ctrl/ctrl_channel.erl

@@ -0,0 +1,116 @@
+%%%-------------------------------------------------------------------
+%% @doc Single control socket channel.
+%% @end
+%%%-------------------------------------------------------------------
+
+-module(ctrl_channel).
+-behaviour(gen_server).
+
+-export([start/0, set_socket/2]).
+
+-export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
+
+-record(state, {
+    socket = undefined :: gen_tcp:socket() | undefined
+}).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec start() -> {ok, pid()} | ignore | {error, term()}.
+start() ->
+    gen_server:start(?MODULE, [], []).
+
+-spec set_socket(pid(), gen_tcp:socket()) -> ok.
+set_socket(Pid, Socket) ->
+    gen_server:call(Pid, {set_socket, Socket}).
+
+%%%===================================================================
+%%% gen_server callbacks
+%%%===================================================================
+
+-spec init([]) -> {ok, #state{}}.
+init([]) ->
+    ok = iot_log:set_metadata(),
+    {ok, #state{}}.
+
+-spec handle_call(term(), {pid(), term()}, #state{}) ->
+    {reply, term(), #state{}} | {stop, term(), term(), #state{}}.
+handle_call({set_socket, Socket}, _From, State = #state{socket = undefined}) ->
+    case inet:setopts(Socket, [{packet, 2}, {active, once}]) of
+        ok ->
+            {reply, ok, State#state{socket = Socket}};
+        {error, Reason} ->
+            logger:warning("[ctrl_channel] failed to activate socket: ~p", [Reason]),
+            gen_tcp:close(Socket),
+            {stop, Reason, {error, Reason}, State}
+    end;
+handle_call(_Request, _From, State = #state{}) ->
+    {reply, ok, State}.
+
+-spec handle_cast(term(), #state{}) -> {noreply, #state{}}.
+handle_cast(_Request, State = #state{}) ->
+    {noreply, State}.
+
+-spec handle_info(term(), #state{}) -> {noreply, #state{}} | {stop, term(), #state{}}.
+handle_info({tcp, Socket, Data}, State = #state{socket = Socket}) ->
+    ok = send_response(Socket, dispatch(Data)),
+    ok = inet:setopts(Socket, [{active, once}]),
+    {noreply, State};
+handle_info({tcp_closed, Socket}, State = #state{socket = Socket}) ->
+    {stop, normal, State};
+handle_info({tcp_error, Socket, Reason}, State = #state{socket = Socket}) ->
+    logger:warning("[ctrl_channel] socket error: ~p", [Reason]),
+    {stop, Reason, State};
+handle_info(Info, State = #state{}) ->
+    logger:warning("[ctrl_channel] ignore unknown info: ~p", [Info]),
+    {noreply, State}.
+
+-spec terminate(term(), #state{}) -> ok.
+terminate(_Reason, #state{socket = undefined}) ->
+    ok;
+terminate(_Reason, #state{socket = Socket}) ->
+    gen_tcp:close(Socket),
+    ok.
+
+-spec code_change(term(), #state{}, term()) -> {ok, #state{}}.
+code_change(_OldVsn, State = #state{}, _Extra) ->
+    {ok, State}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec send_response(gen_tcp:socket(), {ok, iodata()} | {error, iodata()}) -> ok.
+send_response(Socket, {ok, Response}) ->
+    Reply = json:encode(#{<<"result">> => Response}),
+    gen_tcp:send(Socket, Reply);
+send_response(Socket, {error, Response}) ->
+    Reply = json:encode(#{<<"error">> => #{<<"message">> => Response}}),
+    gen_tcp:send(Socket, Reply).
+
+-spec dispatch(binary()) -> {ok, iodata()} | {error, iodata()}.
+dispatch(Data) when is_binary(Data) ->
+    Request = json:decode(Data),
+    handle_request(Request).
+
+-spec handle_request(binary()) -> {ok, iodata()} | {error, iodata()}.
+handle_request(#{<<"method">> := <<"ping">>}) ->
+    {ok, <<"pong">>};
+
+handle_request(#{<<"method">> := <<"add_client">>, <<"params">> := #{<<"uuid">> := UUID, <<"token">> := Token}}) ->
+    case efka_client_store:register(UUID, Token) of
+        ok ->
+            {ok, <<"OK">>};
+        {error, Reason} ->
+            ReadableReason = readable_binary(Reason),
+            {error, <<"add failed: ", ReadableReason/binary>>}
+    end;
+handle_request(Command) ->
+    logger:warning("[ctrl_channel] unsupported command: ~p", [Command]),
+    {error, <<"error unsupported_command\n">>}.
+
+-spec readable_binary(Term :: any()) -> binary().
+readable_binary(Term) ->
+    iolist_to_binary(io_lib:format("~p", [Term])).

apps/iot/src/ctrl/ctrl_server.erl

@@ -0,0 +1,111 @@
+%%%-------------------------------------------------------------------
+%% @doc Unix domain socket control service.
+%% @end
+%%%-------------------------------------------------------------------
+
+-module(ctrl_server).
+-behaviour(gen_server).
+
+-export([start_link/0]).
+
+-export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
+
+-define(SERVER, ?MODULE).
+-define(DEFAULT_SOCKET_PATH, "/var/lib/iot/ctl.sock").
+-define(DEFAULT_BACKLOG, 128).
+
+-record(state, {
+    listen_socket :: gen_tcp:socket(),
+    socket_path :: file:filename_all(),
+    acceptor :: pid()
+}).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec start_link() -> {ok, pid()} | ignore | {error, term()}.
+start_link() ->
+    gen_server:start_link({local, ?SERVER}, ?MODULE, [], []).
+
+%%%===================================================================
+%%% gen_server callbacks
+%%%===================================================================
+
+-spec init([]) -> {ok, #state{}} | {stop, term()}.
+init([]) ->
+    ok = iot_log:set_metadata(),
+    Props = application:get_env(iot, ctrl_server, []),
+    SocketPath = proplists:get_value(socket_path, Props, ?DEFAULT_SOCKET_PATH),
+    Backlog = proplists:get_value(backlog, Props, ?DEFAULT_BACKLOG),
+
+    ok = ensure_socket_dir(SocketPath),
+    ok = delete_stale_socket(SocketPath),
+
+    ListenOpts = [
+        binary,
+        {packet, line},
+        {active, false},
+        {backlog, Backlog},
+        {ifaddr, {local, SocketPath}}
+    ],
+    case gen_tcp:listen(0, ListenOpts) of
+        {ok, ListenSocket} ->
+            {ok, Acceptor} = ctrl_acceptor:start_link(ListenSocket),
+            logger:debug("[ctrl_server] start at socket: ~ts", [SocketPath]),
+            {ok, #state{listen_socket = ListenSocket, socket_path = SocketPath, acceptor = Acceptor}};
+        {error, Reason} ->
+            logger:error("[ctrl_server] failed to listen on socket ~ts: ~p", [SocketPath, Reason]),
+            {stop, Reason}
+    end.
+
+-spec handle_call(term(), {pid(), term()}, #state{}) -> {reply, ok, #state{}}.
+handle_call(_Request, _From, State = #state{}) ->
+    {reply, ok, State}.
+
+-spec handle_cast(term(), #state{}) -> {noreply, #state{}}.
+handle_cast(_Request, State = #state{}) ->
+    {noreply, State}.
+
+-spec handle_info(term(), #state{}) -> {noreply, #state{}}.
+handle_info(Info, State = #state{}) ->
+    logger:warning("[ctrl_server] ignore unknown info: ~p", [Info]),
+    {noreply, State}.
+
+-spec terminate(term(), #state{}) -> ok.
+terminate(_Reason, #state{listen_socket = ListenSocket, socket_path = SocketPath}) ->
+    gen_tcp:close(ListenSocket),
+    ok = delete_stale_socket(SocketPath),
+    ok.
+
+-spec code_change(term(), #state{}, term()) -> {ok, #state{}}.
+code_change(_OldVsn, State = #state{}, _Extra) ->
+    {ok, State}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec ensure_socket_dir(file:filename_all()) -> ok | {error, term()}.
+ensure_socket_dir(SocketPath) ->
+    case filelib:ensure_dir(SocketPath) of
+        ok ->
+            ok;
+        {error, Reason} ->
+            logger:error("[ctrl_server] failed to ensure socket dir for ~ts: ~p",
+                [SocketPath, Reason]),
+            {error, Reason}
+    end.
+
+-spec delete_stale_socket(file:filename_all()) -> ok.
+delete_stale_socket(SocketPath) ->
+    case file:delete(SocketPath) of
+        ok ->
+            ok;
+        {error, enoent} ->
+            ok;
+        {error, Reason} ->
+            logger:warning("[ctrl_server] failed to delete stale socket ~ts: ~p",
+                [SocketPath, Reason]),
+            ok
+    end.

apps/iot/src/database/ai_event_logs_bo.erl

@@ -1,30 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 16. 5月 2023 12:48
-%%%-------------------------------------------------------------------
--module(ai_event_logs_bo).
--author("aresei").
--include("iot.hrl").
-
--export([insert/6]).
-
-%% API
-
--spec insert(HostUUID :: binary(), DeviceUUID :: binary(), SceneId :: integer(), MicroId :: integer(), EventType :: integer(), Content :: binary()) ->
-    ok | {ok, InsertId :: integer()} | {error, Reason :: any()}.
-insert(HostUUID, DeviceUUID, SceneId, MicroId, EventType, Content)
-    when is_integer(EventType), is_binary(HostUUID), is_binary(DeviceUUID), is_integer(SceneId), is_integer(MicroId), is_binary(Content) ->
-
-    mysql_pool:insert(mysql_iot, <<"ai_event_logs">>, #{
-        <<"event_type">> => EventType,
-        <<"host_uuid">> => HostUUID,
-        <<"device_uuid">> => DeviceUUID,
-        <<"scene_id">> => SceneId,
-        <<"micro_id">> => MicroId,
-        <<"content">> => Content,
-        <<"created_at">> => calendar:local_time()
-    }, true).

apps/iot/src/database/device_bo.erl

@@ -1,43 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 16. 5月 2023 12:48
-%%%-------------------------------------------------------------------
--module(device_bo).
--author("aresei").
--include("iot.hrl").
-
-%% API
--export([get_all_devices/0, get_host_devices/1, get_device_by_uuid/1, change_status/2]).
-
--spec get_all_devices() -> {ok, Devices :: [map()]} | {error, Reason :: any()}.
-get_all_devices() ->
-    mysql_pool:get_all(mysql_iot, <<"SELECT * FROM device WHERE device_uuid != ''">>).
-
--spec get_host_devices(HostId :: integer()) -> {ok, Devices :: [map()]} | {error, Reason::any()}.
-get_host_devices(HostId) when is_integer(HostId) ->
-    mysql_pool:get_all(mysql_iot, <<"SELECT device_uuid FROM device WHERE host_id = ? AND device_uuid != ''">>, [HostId]).
-
--spec get_device_by_uuid(DeviceUUID :: binary()) -> {ok, DeviceInfo :: map()} | undefined.
-get_device_by_uuid(DeviceUUID) when is_binary(DeviceUUID) ->
-    mysql_pool:get_row(mysql_iot, <<"SELECT * FROM device WHERE device_uuid = ? LIMIT 1">>, [DeviceUUID]).
-
-%% 修改主机的状态
--spec change_status(DeviceUUID :: binary(), Status :: integer()) -> {ok, AffectedRows :: integer()} | {error, Reason :: any()}.
-change_status(DeviceUUID, NStatus) when is_binary(DeviceUUID), is_integer(NStatus) ->
-    change_status0(DeviceUUID, NStatus).
-change_status0(DeviceUUID, ?DEVICE_ONLINE) when is_binary(DeviceUUID) ->
-    Timestamp = calendar:local_time(),
-    case mysql_pool:get_row(mysql_iot, <<"SELECT status FROM device WHERE device_uuid = ? LIMIT 1">>, [DeviceUUID]) of
-        {ok, #{<<"status">> := -1}} ->
-            mysql_pool:update_by(mysql_iot, <<"UPDATE device SET status = ?, access_at = ?, updated_at = ? WHERE device_uuid = ? LIMIT 1">>, [?DEVICE_ONLINE, Timestamp, Timestamp, DeviceUUID]);
-        {ok, _} ->
-            mysql_pool:update_by(mysql_iot, <<"UPDATE device SET status = ?, updated_at = ? WHERE device_uuid = ? LIMIT 1">>, [?DEVICE_ONLINE, Timestamp, DeviceUUID]);
-        undefined ->
-            {error, <<"device not found">>}
-    end;
-change_status0(DeviceUUID, ?DEVICE_OFFLINE) when is_binary(DeviceUUID) ->
-    mysql_pool:update_by(mysql_iot, <<"UPDATE device SET status = ? WHERE device_uuid = ? LIMIT 1">>, [?DEVICE_OFFLINE, DeviceUUID]).

apps/iot/src/database/endpoint_bo.erl

@@ -1,97 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 16. 5月 2023 12:48
-%%%-------------------------------------------------------------------
--module(endpoint_bo).
--author("aresei").
--include("endpoint.hrl").
-
-%% API
--export([get_all_endpoints/0, get_endpoint/1]).
--export([endpoint_record/1]).
-
--spec get_all_endpoints() -> [Endpoint :: map()].
-get_all_endpoints() ->
-    case mysql_pool:get_all(mysql_iot, <<"SELECT * FROM endpoint where status = 1">>) of
-        {ok, Endpoints} ->
-            Endpoints;
-        {error, _} ->
-            []
-    end.
-
--spec get_endpoint(Id :: integer()) -> undefined | {ok, EndpointInfo :: map()}.
-get_endpoint(Id) when is_integer(Id) ->
-    mysql_pool:get_row(mysql_iot, <<"SELECT * FROM endpoint WHERE id = ? and status = 1 LIMIT 1">>, [Id]).
-
--spec endpoint_record(EndpointInfo :: #{}) -> error | {ok, Endpoint :: #endpoint{}}.
-endpoint_record(#{<<"id">> := Id, <<"name">> := Name, <<"title">> := Title, <<"type">> := Type, <<"config_json">> := ConfigJson,
-    <<"status">> := Status, <<"updated_at">> := UpdatedAt, <<"created_at">> := CreatedAt}) ->
-    try
-        Config = parse_config(Type, catch jiffy:decode(ConfigJson, [return_maps])),
-        {ok, #endpoint {
-            id = Id,
-            name = Name,
-            title = Title,
-            config = Config,
-            status = Status,
-            updated_at = UpdatedAt,
-            created_at = CreatedAt
-        }}
-    catch throw:_ ->
-        error
-    end.
-
-parse_config(<<"mqtt">>, #{<<"host">> := Host, <<"port">> := Port, <<"client_id">> := ClientId, <<"username">> := Username, <<"password">> := Password, <<"topic">> := Topic, <<"qos">> := Qos}) ->
-    #mqtt_endpoint{
-        host = Host,
-        port = Port,
-        client_id = ClientId,
-        username = Username,
-        password = Password,
-        topic = Topic,
-        qos = Qos
-    };
-parse_config(<<"http">>, #{<<"url">> := Url, <<"pool_size">> := PoolSize}) ->
-    #http_endpoint{
-        url = Url,
-        pool_size = PoolSize
-    };
-parse_config(<<"kafka">>, #{<<"sasl_config">> := #{<<"username">> := Username, <<"password">> := Password, <<"mechanism">> := Mechanism0}, <<"bootstrap_servers">> := BootstrapServers, <<"topic">> := Topic}) ->
-    Mechanism = case Mechanism0 of
-                    <<"sha_256">> ->
-                        scram_sha_256;
-                    <<"sha_512">> ->
-                        scram_sha_512;
-                    <<"plain">> ->
-                        plain;
-                    _ ->
-                        plain
-                end,
-
-    #kafka_endpoint{
-        sasl_config = {Mechanism, Username, Password},
-        bootstrap_servers = parse_bootstrap_servers(BootstrapServers),
-        topic = Topic
-    };
-parse_config(<<"kafka">>, #{<<"bootstrap_servers">> := BootstrapServers, <<"topic">> := Topic}) ->
-    #kafka_endpoint{
-        sasl_config = undefined,
-        bootstrap_servers = parse_bootstrap_servers(BootstrapServers),
-        topic = Topic
-    };
-parse_config(_, _) ->
-    throw(invalid_config).
-
-parse_bootstrap_servers(BootstrapServers) when is_list(BootstrapServers) ->
-    lists:filtermap(fun(S) ->
-        case binary:split(S, <<":">>) of
-            [Host0, Port0] ->
-                {true, {binary_to_list(Host0), binary_to_integer(Port0)}};
-            _ ->
-                false
-        end
-                    end, BootstrapServers).

apps/iot/src/database/event_logs_bo.erl

@@ -1,25 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 16. 5月 2023 12:48
-%%%-------------------------------------------------------------------
--module(event_logs_bo).
--author("aresei").
--include("iot.hrl").
-
--export([insert/3]).
-
-%% API
-
--spec insert(EventType :: integer(), AssocUUID :: binary(), Status :: integer()) ->
-    {ok, InsertId :: integer()} | {error, Reason :: any()}.
-insert(EventType, AssocUUID, Status) when is_integer(EventType), is_binary(AssocUUID), is_integer(Status) ->
-    mysql_pool:insert(mysql_iot, <<"event_logs">>, #{
-        <<"event_type">> => EventType,
-        <<"assoc_uuid">> => AssocUUID,
-        <<"status">> => Status,
-        <<"created_at">> => calendar:local_time()
-    }, true).

apps/iot/src/database/host_bo.erl

@@ -1,49 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 16. 5月 2023 12:48
-%%%-------------------------------------------------------------------
--module(host_bo).
--author("aresei").
--include("iot.hrl").
-
-%% API
--export([get_all_hosts/0, change_status/2, get_host_by_uuid/1, get_host_by_id/1]).
-
--spec get_all_hosts() -> UUIDList :: [binary()].
-get_all_hosts() ->
-    case mysql_pool:get_all(mysql_iot, <<"SELECT uuid FROM host where uuid != '' limit 10">>) of
-        {ok, Hosts} ->
-            lists:map(fun(#{<<"uuid">> := UUID}) -> UUID end, Hosts);
-        {error, _} ->
-            []
-    end.
-
--spec get_host_by_uuid(UUID :: binary()) -> undefined | {ok, HostInfo :: map()}.
-get_host_by_uuid(UUID) when is_binary(UUID) ->
-    mysql_pool:get_row(mysql_iot, <<"SELECT * FROM host WHERE uuid = ? LIMIT 1">>, [UUID]).
-
--spec get_host_by_id(HostId :: integer()) -> undefined | {ok, HostInfo :: map()}.
-get_host_by_id(HostId) when is_integer(HostId) ->
-    mysql_pool:get_row(mysql_iot, <<"SELECT * FROM host WHERE id = ? LIMIT 1">>, [HostId]).
-
-%% 修改主机的状态
--spec change_status(UUID :: binary(), Status :: integer()) -> {ok, AffectedRows :: integer()} | {error, Reason :: any()}.
-change_status(UUID, NStatus) when is_binary(UUID), is_integer(NStatus) ->
-    change_status0(UUID, NStatus).
-change_status0(UUID, ?HOST_ONLINE) when is_binary(UUID) ->
-    Timestamp = calendar:local_time(),
-    case mysql_pool:get_row(mysql_iot, <<"SELECT status FROM host WHERE uuid = ? LIMIT 1">>, [UUID]) of
-        %% 第一次更新激活
-        {ok, #{<<"status">> := -1}} ->
-            mysql_pool:update_by(mysql_iot, <<"UPDATE host SET status = ?, access_at = ?, updated_at = ? WHERE uuid = ? LIMIT 1">>, [?HOST_ONLINE, Timestamp, Timestamp, UUID]);
-        {ok, _} ->
-            mysql_pool:update_by(mysql_iot, <<"UPDATE host SET status = ?, updated_at = ? WHERE uuid = ? LIMIT 1">>, [?HOST_ONLINE, Timestamp, UUID]);
-        undefined ->
-            {error, <<"host not found">>}
-    end;
-change_status0(UUID, ?HOST_OFFLINE) when is_binary(UUID) ->
-    mysql_pool:update_by(mysql_iot, <<"UPDATE host SET status = ? WHERE uuid = ? LIMIT 1">>, [?HOST_OFFLINE, UUID]).

apps/iot/src/database/micro_inform_log.erl

@@ -1,17 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 16. 5月 2023 12:48
-%%%-------------------------------------------------------------------
--module(micro_inform_log).
--author("aresei").
--include("iot.hrl").
-
-%% API
--export([insert/1]).
-
-insert(Fields) when is_map(Fields) ->
-    mysql_pool:insert(mysql_iot, <<"micro_inform_log">>, Fields, true).

apps/iot/src/database/micro_service_bo.erl

@@ -1,23 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 16. 5月 2023 12:48
-%%%-------------------------------------------------------------------
--module(micro_service_bo).
--author("aresei").
--export([get_service_config/1]).
-
-%% API
-
-%% TODO
--spec get_service_config(ServiceId :: binary()) -> {ok, ConfigJson :: binary()} | error.
-get_service_config(ServiceId) when is_binary(ServiceId) ->
-    case mysql_pool:get_row(mysql_iot, <<"SELECT * FROM micro_service WHERE id = ? LIMIT 1">>, [ServiceId]) of
-        undefined ->
-            error;
-        {ok, #{<<"config">> := Config}} ->
-            {ok, Config}
-    end.

apps/iot/src/database/micro_set_bo.erl

@@ -1,19 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 16. 5月 2023 12:48
-%%%-------------------------------------------------------------------
--module(micro_set_bo).
--author("aresei").
--include("iot.hrl").
-
-%% API
--export([change_status/4]).
-
-%% 修改主机的状态
--spec change_status(HostId :: integer(), SceneId :: integer(), MircoId :: integer(), Status :: integer()) -> {ok, AffectedRows :: integer()} | {error, Reason :: any()}.
-change_status(HostId, SceneId, MircoId, Status) when is_integer(HostId), is_integer(SceneId), is_integer(MircoId), is_integer(Status) ->
-    mysql_pool:update_by(mysql_iot, <<"UPDATE micro_set SET status = ? WHERE host_id = ? AND scene_id = ? AND micro_id = ? LIMIT 1">>, [Status, HostId, SceneId, MircoId]).

apps/iot/src/database/task_logs_bo.erl

@@ -1,19 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 16. 5月 2023 12:48
-%%%-------------------------------------------------------------------
--module(task_logs_bo).
--author("aresei").
--include("iot.hrl").
-
-%% API
--export([change_status/2]).
-
-%% 修改主机的状态
--spec change_status(TaskId :: integer(), Status :: integer()) -> {ok, AffectedRow :: integer()} | {error, Reason :: any()}.
-change_status(TaskId, Status) when is_integer(TaskId), is_integer(Status) ->
-    mysql_pool:update_by(mysql_iot, <<"UPDATE task_logs SET status = ? WHERE id = ? LIMIT 1">>, [Status, TaskId]).

apps/iot/src/devtools/endpoint_kafka_test.erl

@@ -0,0 +1,60 @@
+%%%-------------------------------------------------------------------
+%%% @author anlicheng
+%%% @copyright (C) 2025, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 14. 11月 2025 23:29
+%%%-------------------------------------------------------------------
+-module(endpoint_kafka_test).
+-author("anlicheng").
+-include_lib("endpoint/include/endpoint.hrl").
+
+%% API
+-export([start_manual/0, test_consumer/0]).
+
+start_manual() ->
+    Name = endpoint:get_name(100),
+    {ok, Pid} = endpoint_kafka:start_link(Name, #endpoint{
+        id = 100,
+        %% 全局唯一，在路由规则中通过名称来指定
+        matcher = <<"/dhlr/device/*/*">>,
+        %% 标题描述
+        title = <<"test_kafka_title">>,
+        %% 配置项, 格式: #{<<"protocol">> => <<"http|https|ws|kafka|mqtt">>, <<"args">> => #{}}
+        config = #kafka_endpoint{
+            sasl_config = {plain, <<"test">>, <<"test123">>},
+            bootstrap_servers = [{"118.178.229.213", 9092}],
+            topic = <<"dhlr_data">>
+        }
+    }),
+
+    Json = iolist_to_binary(json:encode(#{
+        <<"name">> => <<"anlicheng">>,
+        <<"age">> => 30
+    })),
+
+    endpoint:forward(Pid, Json),
+    ok.
+
+test_consumer() ->
+    KafkaBootstrapEndpoints = [{"118.178.229.213", 9092}],
+    Topic = <<"dhlr_data">>,
+
+    ClientConfig = [
+        {sasl, {plain, <<"test">>, <<"test123">>}},
+        {reconnect_cool_down_seconds, 5},
+        {socket_options, [{keepalive, true}]}
+    ],
+
+    ok = brod:start_client(KafkaBootstrapEndpoints, client1, ClientConfig),
+    SubscriberCallbackFun = fun(_Partition, Msg, ShellPid = CallbackState) ->
+        logger:debug("call here msg: ~p", [Msg]),
+        ShellPid ! Msg, {ok, ack, CallbackState}
+                            end,
+    Res = brod_topic_subscriber:start_link(client1, Topic, all,
+        _ConsumerConfig=[{begin_offset, 0}],
+        _CommittedOffsets=[], message, SubscriberCallbackFun,
+        _CallbackState=self()),
+
+    logger:debug("start subscriber res: ~p", [Res]).

apps/iot/src/devtools/endpoint_mqtt_subscriber.erl

@@ -4,56 +4,42 @@
 %%% @doc
 %%%
 %%% @end
-%%% Created : 17. 8月 2025 00:26
+%%% Created : 17. 11月 2025 16:48
 %%%-------------------------------------------------------------------
--module(iot_name_server).
+-module(endpoint_mqtt_subscriber).
 -author("anlicheng").
 
+-author("aresei").
+-include("iot.hrl").
+
 -behaviour(gen_server).
 
 %% API
 -export([start_link/0]).
--export([whereis_alias/1, register/2]).
 
 %% gen_server callbacks
 -export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
 
 -define(SERVER, ?MODULE).
--define(TAB, iot_name_server).
+
+%% 需要订阅的主题信息
+-define(Topics,[
+    {<<"/dhlr/data">>, 0}
+]).
 
 -record(state, {
-    %% #{Pid => Name}
-    pid_names = #{},
-    refs = []
+    conn_pid :: pid()
 }).
 
 %%%===================================================================
 %%% API
 %%%===================================================================
 
--spec register(Name :: atom(), Pid :: pid()) -> ok.
-register(Name, Pid) when is_atom(Name), is_pid(Pid) ->
-    gen_server:call(?SERVER, {register, Name, Pid}).
-
--spec whereis_alias(Name :: atom()) -> undefined | pid().
-whereis_alias(Name) when is_atom(Name) ->
-    case ets:lookup(?TAB, Name) of
-        [] ->
-            undefined;
-        [{Name, Pid}|_] ->
-            case is_process_alive(Pid) of
-                true ->
-                    Pid;
-                false ->
-                    undefined
-            end
-    end.
-
 %% @doc Spawns the server and registers the local name (unique)
 -spec(start_link() ->
     {ok, Pid :: pid()} | ignore | {error, Reason :: term()}).
 start_link() ->
-    gen_server:start_link({local, ?SERVER}, ?MODULE, [], []).
+    gen_server:start_link({local, ?MODULE}, ?MODULE, [], []).
 
 %%%===================================================================
 %%% gen_server callbacks
@@ -65,9 +51,41 @@ start_link() ->
     {ok, State :: #state{}} | {ok, State :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term()} | ignore).
 init([]) ->
-    %% 初始化存储
-    ets:new(?TAB, [named_table, ordered_set, public, {keypos, 1}]),
-    {ok, #state{}}.
+    %% 建立到emqx服务器的连接
+    ClientId = <<"mqtt-client-test-host-subscriber">>,
+    Opts = [
+        {clientid, ClientId},
+        {host, "118.178.229.213"},
+        {port, 1883},
+        {owner, self()},
+        {tcp_opts, []},
+        {username, "admin"},
+        {password, "admin"},
+        {keepalive, 86400},
+        {auto_ack, true},
+        {proto_ver, v5},
+        {retry_interval, 5}
+    ],
+
+    logger:debug("[opts] is: ~p", [Opts]),
+    case emqtt:start_link(Opts) of
+        {ok, ConnPid} ->
+            %% 监听和host相关的全部事件
+            logger:debug("[iot_mqtt_subscriber] start conntecting, pid: ~p", [ConnPid]),
+            {ok, _} = emqtt:connect(ConnPid),
+            logger:debug("[iot_mqtt_subscriber] connect success, pid: ~p", [ConnPid]),
+            SubscribeResult = emqtt:subscribe(ConnPid, ?Topics),
+
+            logger:debug("[iot_mqtt_subscriber] subscribe topics: ~p, result is: ~p", [?Topics, SubscribeResult]),
+
+            {ok, #state{conn_pid = ConnPid}};
+        ignore ->
+            logger:debug("[iot_mqtt_subscriber] connect emqx get ignore"),
+            {stop, ignore};
+        {error, Reason} ->
+            logger:debug("[iot_mqtt_subscriber] connect emqx get error: ~p", [Reason]),
+            {stop, Reason}
+    end.
 
 %% @private
 %% @doc Handling call messages
@@ -79,10 +97,8 @@ init([]) ->
     {noreply, NewState :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term(), Reply :: term(), NewState :: #state{}} |
     {stop, Reason :: term(), NewState :: #state{}}).
-handle_call({register, Name, Pid}, _From, State = #state{refs = Refs, pid_names = PidNames}) ->
-    true = ets:insert(?TAB, {Name, Pid}),
-    MRef = erlang:monitor(process, Pid),
-    {reply, ok, State#state{refs = [MRef|Refs], pid_names = maps:put(Pid, Name, PidNames)}}.
+handle_call(_Info, _From, State = #state{conn_pid = _ConnPid}) ->
+    {reply, ok, State}.
 
 %% @private
 %% @doc Handling cast messages
@@ -99,20 +115,21 @@ handle_cast(_Request, State = #state{}) ->
     {noreply, NewState :: #state{}} |
     {noreply, NewState :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term(), NewState :: #state{}}).
-handle_info({'DOWN', MRef, process, Pid, Reason}, State = #state{refs = Refs, pid_names = PidNames}) ->
-    % lager:debug("[iot_name_server] pid: ~p, down with reason: ~p", [Reason]),
-    case lists:member(MRef, Refs) of
-        true ->
-            case maps:take(Pid, PidNames) of
-                error ->
-                    {noreply, State#state{refs = lists:delete(MRef, Refs)}};
-                {Name, NPidNames} ->
-                    true = ets:delete(?TAB, Name),
-                    {noreply, State#state{pid_names = NPidNames, refs = lists:delete(MRef, Refs)}}
-            end;
-        false ->
-            {noreply, State}
-    end.
+handle_info({disconnect, ReasonCode, Properties}, State = #state{}) ->
+    logger:debug("[iot_mqtt_subscriber] Recv a DISONNECT packet - ReasonCode: ~p, Properties: ~p", [ReasonCode, Properties]),
+    {stop, disconnected, State};
+%% 必须要做到消息的快速分发，数据的json反序列需要在host进程进行
+handle_info({publish, #{packet_id := _PacketId, payload := Payload, qos := Qos, topic := Topic}}, State = #state{conn_pid = _ConnPid}) ->
+    logger:debug("[iot_mqtt_subscriber] Recv a topic: ~p, publish packet: ~p, qos: ~p", [Topic, Payload, Qos]),
+    %% 将消息分发到对应的host进程去处理
+    {noreply, State};
+handle_info({puback, Packet = #{packet_id := _PacketId}}, State = #state{}) ->
+    logger:debug("[iot_mqtt_subscriber] receive puback packet: ~p", [Packet]),
+    {noreply, State};
+
+handle_info(Info, State = #state{}) ->
+    logger:debug("[iot_mqtt_subscriber] get info: ~p", [Info]),
+    {noreply, State}.
 
 %% @private
 %% @doc This function is called by a gen_server when it is about to
@@ -121,7 +138,16 @@ handle_info({'DOWN', MRef, process, Pid, Reason}, State = #state{refs = Refs, pi
 %% with Reason. The return value is ignored.
 -spec(terminate(Reason :: (normal | shutdown | {shutdown, term()} | term()),
     State :: #state{}) -> term()).
-terminate(_Reason, _State = #state{}) ->
+terminate(Reason, _State = #state{conn_pid = ConnPid}) when is_pid(ConnPid) ->
+    %% 取消topic的订阅
+    TopicNames = lists:map(fun({Name, _}) -> Name end, ?Topics),
+    {ok, _Props, _ReasonCode} = emqtt:unsubscribe(ConnPid, #{}, TopicNames),
+
+    ok = emqtt:disconnect(ConnPid),
+    logger:debug("[iot_mqtt_subscriber] terminate with reason: ~p", [Reason]),
+    ok;
+terminate(Reason, _State) ->
+    logger:debug("[iot_mqtt_subscriber] terminate with reason: ~p", [Reason]),
     ok.
 
 %% @private

apps/iot/src/devtools/eval_test.erl

@@ -13,13 +13,5 @@
 -export([test/0]).
 
 test() ->
-
-    {ok, Content} = file:read_file("/tmp/test.erl"),
-
-    {ok, Tokens, _} = erl_scan:string(binary_to_list(Content)),
-    {ok, ExprList} = erl_parse:parse_exprs(Tokens),
-
-    {value, F, _NewBindings} = erl_eval:exprs(ExprList, []),
-    F(#{name => <<"test">>}).
-
+    {error, disabled}.
 

apps/iot/src/devtools/iot_mock.erl

@@ -12,8 +12,7 @@
 
 %% API
 -export([rsa_encode/1]).
--export([insert_services/1]).
--export([test_mqtt/0, test_influxdb/0]).
+-export([test_influxdb/0]).
 
 test_influxdb() ->
     UUID = <<"device123123">>,
@@ -29,41 +28,20 @@ test_influxdb() ->
                                          end)
                   end, lists:seq(1, 100)).
 
-test_mqtt() ->
-    iot_zd_endpoint:forward(<<"location_code_test123">>, [
-        #{<<"key">> => <<"name">>, <<"value">> => <<"anlicheng">>},
-        #{<<"key">> => <<"age">>, <<"value">> => 30},
-        #{<<"key">> => <<"flow">>, <<"value">> => 30}
-    ], iot_util:timestamp_of_seconds()).
-
-insert_services(Num) ->
-    lists:foreach(fun(Id) ->
-        Res = mysql_pool:insert(mysql_iot, <<"micro_service">>,
-            #{
-                <<"name">> => <<"微服务"/utf8, (integer_to_binary(Id))/binary>>,
-                <<"code">> => <<"1223423423423423"/utf8>>,
-                <<"type">> => 1,
-                <<"version">> => <<"v1.0">>,
-                <<"url">> => <<"https://www.baidu.com">>,
-                <<"detail">> => <<"这是一个关于测试的微服务"/utf8>>
-            }, false),
-        lager:debug("insert service result is: ~p", [Res])
-                  end, lists:seq(1, Num)).
-
 rsa_encode(Data) when is_binary(Data) ->
     %% 读取相关配置
     PublicPemFile = "/tmp/keys/public.pem",
 
     %% 私钥保存解析后的
     {ok, PubBin} = file:read_file(PublicPemFile),
-    lager:debug("pub bin is: ~p", [PubBin]),
+    logger:debug("pub bin is: ~p", [PubBin]),
     [Pub] = public_key:pem_decode(PubBin),
-    lager:debug("pub pem bin is: ~p", [Pub]),
+    logger:debug("pub pem bin is: ~p", [Pub]),
     PubKey = public_key:pem_entry_decode(Pub),
-    lager:debug("the public key is: ~p", [PubKey]),
+    logger:debug("the public key is: ~p", [PubKey]),
 
     EncData = public_key:encrypt_public(Data, PubKey),
-    lager:debug("enc data is: ~p", [EncData]),
+    logger:debug("enc data is: ~p", [EncData]),
 
     rsa_decode(EncData),
 
@@ -75,13 +53,13 @@ rsa_decode(EncData) when is_binary(EncData) ->
 
     %% 私钥保存解析后的
     {ok, PubBin} = file:read_file(PublicPemFile),
-    lager:debug("pub bin is: ~p", [PubBin]),
+    logger:debug("pub bin is: ~p", [PubBin]),
     [Pub] = public_key:pem_decode(PubBin),
-    lager:debug("pub pem bin is: ~p", [Pub]),
+    logger:debug("pub pem bin is: ~p", [Pub]),
     PubKey = public_key:pem_entry_decode(Pub),
-    lager:debug("the public key is: ~p", [PubKey]),
+    logger:debug("the public key is: ~p", [PubKey]),
 
     PlainData = public_key:decrypt_private(EncData, PubKey),
-    lager:debug("plain data is: ~p", [PlainData]),
+    logger:debug("plain data is: ~p", [PlainData]),
 
     ok.

apps/iot/src/docker/docker_container_builder.erl

@@ -0,0 +1,588 @@
+%%%-------------------------------------------------------------------
+%%% @author
+%%% @copyright (C) 2026, <COMPANY>
+%%% @doc
+%%% ContainerRequest wire builder helpers.
+%%% @end
+%%%-------------------------------------------------------------------
+-module(docker_container_builder).
+
+-export([list_request/0, config_request/2, deploy_request/2, start_request/1, stop_request/1, kill_request/1, remove_request/1]).
+
+-spec list_request() -> map().
+list_request() ->
+    #{action => list, all => true}.
+
+-spec config_request(binary(), binary()) -> map().
+config_request(ContainerName, ConfigJson) when is_binary(ContainerName), is_binary(ConfigJson) ->
+    #{
+        action => config,
+        target => container_ref(ContainerName),
+        config => ConfigJson
+    }.
+
+-spec deploy_request(integer(), map()) -> {ok, map()} | {error, binary()}.
+deploy_request(TaskId, Config) when is_integer(TaskId), is_map(Config), TaskId >= 0 ->
+    try
+        ensure_supported_deploy_config(Config),
+        validate_deploy_config(Config),
+        Params = build_container_deploy_params(Config),
+        {ok, #{action => deploy, task_id => TaskId, params => Params}}
+    catch
+        throw:{error, Reason} ->
+            {error, Reason}
+    end;
+deploy_request(TaskId, _Config) when is_integer(TaskId), TaskId < 0 ->
+    {error, <<"task_id must be non-negative">>};
+deploy_request(_TaskId, _Config) ->
+    {error, <<"invalid deploy request">>}.
+
+-spec start_request(binary()) -> map().
+start_request(ContainerName) when is_binary(ContainerName) ->
+    #{action => start, target => container_ref(ContainerName)}.
+
+-spec stop_request(binary()) -> map().
+stop_request(ContainerName) when is_binary(ContainerName) ->
+    #{
+        action => stop,
+        target => container_ref(ContainerName),
+        timeout_seconds => 0
+    }.
+
+-spec kill_request(binary()) -> map().
+kill_request(ContainerName) when is_binary(ContainerName) ->
+    #{
+        action => kill,
+        target => container_ref(ContainerName),
+        signal => <<>>
+    }.
+
+-spec remove_request(binary()) -> map().
+remove_request(ContainerName) when is_binary(ContainerName) ->
+    #{
+        action => remove,
+        target => container_ref(ContainerName),
+        force => false,
+        remove_volumes => false
+    }.
+
+-spec container_ref(binary()) -> map().
+container_ref(ContainerName) when is_binary(ContainerName) ->
+    #{name => ContainerName, id => <<>>}.
+
+-spec ensure_supported_deploy_config(map()) -> ok.
+ensure_supported_deploy_config(Config) when is_map(Config) ->
+    UnsupportedKeys = [Key || Key <- [<<"container_dir">>], maps:is_key(Key, Config)],
+    case UnsupportedKeys of
+        [] ->
+            ok;
+        _ ->
+            Unsupported = iolist_to_binary(lists:join(<<", ">>, UnsupportedKeys)),
+            throw({error, <<"unsupported container config keys: ", Unsupported/binary>>})
+    end.
+
+-spec validate_deploy_config(map()) -> ok.
+validate_deploy_config(Config) when is_map(Config) ->
+    Required = [
+        {<<"image">>, binary},
+        {<<"container_name">>, binary},
+        {<<"command">>, {list, binary}},
+        {<<"restart">>, binary}
+    ],
+    Optional = [
+        {<<"privileged">>, boolean},
+        {<<"entrypoint">>, {list, binary}},
+        {<<"envs">>, {list, binary}},
+        {<<"ports">>, {list, binary}},
+        {<<"expose">>, {list, binary}},
+        {<<"volumes">>, {list, binary}},
+        {<<"networks">>, {list, binary}},
+        {<<"labels">>, {map, {binary, binary}}},
+        {<<"user">>, binary},
+        {<<"working_dir">>, binary},
+        {<<"hostname">>, binary},
+        {<<"network_mode">>, binary},
+        {<<"cap_add">>, {list, binary}},
+        {<<"cap_drop">>, {list, binary}},
+        {<<"devices">>, {list, binary}},
+        {<<"mem_limit">>, binary},
+        {<<"mem_reservation">>, binary},
+        {<<"cpu_shares">>, non_neg_integer},
+        {<<"cpus">>, non_neg_number},
+        {<<"ulimits">>, {map, {binary, binary}}},
+        {<<"sysctls">>, {map, {binary, binary}}},
+        {<<"tmpfs">>, {list, binary}},
+        {<<"extra_hosts">>, {list, binary}},
+        {<<"healthcheck">>, map}
+    ],
+    Errors = check_required(Config, Required) ++ check_optional(Config, Optional) ++ check_healthcheck(Config),
+    case Errors of
+        [] ->
+            ok;
+        _ ->
+            throw({error, iolist_to_binary(lists:join(<<"|||">>, Errors))})
+    end.
+
+-spec check_required(map(), list()) -> [binary()].
+check_required(Config, Fields) ->
+    lists:foldl(
+        fun({Key, Type}, ErrAcc) ->
+            case maps:get(Key, Config, undefined) of
+                undefined ->
+                    [iolist_to_binary(io_lib:format("miss requied parameter: ~p", [Key])) | ErrAcc];
+                Value ->
+                    case check_type(Value, Type) of
+                        true ->
+                            ErrAcc;
+                        false ->
+                            [iolist_to_binary(io_lib:format("required parameter: ~p, type must be: ~ts", [Key, type_name(Type)])) | ErrAcc]
+                    end
+            end
+        end,
+        [], Fields).
+
+-spec check_optional(map(), list()) -> [binary()].
+check_optional(Config, Fields) ->
+    lists:foldl(
+        fun({Key, Type}, ErrAcc) ->
+            case maps:get(Key, Config, undefined) of
+                undefined ->
+                    ErrAcc;
+                Value ->
+                    case check_type(Value, Type) of
+                        true ->
+                            ErrAcc;
+                        false ->
+                            [iolist_to_binary(io_lib:format("optional parameter: ~p, type must be: ~ts", [Key, type_name(Type)])) | ErrAcc]
+                    end
+            end
+        end,
+        [], Fields).
+
+-spec type_name(tuple() | atom()) -> binary().
+type_name(binary) ->
+    <<"string">>;
+type_name(integer) ->
+    <<"integer">>;
+type_name(non_neg_integer) ->
+    <<"non-negative integer">>;
+type_name(number) ->
+    <<"number">>;
+type_name(non_neg_number) ->
+    <<"non-negative number">>;
+type_name(duration) ->
+    <<"duration string or non-negative integer">>;
+type_name(list) ->
+    <<"list">>;
+type_name({list, binary}) ->
+    <<"list of string">>;
+type_name({list, number}) ->
+    <<"list of number">>;
+type_name({list, integer}) ->
+    <<"list of integer">>;
+type_name(map) ->
+    <<"map">>;
+type_name({map, {binary, binary}}) ->
+    <<"map of string:string">>;
+type_name({map, {binary, any}}) ->
+    <<"map of string:any">>;
+type_name(boolean) ->
+    <<"boolean">>.
+
+-spec check_type(any(), any()) -> boolean().
+check_type(Value, binary) ->
+    is_binary(Value);
+check_type(Value, integer) ->
+    is_integer(Value);
+check_type(Value, non_neg_integer) ->
+    is_integer(Value) andalso Value >= 0;
+check_type(Value, number) ->
+    is_number(Value);
+check_type(Value, non_neg_number) ->
+    is_number(Value) andalso Value >= 0;
+check_type(Value, duration) ->
+    is_binary(Value) orelse (is_integer(Value) andalso Value >= 0);
+check_type(Value, list) when is_list(Value) ->
+    true;
+check_type(Value, {list, binary}) when is_list(Value) ->
+    lists:all(fun(E) -> is_binary(E) end, Value);
+check_type(Value, {list, number}) when is_list(Value) ->
+    lists:all(fun(E) -> is_number(E) end, Value);
+check_type(Value, {list, integer}) when is_list(Value) ->
+    lists:all(fun(E) -> is_integer(E) end, Value);
+check_type(Value, map) when is_map(Value) ->
+    true;
+check_type(Value, {map, {binary, binary}}) when is_map(Value) ->
+    lists:all(fun({K, V}) -> is_binary(K) andalso is_binary(V) end, maps:to_list(Value));
+check_type(Value, {map, {binary, any}}) when is_map(Value) ->
+    lists:all(fun({K, _}) -> is_binary(K) end, maps:to_list(Value));
+check_type(Value, boolean) ->
+    is_boolean(Value);
+check_type(_, _) ->
+    false.
+
+-spec check_healthcheck(map()) -> [binary()].
+check_healthcheck(Config) ->
+    case maps:get(<<"healthcheck">>, Config, undefined) of
+        undefined ->
+            [];
+        Healthcheck when is_map(Healthcheck) ->
+            Fields = [
+                {<<"test">>, {list, binary}},
+                {<<"interval">>, duration},
+                {<<"timeout">>, duration},
+                {<<"retries">>, non_neg_integer}
+            ],
+            KeyErrors = [<<"optional parameter: <<\"healthcheck\">>, type must be: map of string:any">> ||
+                {Key, _Value} <- maps:to_list(Healthcheck), not is_binary(Key)],
+            KeyErrors ++ check_optional(Healthcheck, Fields);
+        _ ->
+            []
+    end.
+
+-spec build_container_deploy_params(map()) -> map().
+build_container_deploy_params(Config) when is_map(Config) ->
+    #{
+        container_name => maps:get(<<"container_name">>, Config),
+        create => build_docker_create_options(Config)
+    }.
+
+-spec build_docker_create_options(map()) -> map().
+build_docker_create_options(Config) when is_map(Config) ->
+    #{
+        config => build_docker_container_config(Config),
+        host_config => build_docker_host_config(Config),
+        networking_config => build_docker_networking_config(Config)
+    }.
+
+-spec build_docker_container_config(map()) -> map().
+build_docker_container_config(Config) when is_map(Config) ->
+    #{
+        image => maps:get(<<"image">>, Config),
+        cmd => maps:get(<<"command">>, Config),
+        entrypoint => maps:get(<<"entrypoint">>, Config, []),
+        env => maps:get(<<"envs">>, Config, []),
+        labels => maps:get(<<"labels">>, Config, #{}),
+        volumes => build_container_volumes(maps:get(<<"volumes">>, Config, [])),
+        user => maps:get(<<"user">>, Config, <<>>),
+        working_dir => maps:get(<<"working_dir">>, Config, <<>>),
+        hostname => maps:get(<<"hostname">>, Config, <<>>),
+        exposed_ports => build_exposed_ports(Config),
+        healthcheck => build_healthcheck(maps:get(<<"healthcheck">>, Config, undefined))
+    }.
+
+-spec build_docker_host_config(map()) -> map().
+build_docker_host_config(Config) when is_map(Config) ->
+    #{
+        binds => build_host_binds(maps:get(<<"volumes">>, Config, [])),
+        network_mode => maps:get(<<"network_mode">>, Config, <<>>),
+        restart_policy => build_restart_policy(maps:get(<<"restart">>, Config)),
+        privileged => maps:get(<<"privileged">>, Config, false),
+        cap_add => maps:get(<<"cap_add">>, Config, []),
+        cap_drop => maps:get(<<"cap_drop">>, Config, []),
+        devices => build_device_mappings(maps:get(<<"devices">>, Config, [])),
+        memory => default_uint64(parse_optional_size_bytes(maps:get(<<"mem_limit">>, Config, undefined), <<"mem_limit">>)),
+        memory_reservation => default_uint64(parse_optional_size_bytes(maps:get(<<"mem_reservation">>, Config, undefined), <<"mem_reservation">>)),
+        nano_cpus => default_uint64(parse_optional_nano_cpus(maps:get(<<"cpus">>, Config, undefined))),
+        cpu_shares => default_uint64(maps:get(<<"cpu_shares">>, Config, undefined)),
+        port_bindings => build_port_bindings(maps:get(<<"ports">>, Config, [])),
+        ulimits => build_ulimits(maps:get(<<"ulimits">>, Config, #{})),
+        tmpfs => maps:from_list(build_tmpfs_options(maps:get(<<"tmpfs">>, Config, []))),
+        sysctls => maps:get(<<"sysctls">>, Config, #{}),
+        extra_hosts => maps:get(<<"extra_hosts">>, Config, [])
+    }.
+
+-spec build_docker_networking_config(map()) -> map().
+build_docker_networking_config(Config) when is_map(Config) ->
+    Networks = maps:get(<<"networks">>, Config, []),
+    #{endpoints => [#{name => Network} || Network <- Networks]}.
+
+-spec build_restart_policy(binary()) -> map().
+build_restart_policy(Restart0) when is_binary(Restart0) ->
+    case binary:split(Restart0, <<":">>) of
+        [Name, RetryCountBin] ->
+            #{name => Name, maximum_retry_count => parse_uint32(RetryCountBin, <<"restart">>)};
+        [Name] ->
+            #{name => Name, maximum_retry_count => 0}
+    end.
+
+-spec build_healthcheck(undefined | map()) -> undefined | map().
+build_healthcheck(undefined) ->
+    undefined;
+build_healthcheck(Healthcheck) when is_map(Healthcheck) ->
+    #{
+        test => maps:get(<<"test">>, Healthcheck, []),
+        interval_ns => parse_duration_ns(maps:get(<<"interval">>, Healthcheck, <<"0s">>), <<"healthcheck.interval">>),
+        timeout_ns => parse_duration_ns(maps:get(<<"timeout">>, Healthcheck, <<"0s">>), <<"healthcheck.timeout">>),
+        retries => maps:get(<<"retries">>, Healthcheck, 0)
+    }.
+
+-spec default_uint64(undefined | non_neg_integer()) -> non_neg_integer().
+default_uint64(undefined) ->
+    0;
+default_uint64(Value) when is_integer(Value), Value >= 0 ->
+    Value.
+
+-spec parse_optional_nano_cpus(undefined | number()) -> undefined | non_neg_integer().
+parse_optional_nano_cpus(undefined) ->
+    undefined;
+parse_optional_nano_cpus(Cpus) when is_integer(Cpus), Cpus >= 0 ->
+    Cpus * 1000000000;
+parse_optional_nano_cpus(Cpus) when is_float(Cpus), Cpus >= 0 ->
+    trunc(Cpus * 1000000000).
+
+-spec build_container_volumes([binary()]) -> [binary()].
+build_container_volumes(VolumeSpecs) when is_list(VolumeSpecs) ->
+    [ContainerPath || VolumeSpec <- VolumeSpecs, {_HostPath, ContainerPath, _ReadOnly} <- [parse_volume_spec(VolumeSpec)]].
+
+-spec build_host_binds([binary()]) -> [binary()].
+build_host_binds(VolumeSpecs) when is_list(VolumeSpecs) ->
+    [volume_bind(HostPath, ContainerPath, ReadOnly) ||
+        VolumeSpec <- VolumeSpecs,
+        {HostPath, ContainerPath, ReadOnly} <- [parse_volume_spec(VolumeSpec)]].
+
+-spec parse_volume_spec(binary()) -> {binary(), binary(), boolean()}.
+parse_volume_spec(VolumeSpec) when is_binary(VolumeSpec) ->
+    case binary:split(VolumeSpec, <<":">>, [global]) of
+        [HostPath, ContainerPath] when HostPath =/= <<>>, ContainerPath =/= <<>> ->
+            {HostPath, ContainerPath, false};
+        [HostPath, ContainerPath | Modes] when HostPath =/= <<>>, ContainerPath =/= <<>> ->
+            {HostPath, ContainerPath, lists:member(<<"ro">>, Modes)};
+        _ ->
+            throw({error, <<"invalid volume binding">>})
+    end.
+
+-spec volume_bind(binary(), binary(), boolean()) -> binary().
+volume_bind(HostPath, ContainerPath, true) when is_binary(HostPath), is_binary(ContainerPath) ->
+    <<HostPath/binary, ":", ContainerPath/binary, ":ro">>;
+volume_bind(HostPath, ContainerPath, false) when is_binary(HostPath), is_binary(ContainerPath) ->
+    <<HostPath/binary, ":", ContainerPath/binary>>.
+
+-spec build_exposed_ports(map()) -> [map()].
+build_exposed_ports(Config) when is_map(Config) ->
+    ExposePorts = [build_exposed_port(ExposeSpec) || ExposeSpec <- maps:get(<<"expose">>, Config, [])],
+    BoundPorts = [
+        #{container_port => ContainerPort, protocol => Protocol} ||
+        #{container_port := ContainerPort, protocol := Protocol} <- build_port_bindings(maps:get(<<"ports">>, Config, []))
+    ],
+    unique_ports(ExposePorts ++ BoundPorts).
+
+-spec build_exposed_port(binary()) -> map().
+build_exposed_port(ExposeSpec) when is_binary(ExposeSpec) ->
+    case binary:split(ExposeSpec, <<"/">>) of
+        [PortBin] ->
+            #{container_port => parse_uint32(PortBin, <<"expose">>), protocol => <<"tcp">>};
+        [PortBin, Protocol] ->
+            #{container_port => parse_uint32(PortBin, <<"expose">>), protocol => Protocol}
+    end.
+
+-spec build_port_bindings([binary()]) -> [map()].
+build_port_bindings(PortSpecs) when is_list(PortSpecs) ->
+    [build_port_binding(PortSpec) || PortSpec <- PortSpecs].
+
+-spec build_port_binding(binary()) -> map().
+build_port_binding(PortSpec) when is_binary(PortSpec) ->
+    case binary:split(PortSpec, <<":">>, [global]) of
+        [HostPortBin, ContainerPortSpec] when HostPortBin =/= <<>>, ContainerPortSpec =/= <<>> ->
+            {ContainerPort, Protocol} = parse_container_port_spec(ContainerPortSpec, <<"ports">>),
+            #{
+                host_ip => <<>>,
+                host_port => parse_tcp_port(HostPortBin, <<"ports.host_port">>),
+                container_port => ContainerPort,
+                protocol => Protocol
+            };
+        _ ->
+            throw({error, <<"invalid port binding">>})
+    end.
+
+-spec parse_container_port_spec(binary(), binary()) -> {non_neg_integer(), binary()}.
+parse_container_port_spec(PortSpec, Field) when is_binary(PortSpec), is_binary(Field) ->
+    case binary:split(PortSpec, <<"/">>) of
+        [PortBin] ->
+            {parse_tcp_port(PortBin, Field), <<"tcp">>};
+        [PortBin, Protocol] when Protocol =/= <<>> ->
+            {parse_tcp_port(PortBin, Field), Protocol};
+        _ ->
+            throw({error, <<"invalid port binding">>})
+    end.
+
+-spec unique_ports([map()]) -> [map()].
+unique_ports(Ports) when is_list(Ports) ->
+    {_Seen, Result} = lists:foldl(
+        fun(Port = #{container_port := ContainerPort, protocol := Protocol}, {Seen, Acc}) ->
+            Key = {ContainerPort, Protocol},
+            case maps:is_key(Key, Seen) of
+                true ->
+                    {Seen, Acc};
+                false ->
+                    {Seen#{Key => true}, [Port | Acc]}
+            end
+        end,
+        {#{}, []},
+        Ports),
+    lists:reverse(Result).
+
+-spec build_device_mappings([binary()]) -> [map()].
+build_device_mappings(DeviceSpecs) when is_list(DeviceSpecs) ->
+    [build_device_mapping(DeviceSpec) || DeviceSpec <- DeviceSpecs].
+
+-spec build_device_mapping(binary()) -> map().
+build_device_mapping(DeviceSpec) when is_binary(DeviceSpec) ->
+    case binary:split(DeviceSpec, <<":">>, [global]) of
+        [HostPath, ContainerPath] when HostPath =/= <<>>, ContainerPath =/= <<>> ->
+            #{path_on_host => HostPath, path_in_container => ContainerPath, cgroup_permissions => <<"rwm">>};
+        [HostPath, ContainerPath, Permissions] when HostPath =/= <<>>, ContainerPath =/= <<>>, Permissions =/= <<>> ->
+            #{path_on_host => HostPath, path_in_container => ContainerPath, cgroup_permissions => Permissions};
+        _ ->
+            throw({error, <<"invalid device mapping">>})
+    end.
+
+-spec build_ulimits(map()) -> [map()].
+build_ulimits(Ulimits) when is_map(Ulimits) ->
+    [build_ulimit(Name, Value) || {Name, Value} <- maps:to_list(Ulimits)].
+
+-spec build_ulimit(binary(), binary()) -> map().
+build_ulimit(Name, Value) when is_binary(Name), is_binary(Value) ->
+    case binary:split(Value, <<":">>) of
+        [SoftBin, HardBin] ->
+            #{name => Name, soft => parse_uint64(SoftBin, <<"ulimits.soft">>), hard => parse_uint64(HardBin, <<"ulimits.hard">>)};
+        [LimitBin] ->
+            Limit = parse_uint64(LimitBin, <<"ulimits.limit">>),
+            #{name => Name, soft => Limit, hard => Limit}
+    end.
+
+-spec build_tmpfs_options([binary()]) -> [{binary(), binary()}].
+build_tmpfs_options(TmpfsSpecs) when is_list(TmpfsSpecs) ->
+    [build_tmpfs_option(TmpfsSpec) || TmpfsSpec <- TmpfsSpecs].
+
+-spec build_tmpfs_option(binary()) -> {binary(), binary()}.
+build_tmpfs_option(TmpfsSpec) when is_binary(TmpfsSpec) ->
+    case binary:split(TmpfsSpec, <<":">>) of
+        [Path] when Path =/= <<>> ->
+            {Path, <<>>};
+        [Path, Options] when Path =/= <<>> ->
+            {Path, Options};
+        _ ->
+            throw({error, <<"invalid tmpfs mount">>})
+    end.
+
+-spec parse_optional_size_bytes(undefined | binary(), binary()) -> undefined | non_neg_integer().
+parse_optional_size_bytes(undefined, _Field) ->
+    undefined;
+parse_optional_size_bytes(Value, Field) when is_binary(Value) ->
+    parse_size_bytes(Value, Field).
+
+-spec parse_duration_ns(binary() | integer(), binary()) -> non_neg_integer().
+parse_duration_ns(Value, _Field) when is_integer(Value), Value >= 0 ->
+    Value;
+parse_duration_ns(Value, Field) when is_binary(Value) ->
+    parse_scaled_uint64(Value, Field, #{
+        <<"ns">> => 1,
+        <<"us">> => 1000,
+        <<"ms">> => 1000000,
+        <<"s">> => 1000000000,
+        <<"m">> => 60000000000,
+        <<"h">> => 3600000000000,
+        <<>> => 1000000000
+    }).
+
+-spec parse_size_bytes(binary(), binary()) -> non_neg_integer().
+parse_size_bytes(Value, Field) when is_binary(Value) ->
+    parse_scaled_uint64(Value, Field, #{
+        <<"b">> => 1,
+        <<"k">> => 1024,
+        <<"kb">> => 1024,
+        <<"ki">> => 1024,
+        <<"kib">> => 1024,
+        <<"m">> => 1048576,
+        <<"mb">> => 1048576,
+        <<"mi">> => 1048576,
+        <<"mib">> => 1048576,
+        <<"g">> => 1073741824,
+        <<"gb">> => 1073741824,
+        <<"gi">> => 1073741824,
+        <<"gib">> => 1073741824,
+        <<"t">> => 1099511627776,
+        <<"tb">> => 1099511627776,
+        <<"ti">> => 1099511627776,
+        <<"tib">> => 1099511627776,
+        <<>> => 1
+    }).
+
+-spec parse_scaled_uint64(binary(), binary(), map()) -> non_neg_integer().
+parse_scaled_uint64(Value0, Field, Multipliers) when is_binary(Value0), is_binary(Field), is_map(Multipliers) ->
+    Value = trim_binary(Value0),
+    LowerValue = lower_binary(Value),
+    {NumberBin, Unit} = split_numeric_suffix(LowerValue),
+    case maps:get(Unit, Multipliers, undefined) of
+        undefined ->
+            throw({error, <<"invalid value for ", Field/binary, ": ", Value0/binary>>});
+        Multiplier ->
+            trunc(parse_decimal(NumberBin, Field) * Multiplier)
+    end.
+
+-spec parse_uint32(binary(), binary()) -> non_neg_integer().
+parse_uint32(Value, Field) when is_binary(Value), is_binary(Field) ->
+    Parsed = parse_uint64(Value, Field),
+    case Parsed =< 16#FFFFFFFF of
+        true ->
+            Parsed;
+        false ->
+            throw({error, <<"value overflow for ", Field/binary>>})
+    end.
+
+-spec parse_tcp_port(binary(), binary()) -> non_neg_integer().
+parse_tcp_port(Value, Field) when is_binary(Value), is_binary(Field) ->
+    Parsed = parse_uint32(Value, Field),
+    case Parsed =< 65535 of
+        true ->
+            Parsed;
+        false ->
+            throw({error, <<"port out of range for ", Field/binary>>})
+    end.
+
+-spec parse_uint64(binary(), binary()) -> non_neg_integer().
+parse_uint64(Value0, Field) when is_binary(Value0), is_binary(Field) ->
+    Value = trim_binary(Value0),
+    case catch binary_to_integer(Value) of
+        Parsed when is_integer(Parsed), Parsed >= 0 ->
+            Parsed;
+        _ ->
+            throw({error, <<"invalid unsigned integer for ", Field/binary, ": ", Value0/binary>>})
+    end.
+
+-spec parse_decimal(binary(), binary()) -> float().
+parse_decimal(Value, Field) when is_binary(Value), is_binary(Field) ->
+    case catch binary_to_integer(Value) of
+        ParsedInt when is_integer(ParsedInt), ParsedInt >= 0 ->
+            float(ParsedInt);
+        _ ->
+            case catch binary_to_float(Value) of
+                ParsedFloat when is_float(ParsedFloat), ParsedFloat >= 0 ->
+                    ParsedFloat;
+                _ ->
+                    throw({error, <<"invalid number for ", Field/binary, ": ", Value/binary>>})
+            end
+    end.
+
+-spec split_numeric_suffix(binary()) -> {binary(), binary()}.
+split_numeric_suffix(Value) when is_binary(Value) ->
+    split_numeric_suffix(Value, <<>>).
+
+-spec split_numeric_suffix(binary(), binary()) -> {binary(), binary()}.
+split_numeric_suffix(<<Char, Rest/binary>>, Acc)
+  when (Char >= $0 andalso Char =< $9) orelse Char =:= $. ->
+    split_numeric_suffix(Rest, <<Acc/binary, Char>>);
+split_numeric_suffix(Rest, <<>>) ->
+    throw({error, <<"invalid numeric value: ", Rest/binary>>});
+split_numeric_suffix(Rest, Acc) ->
+    {Acc, Rest}.
+
+-spec trim_binary(binary()) -> binary().
+trim_binary(Value) when is_binary(Value) ->
+    Trimmed = string:trim(binary_to_list(Value)),
+    list_to_binary(Trimmed).
+
+-spec lower_binary(binary()) -> binary().
+lower_binary(Value) when is_binary(Value) ->
+    list_to_binary(string:lowercase(binary_to_list(Value))).

apps/iot/src/endpoint/endpoint.erl

@@ -1,61 +0,0 @@
-
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 06. 7月 2023 12:02
-%%%-------------------------------------------------------------------
--module(endpoint).
--include("endpoint.hrl").
-
-%% API
--export([start_link/1]).
--export([get_name/1, get_pid/1, forward/3, reload/2, clean_up/1]).
--export([get_alias_pid/1]).
-
-%%%===================================================================
-%%% API
-%%%===================================================================
-
--spec start_link(Endpoint :: #endpoint{}) -> {'ok', pid()} | 'ignore' | {'error', term()}.
-start_link(Endpoint = #endpoint{id = Id, name = Name, config = #http_endpoint{}}) ->
-    LocalName = get_name(Id),
-    AliasName = get_alias_name(Name),
-    endpoint_http:start_link(LocalName, AliasName, Endpoint);
-start_link(Endpoint = #endpoint{id = Id, name = Name, config = #mqtt_endpoint{}}) ->
-    LocalName = get_name(Id),
-    AliasName = get_alias_name(Name),
-    endpoint_mqtt:start_link(LocalName, AliasName, Endpoint);
-start_link(Endpoint = #endpoint{id = Id, name = Name, config = #kafka_endpoint{}}) ->
-    LocalName = get_name(Id),
-    AliasName = get_alias_name(Name),
-    endpoint_kafka:start_link(LocalName, AliasName, Endpoint).
-
--spec get_name(Id :: integer()) -> atom().
-get_name(Id) when is_integer(Id) ->
-    list_to_atom("endpoint:" ++ integer_to_list(Id)).
-
--spec get_pid(Id :: integer()) -> undefined | pid().
-get_pid(Id) when is_integer(Id) ->
-    whereis(get_name(Id)).
-
--spec get_alias_name(Name :: binary()) -> atom().
-get_alias_name(Name) when is_binary(Name) ->
-    list_to_atom("endpoint:" ++ binary_to_list(Name)).
-
--spec get_alias_pid(Name :: binary()) -> undefined | pid().
-get_alias_pid(Name) when is_binary(Name) ->
-    gproc:whereis_name({n, l, get_alias_name(Name)}).
-
--spec forward(Pid :: pid(), ServiceId :: binary(), Metric :: binary()) -> no_return().
-forward(Pid, ServiceId, Metric) when is_pid(Pid), is_binary(ServiceId), is_binary(Metric) ->
-    gen_server:cast(Pid, {forward, ServiceId, Metric}).
-
-reload(Pid, NEndpoint = #endpoint{}) when is_pid(Pid) ->
-    gen_statem:cast(Pid, {reload, NEndpoint}).
-
--spec clean_up(Pid :: pid()) -> ok.
-clean_up(Pid) when is_pid(Pid) ->
-    gen_server:call(Pid, clean_up, 5000).

apps/iot/src/endpoint/endpoint_buffer.erl

@@ -1,105 +0,0 @@
-
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 06. 7月 2023 12:02
-%%%-------------------------------------------------------------------
--module(endpoint_buffer).
-
--include("endpoint.hrl").
-
-%% 消息重发间隔
--define(RETRY_INTERVAL, 5000).
-
--export([new/2, append/2, trigger_next/1, trigger_n/1, cleanup/1, ack/2, stat/1]).
--export_type([buffer/0]).
-
--record(buffer, {
-    endpoint :: #endpoint{},
-    next_id = 1 :: integer(),
-    %% 当前数据所在的游标
-    cursor = 0 :: integer(),
-    %% ets存储的引用
-    tid :: ets:tid(),
-    %% 定时器
-    timer_pid :: pid(),
-    %% 窗口大小，允许最大的未确认消息数
-    window_size = 10,
-    %% 未确认的消息数
-    flight_num = 0,
-    %% 记录成功处理的消息数
-    acc_num = 0
-}).
-
--record(north_data, {
-    id :: integer(),
-    tuple :: any()
-}).
-
--type buffer() :: #buffer{}.
-
--spec new(Endpoint :: #endpoint{}, WindowSize :: integer()) -> Buffer :: #buffer{}.
-new(Endpoint = #endpoint{id = Id}, WindowSize) when is_integer(WindowSize), WindowSize > 0 ->
-    %% 初始化存储
-    EtsName = list_to_atom("endpoint_buffer_ets:" ++ integer_to_list(Id)),
-    Tid = ets:new(EtsName, [ordered_set, private, {keypos, 2}]),
-    %% 定义重发器
-    {ok, TimerPid} = endpoint_timer:start_link(?RETRY_INTERVAL),
-
-    #buffer{cursor = 0, tid = Tid, timer_pid = TimerPid, endpoint = Endpoint, window_size = WindowSize}.
-
--spec append(tuple(), Buffer :: #buffer{}) -> NBuffer :: #buffer{}.
-append(Tuple, Buffer = #buffer{tid = Tid, next_id = NextId, window_size = WindowSize, flight_num = FlightNum}) ->
-    NorthData = #north_data{id = NextId, tuple = Tuple},
-    true = ets:insert(Tid, NorthData),
-    NBuffer = Buffer#buffer{next_id = NextId + 1},
-    case FlightNum < WindowSize of
-        true ->
-            trigger_next(NBuffer);
-        false ->
-            NBuffer
-    end.
-
--spec trigger_n(Buffer :: #buffer{}) -> NBuffer :: #buffer{}.
-trigger_n(Buffer = #buffer{window_size = WindowSize}) ->
-    %% 最多允许window_size
-    lists:foldl(fun(_, Buffer0) -> trigger_next(Buffer0) end, Buffer, lists:seq(1, WindowSize)).
-
-%% 触发读取下一条数据
--spec trigger_next(Buffer :: #buffer{}) -> NBuffer :: #buffer{}.
-trigger_next(Buffer = #buffer{tid = Tid, cursor = Cursor, timer_pid = TimerPid, flight_num = FlightNum}) ->
-    case ets:first(Tid) of
-        '$end_of_table' ->
-            Buffer;
-        Key ->
-            [#north_data{id = Id, tuple = Tuple}|_] = ets:take(Tid, Key),
-            ReceiverPid = self(),
-            ReceiverPid ! {next_data, Id, Tuple},
-            endpoint_timer:task(TimerPid, Id, fun() -> ReceiverPid ! {next_data, Id, Tuple} end),
-            Buffer#buffer{flight_num = FlightNum + 1, cursor = Cursor + 1}
-    end.
-
--spec ack(Id :: integer(), Buffer :: #buffer{}) -> NBuffer :: #buffer{}.
-ack(Id, Buffer = #buffer{timer_pid = TimerPid, acc_num = AccNum, flight_num = FlightNum}) when is_integer(Id) ->
-    endpoint_timer:ack(TimerPid, Id),
-    trigger_next(Buffer#buffer{acc_num = AccNum + 1, flight_num = FlightNum - 1}).
-
-%% 获取当前统计信息
--spec stat(Buffer :: #buffer{}) -> map().
-stat(#buffer{acc_num = AccNum, tid = Tid}) ->
-    #{
-        <<"acc_num">> => AccNum,
-        <<"queue_num">> => ets:info(Tid, size)
-    }.
-
--spec cleanup(Buffer :: #buffer{}) -> ok.
-cleanup(#buffer{timer_pid = TimerPid}) ->
-    endpoint_timer:cleanup(TimerPid),
-    ok.
-
-%%%===================================================================
-%%% Internal functions
-%%%===================================================================

apps/iot/src/endpoint/endpoint_kafka.erl

@@ -1,185 +0,0 @@
-
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 06. 7月 2023 12:02
-%%%-------------------------------------------------------------------
--module(endpoint_kafka).
-
--include("endpoint.hrl").
--behaviour(gen_server).
-
-%% API
--export([start_link/3]).
-
-%% gen_server callbacks
--export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
-
-%% 消息重发间隔
--define(RETRY_INTERVAL, 5000).
-
--define(DISCONNECTED, disconnected).
--define(CONNECTED, connected).
-
--record(state, {
-    endpoint :: #endpoint{},
-    buffer :: endpoint_buffer:buffer(),
-    client_id :: atom(),
-    client_pid :: undefined | pid(),
-    status = ?DISCONNECTED
-}).
-
-%%%===================================================================
-%%% API
-%%%===================================================================
-
-%% @doc Creates a gen_statem process which calls Module:init/1 to
-%% initialize. To ensure a synchronized start-up procedure, this
-%% function does not return until Module:init/1 has returned.
-start_link(LocalName, AliasName, Endpoint = #endpoint{}) when is_atom(LocalName), is_atom(AliasName) ->
-    gen_server:start_link({local, LocalName}, ?MODULE, [AliasName, Endpoint], []).
-
-%%%===================================================================
-%%% gen_statem callbacks
-%%%===================================================================
-
-%% @private
-%% @doc Whenever a gen_statem is started using gen_statem:start/[3,4] or
-%% gen_statem:start_link/[3,4], this function is called by the new
-%% process to initialize.
-init([AliasName, Endpoint = #endpoint{id = Id}]) ->
-    erlang:process_flag(trap_exit, true),
-    true = gproc:reg({n, l, AliasName}),
-
-    %% 创建转发器, 避免阻塞当前进程的创建，因此采用了延时初始化的机制
-    erlang:start_timer(0, self(), connect),
-    %% 初始化存储
-    Buffer = endpoint_buffer:new(Endpoint, 10),
-    ClientId = list_to_atom("brod_client:" ++ integer_to_list(Id)),
-
-    {ok, #state{endpoint = Endpoint, buffer = Buffer, status = ?DISCONNECTED, client_id = ClientId}}.
-
-%% @private
-%% @doc Handling call messages
--spec(handle_call(Request :: term(), From :: {pid(), Tag :: term()},
-    State :: #state{}) ->
-    {reply, Reply :: term(), NewState :: #state{}} |
-    {reply, Reply :: term(), NewState :: #state{}, timeout() | hibernate} |
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), Reply :: term(), NewState :: #state{}} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_call(get_stat, _From, State = #state{buffer = Buffer}) ->
-    Stat = endpoint_buffer:stat(Buffer),
-    {reply, {ok, Stat}, State}.
-
-%% @private
-%% @doc Handling cast messages
--spec(handle_cast(Request :: term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_cast({forward, ServiceId, Metric}, State = #state{buffer = Buffer}) ->
-    NBuffer = endpoint_buffer:append({ServiceId, Metric}, Buffer),
-    {noreply, State#state{buffer = NBuffer}}.
-
-%% @private
-%% @doc Handling all non call/cast messages
--spec(handle_info(Info :: timeout() | term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_info({timeout, _, connect}, State = #state{buffer = Buffer, status = ?DISCONNECTED, client_id = ClientId,
-    endpoint = #endpoint{title = Title, config = #kafka_endpoint{sasl_config = SaslConfig, bootstrap_servers = BootstrapServers, topic = Topic}}}) ->
-    lager:debug("[endpoint_kafka] endpoint: ~p, create postman", [Title]),
-
-    BaseConfig = [
-        {reconnect_cool_down_seconds, 5},
-        {socket_options, [{keepalive, true}]}
-    ],
-
-    ClientConfig = case SaslConfig of
-                     {Mechanism, Username, Password} ->
-                         [{sasl, {Mechanism, Username, Password}}|BaseConfig];
-                     undefined ->
-                         BaseConfig
-                 end,
-
-    case brod:start_link_client(BootstrapServers, ClientId, ClientConfig) of
-        {ok, ClientPid} ->
-            ok = brod:start_producer(ClientId, Topic, _ProducerConfig = []),
-            NBuffer = endpoint_buffer:trigger_next(Buffer),
-            {noreply, State#state{buffer = NBuffer, client_pid = ClientPid, status = ?CONNECTED}};
-        {error, Reason} ->
-            lager:debug("[endpoint_kafka] start_client: ~p, get error: ~p", [ClientId, Reason]),
-            retry_connect(),
-            {noreply, State#state{status = ?DISCONNECTED, client_pid = undefined}}
-    end;
-
-%% 离线时，忽略数据发送逻辑
-handle_info({next_data, _Id, _Tuple}, State = #state{status = ?DISCONNECTED}) ->
-    {noreply, State};
-%% 发送数据到mqtt服务器
-handle_info({next_data, Id, {_ServiceId, Metric}}, State = #state{status = ?CONNECTED, client_pid = ClientPid,
-    endpoint = #endpoint{config = #kafka_endpoint{topic = Topic}}}) ->
-
-    ReceiverPid = self(),
-    AckCb = fun(Partition, BaseOffset) ->
-        lager:debug("[endpoint_kafka] ack partion: ~p, offset: ~p", [Partition, BaseOffset]),
-        ReceiverPid ! {ack, Id}
-            end,
-    _ = brod:produce_cb(ClientPid, Topic, random, <<>>, Metric, AckCb),
-
-    {noreply, State};
-
-handle_info({ack, Id}, State = #state{buffer = Buffer}) ->
-    NBuffer = endpoint_buffer:ack(Id, Buffer),
-    {noreply, State#state{buffer = NBuffer}};
-
-%% postman进程挂掉时，重新建立新的
-handle_info({'EXIT', ClientPid, Reason}, State = #state{client_pid = ClientPid, endpoint = #endpoint{title = Title}}) ->
-    lager:warning("[endpoint_kafka] endpoint: ~p, conn pid exit with reason: ~p", [Title, Reason]),
-    retry_connect(),
-    {noreply, State#state{client_pid = undefined, status = ?DISCONNECTED}};
-
-handle_info(Info, State = #state{status = Status}) ->
-    lager:warning("[endpoint_kafka] unknown message: ~p, status: ~p", [Info, Status]),
-    {noreply, State}.
-
-%% @private
-%% @doc This function is called by a gen_server when it is about to
-%% terminate. It should be the opposite of Module:init/1 and do any
-%% necessary cleaning up. When it returns, the gen_server terminates
-%% with Reason. The return value is ignored.
--spec(terminate(Reason :: (normal | shutdown | {shutdown, term()} | term()),
-    State :: #state{}) -> term()).
-terminate(Reason, #state{endpoint = #endpoint{title = Title}, buffer = Buffer}) ->
-    lager:debug("[endpoint_kafka] endpoint: ~p, terminate with reason: ~p", [Title, Reason]),
-    endpoint_buffer:cleanup(Buffer),
-    ok.
-
-%% @private
-%% @doc Convert process state when code is changed
--spec(code_change(OldVsn :: term() | {down, term()}, State :: #state{},
-    Extra :: term()) ->
-    {ok, NewState :: #state{}} | {error, Reason :: term()}).
-code_change(_OldVsn, State = #state{}, _Extra) ->
-    {ok, State}.
-
-%%%===================================================================
-%%% Internal functions
-%%%===================================================================
-
-retry_connect() ->
-    erlang:start_timer(?RETRY_INTERVAL, self(), connect).
-
-check_produce_result(ok) ->
-    true;
-check_produce_result({ok, _}) ->
-    true;
-check_produce_result({ok, _}) ->
-    false.
-

apps/iot/src/endpoint/endpoint_mqtt.erl

@@ -1,197 +0,0 @@
-
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 06. 7月 2023 12:02
-%%%-------------------------------------------------------------------
--module(endpoint_mqtt).
-
--include("endpoint.hrl").
--behaviour(gen_server).
-
-%% API
--export([start_link/3]).
-
-%% gen_server callbacks
--export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
-
-%% 消息重发间隔
--define(RETRY_INTERVAL, 5000).
-
--define(DISCONNECTED, disconnected).
--define(CONNECTED, connected).
-
--record(state, {
-    endpoint :: #endpoint{},
-    buffer :: endpoint_buffer:buffer(),
-    conn_pid :: undefined | pid(),
-    %% 待确认的数据, #{PacketId :: integer() => Id :: integer()}
-    inflight = #{},
-
-    status = disconnected
-}).
-
-%%%===================================================================
-%%% API
-%%%===================================================================
-
-%% @doc Creates a gen_statem process which calls Module:init/1 to
-%% initialize. To ensure a synchronized start-up procedure, this
-%% function does not return until Module:init/1 has returned.
-start_link(LocalName, AliasName, Endpoint = #endpoint{}) when is_atom(LocalName), is_atom(AliasName) ->
-    gen_server:start_link({local, LocalName}, ?MODULE, [AliasName, Endpoint], []).
-
-%%%===================================================================
-%%% gen_statem callbacks
-%%%===================================================================
-
-%% @private
-%% @doc Whenever a gen_statem is started using gen_statem:start/[3,4] or
-%% gen_statem:start_link/[3,4], this function is called by the new
-%% process to initialize.
-init([AliasName, Endpoint]) ->
-    erlang:process_flag(trap_exit, true),
-    true = gproc:reg({n, l, AliasName}),
-    %% 创建转发器, 避免阻塞当前进程的创建，因此采用了延时初始化的机制
-    erlang:start_timer(0, self(), create_postman),
-    %% 初始化存储
-    Buffer = endpoint_buffer:new(Endpoint, 10),
-
-    {ok, #state{endpoint = Endpoint, buffer = Buffer, status = ?DISCONNECTED}}.
-
-%% @private
-%% @doc Handling call messages
--spec(handle_call(Request :: term(), From :: {pid(), Tag :: term()},
-    State :: #state{}) ->
-    {reply, Reply :: term(), NewState :: #state{}} |
-    {reply, Reply :: term(), NewState :: #state{}, timeout() | hibernate} |
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), Reply :: term(), NewState :: #state{}} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_call(get_stat, _From, State = #state{buffer = Buffer}) ->
-    Stat = endpoint_buffer:stat(Buffer),
-    {reply, {ok, Stat}, State}.
-
-%% @private
-%% @doc Handling cast messages
--spec(handle_cast(Request :: term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_cast({forward, ServiceId, Metric}, State = #state{buffer = Buffer}) ->
-    NBuffer = endpoint_buffer:append({ServiceId, Metric}, Buffer),
-    {noreply, State#state{buffer = NBuffer}}.
-
-%% @private
-%% @doc Handling all non call/cast messages
--spec(handle_info(Info :: timeout() | term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_info({timeout, _, create_postman}, State = #state{buffer = Buffer, status = ?DISCONNECTED,
-    endpoint = #endpoint{title = Title, config = #mqtt_endpoint{host = Host, port = Port, username = Username, password = Password, client_id = ClientId}}}) ->
-    lager:debug("[endpoint_mqtt] endpoint: ~p, create postman", [Title]),
-    Opts = [
-        {owner, self()},
-        {clientid, ClientId},
-        {host, binary_to_list(Host)},
-        {port, Port},
-        {tcp_opts, []},
-        {username, binary_to_list(Username)},
-        {password, binary_to_list(Password)},
-        {keepalive, 86400},
-        {auto_ack, true},
-        {connect_timeout, 5000},
-        {proto_ver, v5},
-        {retry_interval, 5000}
-    ],
-
-    {ok, ConnPid} = emqtt:start_link(Opts),
-    lager:debug("[endpoint_mqtt] start connect, options: ~p", [Opts]),
-    case emqtt:connect(ConnPid, 5000) of
-        {ok, _} ->
-            lager:debug("[endpoint_mqtt] connect success, pid: ~p", [ConnPid]),
-            NBuffer = endpoint_buffer:trigger_n(Buffer),
-            {noreply, State#state{conn_pid = ConnPid, buffer = NBuffer, status = ?CONNECTED}};
-        {error, Reason} ->
-            lager:warning("[endpoint_mqtt] connect get error: ~p", [Reason]),
-            erlang:start_timer(5000, self(), create_postman),
-            {noreply, State}
-    end;
-
-%% 离线时，忽略数据发送逻辑
-handle_info({next_data, _Id, _Tuple}, State = #state{status = ?DISCONNECTED}) ->
-    {keep_state, State};
-%% 发送数据到mqtt服务器
-handle_info({next_data, Id, {ServiceId, Metric}}, State = #state{status = ?CONNECTED, conn_pid = ConnPid, buffer = Buffer, inflight = InFlight,
-    endpoint = #endpoint{config = #mqtt_endpoint{topic = Topic0, qos = Qos}}}) ->
-
-    Topic = re:replace(Topic0, <<"\\${service_id}">>, ServiceId, [global, {return, binary}]),
-    lager:debug("[endpoint_mqtt] will publish topic: ~p, metric: ~p, qos: ~p", [Topic, Metric, Qos]),
-    case emqtt:publish(ConnPid, Topic, #{}, Metric, [{qos, Qos}, {retain, true}]) of
-        ok ->
-            NBuffer = endpoint_buffer:ack(Id, Buffer),
-            {noreply, State#state{buffer = NBuffer}};
-        {ok, PacketId} ->
-            {noreply, State#state{inflight = maps:put(PacketId, Id, InFlight)}};
-        {error, Reason} ->
-            lager:warning("[endpoint_mqtt] send message to topic: ~p, get error: ~p", [Topic, Reason]),
-            {stop, Reason, State}
-    end;
-
-handle_info({disconnected, ReasonCode, Properties}, State = #state{status = ?CONNECTED}) ->
-    lager:debug("[endpoint_mqtt] Recv a DISONNECT packet - ReasonCode: ~p, Properties: ~p", [ReasonCode, Properties]),
-    erlang:start_timer(?RETRY_INTERVAL, self(), create_postman),
-    {noreply, State#state{conn_pid = undefined, status = ?DISCONNECTED}};
-
-handle_info({publish, Message = #{packet_id := _PacketId, payload := Payload}}, State = #state{status = ?CONNECTED}) ->
-    lager:debug("[endpoint_mqtt] Recv a publish packet: ~p, payload: ~p", [Message, Payload]),
-    {noreply, State};
-
-%% 收到确认的消息
-handle_info({puback, #{packet_id := PacketId}}, State = #state{status = ?CONNECTED, inflight = Inflight, buffer = Buffer}) ->
-    case maps:take(PacketId, Inflight) of
-        {Id, RestInflight} ->
-            NBuffer = endpoint_buffer:ack(Id, Buffer),
-            {noreply, State#state{buffer = NBuffer, inflight = RestInflight}};
-        error ->
-            {noreply, State}
-    end;
-
-%% postman进程挂掉时，重新建立新的
-handle_info({'EXIT', ConnPid, Reason}, State = #state{endpoint = #endpoint{title = Title}, conn_pid = ConnPid}) ->
-    lager:warning("[endpoint_mqtt] endpoint: ~p, conn pid exit with reason: ~p", [Title, Reason]),
-    erlang:start_timer(?RETRY_INTERVAL, self(), create_postman),
-    {noreply, State#state{conn_pid = undefined, status = ?DISCONNECTED}};
-
-handle_info(Info, State = #state{status = Status}) ->
-    lager:warning("[endpoint_mqtt] unknown message: ~p, status: ~p", [Info, Status]),
-    {noreply, State}.
-
-%% @private
-%% @doc This function is called by a gen_server when it is about to
-%% terminate. It should be the opposite of Module:init/1 and do any
-%% necessary cleaning up. When it returns, the gen_server terminates
-%% with Reason. The return value is ignored.
--spec(terminate(Reason :: (normal | shutdown | {shutdown, term()} | term()),
-    State :: #state{}) -> term()).
-terminate(Reason, #state{endpoint = #endpoint{title = Title}, buffer = Buffer}) ->
-    lager:debug("[endpoint_mqtt] endpoint: ~p, terminate with reason: ~p", [Title, Reason]),
-    endpoint_buffer:cleanup(Buffer),
-    ok.
-
-%% @private
-%% @doc Convert process state when code is changed
--spec(code_change(OldVsn :: term() | {down, term()}, State :: #state{},
-    Extra :: term()) ->
-    {ok, NewState :: #state{}} | {error, Reason :: term()}).
-code_change(_OldVsn, State = #state{}, _Extra) ->
-    {ok, State}.
-
-%%%===================================================================
-%%% Internal functions
-%%%===================================================================

apps/iot/src/endpoint/endpoint_sup.erl

@@ -1,110 +0,0 @@
-%%%-------------------------------------------------------------------
-%% @doc endpoint top level supervisor.
-%% @end
-%%%-------------------------------------------------------------------
-
--module(endpoint_sup).
-
--behaviour(supervisor).
--include("endpoint.hrl").
-
--export([start_link/0]).
--export([ensured_endpoint_started/1, delete_endpoint/1]).
-
--export([init/1]).
--export([kafka_test/0]).
-
--define(SERVER, ?MODULE).
-
-start_link() ->
-    supervisor:start_link({local, ?SERVER}, ?MODULE, []).
-
-%% sup_flags() = #{strategy => strategy(),         % optional
-%%                 intensity => non_neg_integer(), % optional
-%%                 period => pos_integer()}        % optional
-%% child_spec() = #{id => child_id(),       % mandatory
-%%                  start => mfargs(),      % mandatory
-%%                  restart => restart(),   % optional
-%%                  shutdown => shutdown(), % optional
-%%                  type => worker(),       % optional
-%%                  modules => modules()}   % optional
-init([]) ->
-    SupFlags = #{strategy => one_for_one, intensity => 1000, period => 3600},
-    Endpoints = endpoint_bo:get_all_endpoints(),
-    ChildSpecs = lists:flatmap(fun(EndpointInfo) ->
-        case endpoint_bo:endpoint_record(EndpointInfo) of
-            error ->
-                [];
-            {ok, Endpoint} ->
-                [Endpoint]
-        end
-                               end, Endpoints),
-    {ok, {SupFlags, ChildSpecs}}.
-
-%% internal functions
-
-kafka_test() ->
-    Endpoint = #endpoint{
-        id = 1,
-        %% 全局唯一，在路由规则中通过名称来指定
-        name = <<"kafka_test">>,
-        %% 标题描述
-        title = <<"kafka测试"/utf8>>,
-        %% 配置项, 格式: #{<<"protocol">> => <<"http|https|ws|kafka|mqtt">>, <<"args">> => #{}}
-        config = #kafka_endpoint{
-            %sasl_config = {
-            %    scram_sha_256,
-            %    <<"admin">>,
-            %    <<"lz4rP5UavRTiGZEZK8G51mxHcM5iPC">>
-            %},
-
-            sasl_config = undefined,
-
-            bootstrap_servers = [
-                {"127.0.0.1", 19092}
-            ],
-            topic = <<"metric">>
-        },
-        status = 0,
-        updated_at = 0,
-        created_at = 0
-    },
-    {ok, Pid} = ensured_endpoint_started(Endpoint),
-    ServiceId = <<"service_id_123">>,
-    Metric = <<"this is a test">>,
-    endpoint:forward(Pid, ServiceId, Metric),
-    endpoint:forward(Pid, ServiceId, Metric),
-    endpoint:forward(Pid, ServiceId, Metric),
-    endpoint:forward(Pid, ServiceId, Metric),
-    endpoint:forward(Pid, ServiceId, Metric),
-    endpoint:forward(Pid, ServiceId, Metric),
-    endpoint:forward(Pid, ServiceId, Metric),
-    endpoint:forward(Pid, ServiceId, Metric),
-    endpoint:forward(Pid, ServiceId, Metric),
-    endpoint:forward(Pid, ServiceId, Metric).
-
--spec ensured_endpoint_started(Endpoint :: #endpoint{}) -> {ok, Pid :: pid()} | {error, Reason :: any()}.
-ensured_endpoint_started(Endpoint = #endpoint{}) ->
-    case supervisor:start_child(?MODULE, child_spec(Endpoint)) of
-        {ok, Pid} when is_pid(Pid) ->
-            {ok, Pid};
-        {error, {'already_started', Pid}} when is_pid(Pid) ->
-            {ok, Pid};
-        {error, Error} ->
-            {error, Error}
-    end.
-
--spec delete_endpoint(Id :: integer()) -> ok | {error, Reason :: any()}.
-delete_endpoint(Id) when is_integer(Id) ->
-    Name = endpoint:get_name(Id),
-    supervisor:terminate_child(?MODULE, Name),
-    supervisor:delete_child(?MODULE, Name).
-
-child_spec(Endpoint = #endpoint{id = Id}) ->
-    Name = endpoint:get_name(Id),
-    #{id => Name,
-        start => {endpoint, start_link, [Endpoint]},
-        restart => permanent,
-        shutdown => 2000,
-        type => worker,
-        modules => ['endpoint']}.

apps/iot/src/endpoint/endpoint_timer.erl

@@ -1,128 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author anlicheng
-%%% @copyright (C) 2024, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 07. 5月 2024 10:30
-%%%-------------------------------------------------------------------
--module(endpoint_timer).
--author("anlicheng").
--include("endpoint.hrl").
-
--behaviour(gen_server).
-
-%% API
--export([start_link/1]).
--export([task/3, ack/2, cleanup/1]).
-
-%% gen_server callbacks
--export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
-
--record(state, {
-    retry_interval = 0,
-    %% 定时器
-    timer_map = #{}
-}).
-
-%%%===================================================================
-%%% API
-%%%===================================================================
-
-task(Pid, Id, Task) when is_pid(Pid), is_integer(Id), is_function(Task, 0) ->
-    gen_server:cast(Pid, {task, Id, Task}).
-
-ack(Pid, Id) when is_pid(Pid), is_integer(Id) ->
-    gen_server:cast(Pid, {ack, Id}).
-
-cleanup(Pid) when is_pid(Pid) ->
-    gen_server:cast(Pid, cleanup).
-
-%% @doc Spawns the server and registers the local name (unique)
--spec(start_link(RetryInterval :: integer()) ->
-    {ok, Pid :: pid()} | ignore | {error, Reason :: term()}).
-start_link(RetryInterval) when is_integer(RetryInterval) ->
-    gen_server:start_link(?MODULE, [RetryInterval], []).
-
-%%%===================================================================
-%%% gen_server callbacks
-%%%===================================================================
-
-%% @private
-%% @doc Initializes the server
--spec(init(Args :: term()) ->
-    {ok, State :: #state{}} | {ok, State :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term()} | ignore).
-init([RetryInterval]) ->
-    {ok, #state{retry_interval = RetryInterval}}.
-
-%% @private
-%% @doc Handling call messages
--spec(handle_call(Request :: term(), From :: {pid(), Tag :: term()},
-    State :: #state{}) ->
-    {reply, Reply :: term(), NewState :: #state{}} |
-    {reply, Reply :: term(), NewState :: #state{}, timeout() | hibernate} |
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), Reply :: term(), NewState :: #state{}} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_call(_Request, _From, State = #state{}) ->
-    {reply, ok, State}.
-
-%% @private
-%% @doc Handling cast messages
--spec(handle_cast(Request :: term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_cast({task, Id, Task}, State = #state{retry_interval = RetryInterval, timer_map = TimerMap}) ->
-    TimerRef = erlang:start_timer(RetryInterval, self(), {repost_ticker, {Id, Task}}),
-    {noreply, State#state{timer_map = maps:put(Id, TimerRef, TimerMap)}};
-
-%% 取消
-handle_cast({ack, Id}, State = #state{timer_map = TimerMap}) ->
-    case maps:take(Id, TimerMap) of
-        error ->
-            {noreply, State};
-        {TimerRef, NTimerMap} ->
-            is_reference(TimerRef) andalso erlang:cancel_timer(TimerRef),
-            {noreply, State#state{timer_map = NTimerMap}}
-    end;
-
-handle_cast(cleanup, State = #state{timer_map = TimerMap}) ->
-    lists:foreach(fun({_, TimerRef}) -> catch erlang:cancel_timer(TimerRef) end, maps:to_list(TimerMap)),
-    {noreply, State#state{timer_map = #{}}}.
-
-%% @private
-%% @doc Handling all non call/cast messages
--spec(handle_info(Info :: timeout() | term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_info({timeout, _, {repost_ticker, {Id, Task}}}, State = #state{retry_interval = RetryInterval, timer_map = TimerMap}) ->
-    Task(),
-    TimerRef = erlang:start_timer(RetryInterval, self(), {repost_ticker, {Id, Task}}),
-
-    {noreply, State#state{timer_map = maps:put(Id, TimerRef, TimerMap)}}.
-
-%% @private
-%% @doc This function is called by a gen_server when it is about to
-%% terminate. It should be the opposite of Module:init/1 and do any
-%% necessary cleaning up. When it returns, the gen_server terminates
-%% with Reason. The return value is ignored.
--spec(terminate(Reason :: (normal | shutdown | {shutdown, term()} | term()),
-    State :: #state{}) -> term()).
-terminate(_Reason, _State = #state{}) ->
-    ok.
-
-%% @private
-%% @doc Convert process state when code is changed
--spec(code_change(OldVsn :: term() | {down, term()}, State :: #state{},
-    Extra :: term()) ->
-    {ok, NewState :: #state{}} | {error, Reason :: term()}).
-code_change(_OldVsn, State = #state{}, _Extra) ->
-    {ok, State}.
-
-%%%===================================================================
-%%% Internal functions
-%%%===================================================================

apps/iot/src/host/iot_container_task.erl

@@ -0,0 +1,147 @@
+%%%-------------------------------------------------------------------
+%%% @doc Owns one container deployment task stream.
+%%% @end
+%%%-------------------------------------------------------------------
+-module(iot_container_task).
+
+-behaviour(gen_server).
+
+%% API
+-export([start_link/2]).
+-export([get_pid/2, subscribe/3, stream/3, close/2]).
+
+%% gen_server callbacks
+-export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
+
+-define(MAX_EVENTS, 200).
+-define(STOP_AFTER_CLOSE, 300000).
+
+-record(state, {
+    uuid :: binary(),
+    task_id :: integer(),
+    status = pending :: pending | running | success | fail,
+    events = [] :: [{binary(), binary()}],
+    listeners = #{} :: #{pid() => reference()},
+    close_reason = undefined :: undefined | binary(),
+    stop_ref = undefined :: undefined | reference()
+}).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec start_link(binary(), integer()) -> {ok, pid()} | ignore | {error, term()}.
+start_link(UUID, TaskId) when is_binary(UUID), is_integer(TaskId) ->
+    gen_server:start_link({via, gproc, {n, l, name(UUID, TaskId)}}, ?MODULE, [UUID, TaskId], []).
+
+-spec get_pid(binary(), integer()) -> undefined | pid().
+get_pid(UUID, TaskId) when is_binary(UUID), is_integer(TaskId) ->
+    gproc:whereis_name({n, l, name(UUID, TaskId)}).
+
+-spec subscribe(binary(), integer(), pid()) -> ok | {error, term()}.
+subscribe(UUID, TaskId, ListenerPid) when is_binary(UUID), is_integer(TaskId), is_pid(ListenerPid) ->
+    case iot_container_task_sup:ensure_started(UUID, TaskId) of
+        {ok, Pid} ->
+            gen_server:call(Pid, {subscribe, ListenerPid});
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec stream(pid(), binary(), binary()) -> ok.
+stream(Pid, Type, Stream) when is_pid(Pid), is_binary(Type), is_binary(Stream) ->
+    gen_server:cast(Pid, {stream, Type, Stream}).
+
+-spec close(pid(), binary()) -> ok.
+close(Pid, Reason) when is_pid(Pid), is_binary(Reason) ->
+    gen_server:cast(Pid, {close, Reason}).
+
+%%%===================================================================
+%%% gen_server callbacks
+%%%===================================================================
+
+-spec init([binary() | integer()]) -> {ok, #state{}}.
+init([UUID, TaskId]) ->
+    ok = iot_log:set_metadata(),
+    {ok, #state{uuid = UUID, task_id = TaskId}}.
+
+-spec handle_call(term(), {pid(), term()}, #state{}) ->
+    {reply, term(), #state{}}.
+handle_call({subscribe, ListenerPid}, _From, State0 = #state{task_id = TaskId, events = Events, listeners = Listeners, close_reason = CloseReason}) ->
+    MonRef = erlang:monitor(process, ListenerPid),
+    maps:get(ListenerPid, Listeners, undefined) =/= undefined andalso erlang:demonitor(maps:get(ListenerPid, Listeners), [flush]),
+    lists:foreach(fun({Type, Stream}) ->
+        ListenerPid ! {stream_data, TaskId, Type, Stream}
+    end, Events),
+    case CloseReason of
+        undefined ->
+            ok;
+        Reason ->
+            ListenerPid ! {stream_close, TaskId, Reason}
+    end,
+    {reply, ok, State0#state{listeners = maps:put(ListenerPid, MonRef, Listeners)}};
+handle_call(_Request, _From, State) ->
+    {reply, ok, State}.
+
+-spec handle_cast(term(), #state{}) -> {noreply, #state{}}.
+handle_cast({stream, Type, Stream}, State0 = #state{task_id = TaskId, events = Events0, listeners = Listeners}) ->
+    Event = {Type, Stream},
+    Events = trim_events(Events0 ++ [Event]),
+    maps:foreach(fun(ListenerPid, _MonRef) ->
+        ListenerPid ! {stream_data, TaskId, Type, Stream}
+    end, Listeners),
+    {noreply, State0#state{status = running, events = Events}};
+handle_cast({close, Reason}, State0 = #state{task_id = TaskId, listeners = Listeners}) ->
+    maps:foreach(fun(ListenerPid, _MonRef) ->
+        ListenerPid ! {stream_close, TaskId, Reason}
+    end, Listeners),
+    Status = close_status(Reason),
+    {noreply, ensure_stop_timer(State0#state{status = Status, close_reason = Reason})};
+handle_cast(_Request, State) ->
+    {noreply, State}.
+
+-spec handle_info(term(), #state{}) -> {noreply, #state{}} | {stop, normal, #state{}}.
+handle_info({'DOWN', _Ref, process, ListenerPid, _Reason}, State = #state{listeners = Listeners}) ->
+    {noreply, State#state{listeners = maps:remove(ListenerPid, Listeners)}};
+handle_info(stop_after_close, State) ->
+    {stop, normal, State};
+handle_info(_Info, State) ->
+    {noreply, State}.
+
+-spec terminate(term(), #state{}) -> ok.
+terminate(_Reason, _State) ->
+    ok.
+
+-spec code_change(term(), #state{}, term()) -> {ok, #state{}}.
+code_change(_OldVsn, State, _Extra) ->
+    {ok, State}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec name(binary(), integer()) -> term().
+name(UUID, TaskId) ->
+    {iot_container_task, UUID, TaskId}.
+
+-spec trim_events([{binary(), binary()}]) -> [{binary(), binary()}].
+trim_events(Events) ->
+    Len = length(Events),
+    case Len > ?MAX_EVENTS of
+        true ->
+            lists:nthtail(Len - ?MAX_EVENTS, Events);
+        false ->
+            Events
+    end.
+
+-spec close_status(binary()) -> success | fail.
+close_status(<<"success">>) ->
+    success;
+close_status(_) ->
+    fail.
+
+-spec ensure_stop_timer(#state{}) -> #state{}.
+ensure_stop_timer(State = #state{stop_ref = undefined}) ->
+    Ref = erlang:send_after(?STOP_AFTER_CLOSE, self(), stop_after_close),
+    State#state{stop_ref = Ref};
+ensure_stop_timer(State) ->
+    State.

apps/iot/src/host/iot_container_task_sup.erl

@@ -0,0 +1,100 @@
+%%%-------------------------------------------------------------------
+%%% @doc Dynamic supervisor for container deployment tasks.
+%%% @end
+%%%-------------------------------------------------------------------
+-module(iot_container_task_sup).
+
+-behaviour(supervisor).
+
+%% API
+-export([start_link/0]).
+-export([ensure_started/2, stream/4, close/3, fail/3]).
+
+%% supervisor callbacks
+-export([init/1]).
+
+-define(SERVER, ?MODULE).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+
+-spec start_link() -> {ok, pid()} | ignore | {error, term()}.
+start_link() ->
+    supervisor:start_link({local, ?SERVER}, ?MODULE, []).
+
+-spec ensure_started(binary(), integer()) -> {ok, pid()} | {error, term()}.
+ensure_started(UUID, TaskId) when is_binary(UUID), is_integer(TaskId) ->
+    case iot_container_task:get_pid(UUID, TaskId) of
+        undefined ->
+            case supervisor:start_child(?SERVER, [UUID, TaskId]) of
+                {ok, Pid} ->
+                    {ok, Pid};
+                {ok, Pid, _Info} ->
+                    {ok, Pid};
+                {error, {already_started, Pid}} ->
+                    {ok, Pid};
+                {error, {shutdown, {failed_to_start_child, _Id, {already_started, Pid}}}} ->
+                    {ok, Pid};
+                {error, Reason} ->
+                    {error, Reason}
+            end;
+        Pid when is_pid(Pid) ->
+            {ok, Pid}
+    end.
+
+-spec stream(binary(), integer(), binary(), binary()) -> ok.
+stream(UUID, TaskId, Type, Stream)
+  when is_binary(UUID), is_integer(TaskId), is_binary(Type), is_binary(Stream) ->
+    case ensure_started(UUID, TaskId) of
+        {ok, Pid} ->
+            iot_container_task:stream(Pid, Type, Stream);
+        {error, Reason} ->
+            logger:warning("[iot_container_task_sup] start task failed, uuid: ~p, task_id: ~p, reason: ~p", [UUID, TaskId, Reason]),
+            ok
+    end.
+
+-spec close(binary(), integer(), binary()) -> ok.
+close(UUID, TaskId, Reason) when is_binary(UUID), is_integer(TaskId), is_binary(Reason) ->
+    case ensure_started(UUID, TaskId) of
+        {ok, Pid} ->
+            iot_container_task:close(Pid, Reason);
+        {error, StartReason} ->
+            logger:warning("[iot_container_task_sup] start task before close failed, uuid: ~p, task_id: ~p, reason: ~p", [UUID, TaskId, StartReason]),
+            ok
+    end.
+
+-spec fail(binary(), integer(), term()) -> ok.
+fail(UUID, TaskId, Reason) when is_binary(UUID), is_integer(TaskId) ->
+    ReasonBin = reason_to_binary(Reason),
+    ok = stream(UUID, TaskId, <<"error">>, ReasonBin),
+    close(UUID, TaskId, <<"fail">>).
+
+%%%===================================================================
+%%% supervisor callbacks
+%%%===================================================================
+
+-spec init([]) -> {ok, {{simple_one_for_one, non_neg_integer(), pos_integer()}, [supervisor:child_spec()]}}.
+init([]) ->
+    SupFlags = {simple_one_for_one, 10, 60},
+    ChildSpec = #{
+        id => iot_container_task,
+        start => {iot_container_task, start_link, []},
+        restart => temporary,
+        shutdown => 5000,
+        type => worker,
+        modules => [iot_container_task]
+    },
+    {ok, {SupFlags, [ChildSpec]}}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec reason_to_binary(term()) -> binary().
+reason_to_binary(Reason) when is_binary(Reason) ->
+    Reason;
+reason_to_binary(timeout) ->
+    <<"timeout">>;
+reason_to_binary(Reason) ->
+    unicode:characters_to_binary(io_lib:format("~p", [Reason])).

apps/iot/src/host/iot_host.erl

@@ -0,0 +1,458 @@
+%%%-------------------------------------------------------------------
+%%% @author
+%%% @copyright (C) 2023, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 22. 9月 2023 16:38
+%%%-------------------------------------------------------------------
+-module(iot_host).
+-author("aresei").
+-include("iot.hrl").
+-include("domain_model.hrl").
+
+-behaviour(gen_statem).
+
+%% 心跳包检测时间间隔, 15分钟检测一次
+-define(HEARTBEAT_INTERVAL, 900 * 1000).
+
+%% 状态
+-define(STATE_DENIED, denied).
+-define(STATE_ACTIVATED, activated).
+
+%% API
+-export([start_link/2, get_name/1, get_alias_name/1, get_pid/1, handle/2, activate/2]).
+-export([lookup_pid/1]).
+-export([get_metric/1, get_status/1, kill/1]).
+%% 通讯相关
+-export([pub/4, attach_channel/2]).
+-export([deploy_container/3, start_container/2, stop_container/2, remove_container/2, kill_container/2, config_container/3, get_containers/1, await_reply/3]).
+-export([heartbeat/1]).
+
+%% gen_statem callbacks
+-export([init/1, handle_event/4, terminate/3, code_change/4, callback_mode/0]).
+
+-record(state, {
+    host_id :: integer(),
+    %% 从数据库里面读取到的数据
+    uuid :: binary(),
+    %% 主机状态
+    host_status :: integer(),
+
+    has_session = false :: boolean(),
+    %% 心跳计数器
+    heartbeat_counter = 0 :: integer(),
+    %% websocket相关
+    channel_pid :: undefined | pid(),
+    %% 主机的相关信息
+    metrics = #{} :: map()
+}).
+
+%%%===================================================================
+%%% API
+%%%===================================================================
+-spec get_pid(UUID :: binary()) -> undefined | pid().
+get_pid(UUID) when is_binary(UUID) ->
+    Name = get_name(UUID),
+    gproc:whereis_name({n, l, Name}).
+
+-spec lookup_pid(UUID :: binary()) -> {ok, pid()} | {error, Reason :: any()}.
+lookup_pid(UUID) when is_binary(UUID) ->
+    case get_pid(UUID) of
+        undefined ->
+            {error, <<"host not found">>};
+        HostPid ->
+            {ok, HostPid}
+    end.
+
+-spec get_name(UUID :: binary()) -> term().
+get_name(UUID) when is_binary(UUID) ->
+    {iot_host, UUID}.
+
+-spec get_alias_name(HostId :: integer()) -> term().
+get_alias_name(HostId0) when is_integer(HostId0) ->
+    {iot_host_id, HostId0}.
+
+-spec kill(UUID :: binary()) -> ok | true.
+kill(UUID) when is_binary(UUID) ->
+    case get_pid(UUID) of
+        undefined ->
+            ok;
+        Pid ->
+            exit(Pid, kill)
+    end.
+
+%% 处理消息
+-spec handle(Pid :: pid(), Packet :: tuple()) -> ok.
+handle(Pid, Packet) when is_pid(Pid) ->
+    gen_statem:cast(Pid, {handle, Packet}).
+
+-spec get_status(Pid :: pid()) -> {ok, Status :: map()}.
+get_status(Pid) when is_pid(Pid) ->
+    gen_statem:call(Pid, get_status).
+
+%% 激活主机, true 表示激活； false表示关闭激活
+-spec activate(Pid :: pid(), Auth :: boolean()) -> ok | {error, term()}.
+activate(Pid, Auth) when is_pid(Pid), is_boolean(Auth) ->
+    gen_statem:call(Pid, {activate, Auth}).
+
+-spec get_metric(Pid :: pid()) -> {ok, MetricInfo :: map()}.
+get_metric(Pid) when is_pid(Pid) ->
+    gen_statem:call(Pid, get_metric).
+
+-spec attach_channel(pid(), pid()) -> ok | {error, Reason :: binary()}.
+attach_channel(Pid, ChannelPid) when is_pid(Pid), is_pid(ChannelPid) ->
+    gen_statem:call(Pid, {attach_channel, ChannelPid}).
+
+-type request_ref() :: binary().
+
+-spec get_containers(Pid :: pid()) -> {ok, Ref :: request_ref()} | {error, Reason :: any()}.
+get_containers(Pid) when is_pid(Pid) ->
+    container_call(Pid, docker_container_builder:list_request()).
+
+-spec config_container(Pid :: pid(), ContainerName :: binary(), ConfigJson :: binary()) -> {ok, Ref :: request_ref()} | {error, Reason :: any()}.
+config_container(Pid, ContainerName, ConfigJson) when is_pid(Pid), is_binary(ContainerName), is_binary(ConfigJson) ->
+    container_call(Pid, docker_container_builder:config_request(ContainerName, ConfigJson)).
+
+-spec deploy_container(Pid :: pid(), TaskId :: integer(), Config :: map()) -> {ok, Ref :: request_ref()} | {error, Reason :: any()}.
+deploy_container(Pid, TaskId, Config) when is_pid(Pid), is_integer(TaskId), is_map(Config) ->
+    case docker_container_builder:deploy_request(TaskId, Config) of
+        {ok, Request} ->
+            container_call(Pid, Request);
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec start_container(Pid :: pid(), ContainerName :: binary()) -> {ok, Ref :: request_ref()} | {error, Reason :: any()}.
+start_container(Pid, ContainerName) when is_pid(Pid), is_binary(ContainerName) ->
+    container_call(Pid, docker_container_builder:start_request(ContainerName)).
+
+-spec stop_container(Pid :: pid(), ContainerName :: binary()) -> {ok, Ref :: request_ref()} | {error, Reason :: any()}.
+stop_container(Pid, ContainerName) when is_pid(Pid), is_binary(ContainerName) ->
+    container_call(Pid, docker_container_builder:stop_request(ContainerName)).
+
+-spec kill_container(Pid :: pid(), ContainerName :: binary()) -> {ok, Ref :: request_ref()} | {error, Reason :: any()}.
+kill_container(Pid, ContainerName) when is_pid(Pid), is_binary(ContainerName) ->
+    container_call(Pid, docker_container_builder:kill_request(ContainerName)).
+
+-spec remove_container(Pid :: pid(), ContainerName :: binary()) -> {ok, Ref :: request_ref()} | {error, Reason :: any()}.
+remove_container(Pid, ContainerName) when is_pid(Pid), is_binary(ContainerName) ->
+    container_call(Pid, docker_container_builder:remove_request(ContainerName)).
+
+-spec await_reply(Pid :: pid(), Ref :: request_ref(), Timeout :: integer()) ->
+    ok | {ok, Result :: term()} | {error, Reason :: term()}.
+await_reply(Pid, Ref, Timeout) when is_pid(Pid), is_binary(Ref), is_integer(Timeout) ->
+    receive
+        {command_reply, Ref, ok} ->
+            ok;
+        {command_reply, Ref, {ok, Result}} ->
+            {ok, Result};
+        {command_reply, Ref, {error, Reason}} ->
+            {error, Reason}
+    after Timeout ->
+        ok = gen_statem:call(Pid, {cancel_command_call, Ref}),
+        flush_reply(Ref),
+        {error, timeout}
+    end.
+
+-spec pub(Pid :: pid(), Topic :: binary(), Qos :: integer(),  Content :: binary()) -> ok | {error, Reason :: any()}.
+pub(Pid, Topic, Qos, Content) when is_pid(Pid), is_binary(Topic), is_integer(Qos), is_binary(Content) ->
+    gen_statem:call(Pid, {pub, Topic, Qos, Content}).
+
+-spec heartbeat(Pid :: undefined | pid()) -> ok.
+heartbeat(undefined) ->
+    ok;
+heartbeat(Pid) when is_pid(Pid) ->
+    gen_statem:cast(Pid, heartbeat).
+
+%% @doc Creates a gen_statem process which calls Module:init/1 to
+%% initialize. To ensure a synchronized start-up procedure, this
+%% function does not return until Module:init/1 has returned.
+start_link(Name, UUID) when is_binary(UUID) ->
+    gen_statem:start_link({via, gproc, {n, l, Name}}, ?MODULE, [UUID], []).
+
+%%%===================================================================
+%%% gen_statem callbacks
+%%%===================================================================
+
+%% @private
+%% @doc Whenever a gen_statem is started using gen_statem:start/[3,4] or
+%% gen_statem:start_link/[3,4], this function is called by the new
+%% process to initialize.
+init([UUID]) ->
+    ok = iot_log:set_metadata(),
+    case iot_api_client:get_host_by_uuid(UUID) of
+        {ok, #host_info{id = HostId, authorize_status = AuthorizeStatus, status = HostStatus}} ->
+            %% 通过host_id注册别名, 可以避免通过查询数据库获取HostPid
+            AliasName = get_alias_name(HostId),
+            global:register_name(AliasName, self()),
+
+            %% 心跳检测机制
+            erlang:start_timer(?HEARTBEAT_INTERVAL, self(), heartbeat_ticker),
+
+            StateName = case AuthorizeStatus =:= ?HOST_AUTHORIZED of
+                            true  -> ?STATE_ACTIVATED;
+                            false -> ?STATE_DENIED
+                        end,
+            {ok, StateName, #state{host_id = HostId, uuid = UUID, host_status = HostStatus, has_session = false}};
+        undefined ->
+            logger:warning("[iot_host] host uuid: ~p, load host info failed", [UUID]),
+            ignore
+    end.
+
+%% @private
+%% @doc This function is called by a gen_statem when it needs to find out
+%% the callback mode of the callback module.
+callback_mode() ->
+    handle_event_function.
+
+%% @private
+%% @doc If callback_mode is handle_event_function, then whenever a
+%% gen_statem receives an event from call/2, cast/2, or as a normal
+%% process message, this function is called.
+
+handle_event({call, From}, get_metric, _, State = #state{metrics = Metrics}) ->
+    {keep_state, State, [{reply, From, {ok, Metrics}}]};
+
+%% 获取主机的状态
+handle_event({call, From}, get_status, _, State = #state{channel_pid = ChannelPid, heartbeat_counter = HeartbeatCounter, metrics = Metrics, has_session = HasSession, host_status = HostStatus}) ->
+    HasChannel = (ChannelPid /= undefined),
+    Reply = #{
+        <<"has_channel">> => HasChannel,
+        <<"has_session">> => HasSession,
+        <<"host_status">> => HostStatus,
+        <<"heartbeat_counter">> => HeartbeatCounter,
+        <<"metrics">> => Metrics
+    },
+    {keep_state, State, [{reply, From, {ok, Reply}}]};
+
+handle_event({call, From}, {container_call, ReceiverPid, Request}, _, State = #state{uuid = UUID, channel_pid = ChannelPid, has_session = HasSession}) ->
+    case HasSession andalso is_pid(ChannelPid) of
+        true ->
+            Ref = request_ref(),
+            ok = ssl_channel:container_call(ChannelPid, ReceiverPid, Ref, Request),
+            {keep_state, State, [{reply, From, {ok, Ref}}]};
+        false ->
+            logger:debug("[iot_host] uuid: ~p, invalid state: ~p", [UUID, state_map(State)]),
+            {keep_state, State, [{reply, From, {error, <<"主机离线，发送命令失败"/utf8>>}}]}
+    end;
+
+handle_event({call, From}, {cancel_command_call, Ref}, _, State = #state{channel_pid = ChannelPid}) ->
+    case is_pid(ChannelPid) of
+        true ->
+            ok = ssl_channel:cancel_command_call(ChannelPid, Ref),
+            {keep_state, State, [{reply, From, ok}]};
+        false ->
+            {keep_state, State, [{reply, From, ok}]}
+    end;
+
+%% 发送指令时, pub/sub
+handle_event({call, From}, {pub, Topic, Qos, Content}, StateName, State = #state{uuid = UUID, channel_pid = ChannelPid, has_session = HasSession}) ->
+    case StateName of
+        ?STATE_ACTIVATED ->
+            case HasSession andalso is_pid(ChannelPid) of
+                true ->
+                    logger:debug("[iot_host] host: ~p, publish to topic: ~p, content: ~p", [UUID, Topic, Content]),
+                    %% 通过websocket发送消息
+                    ssl_channel:pub(ChannelPid, Topic, Qos, Content),
+
+                    {keep_state, State, [{reply, From, ok}]};
+                false ->
+                    logger:debug("[iot_host] uuid: ~p, publish to topic: ~p, content: ~p, invalid state: ~p", [UUID, Topic, Content, state_map(State)]),
+                    {keep_state, State, [{reply, From, {error, <<"主机离线，发送失败"/utf8>>}}]}
+            end;
+        ?STATE_DENIED ->
+            {keep_state, State, [{reply, From, {error, <<"主机未激活，发送失败"/utf8>>}}]}
+    end;
+
+%% 激活/关闭授权只修改 iot 本地授权状态；efka 连接保持在线，数据是否处理由 host 状态决定。
+handle_event({call, From}, {activate, Auth}, _, State = #state{uuid = UUID}) ->
+    NStateName = case Auth of
+                     true ->
+                         ?STATE_ACTIVATED;
+                     false ->
+                         ?STATE_DENIED
+                 end,
+    logger:debug("[iot_host] uuid: ~p, state_name change to ~p", [UUID, NStateName]),
+    {next_state, NStateName, State, [{reply, From, ok}]};
+
+%% 绑定channel
+handle_event({call, From}, {attach_channel, ChannelPid}, StateName, State = #state{uuid = UUID, channel_pid = OldChannelPid, host_status = HostStatus}) ->
+    case OldChannelPid == undefined orelse OldChannelPid =:= ChannelPid of
+        true ->
+            case StateName of
+                ?STATE_ACTIVATED ->
+                    erlang:monitor(process, ChannelPid),
+                    %% 更新主机为在线状态
+                    NState = mark_host_online(UUID, HostStatus, State#state{channel_pid = ChannelPid, has_session = true}),
+                    {keep_state, NState, [{reply, From, ok}]};
+                %% 主机未激活
+                ?STATE_DENIED ->
+                    logger:notice("[iot_host] attach_channel host_id uuid: ~p, channel: ~p, host denied locally", [UUID, ChannelPid]),
+                    erlang:monitor(process, ChannelPid),
+                    NState = mark_host_online(UUID, HostStatus, State#state{channel_pid = ChannelPid, has_session = true}),
+                    {keep_state, NState, [{reply, From, ok}]}
+            end;
+        false ->
+            logger:notice("[iot_host] attach_channel host_id uuid: ~p, old channel exists: ~p", [UUID, OldChannelPid]),
+            {keep_state, State, [{reply, From, {error, <<"channel existed">>}}]}
+    end;
+
+%% 数据分发
+handle_event(cast, {handle, {data, RouteKey, MetricBin}}, ?STATE_ACTIVATED,
+    State = #state{uuid = UUID, has_session = true}) ->
+    logger:debug("[iot_host] metric_data host: ~p, route_key: ~p, metric: ~p", [UUID, RouteKey, MetricBin]),
+    endpoint_subscription:publish(get_route_key(RouteKey), MetricBin),
+    {keep_state, State};
+
+handle_event(cast, {handle, {data, RouteKey, MetricBin}}, StateName,
+    State = #state{uuid = UUID, has_session = true}) ->
+    logger:notice("[iot_host] host_uuid: ~p, state_name: ~p, metric_data route_key: ~p, metric: ~p, discard", [UUID, StateName, RouteKey, MetricBin]),
+    {keep_state, State};
+
+%% ping的数据是通过aes加密后的，因此需要在有会话的情况下才行
+handle_event(cast, {handle, {ping, Metrics}}, ?STATE_ACTIVATED, State = #state{uuid = UUID, has_session = true}) ->
+    logger:debug("[iot_host] ping host_id uuid: ~p, get ping: ~p", [UUID, Metrics]),
+    {keep_state, State#state{metrics = Metrics}};
+
+%% 心跳机制
+handle_event(cast, heartbeat, _, State = #state{uuid = UUID, heartbeat_counter = HeartbeatCounter, host_status = HostStatus}) ->
+    NState = mark_host_online(UUID, HostStatus, State#state{heartbeat_counter = HeartbeatCounter + 1}),
+    {keep_state, NState};
+
+%% UDP 心跳丢失但 SSL channel 仍在时，不能把 host 标记为离线。
+handle_event(info, {timeout, _, heartbeat_ticker}, _,
+    State = #state{uuid = UUID, heartbeat_counter = 0, host_status = HostStatus, channel_pid = ChannelPid}) when is_pid(ChannelPid) ->
+
+    erlang:start_timer(?HEARTBEAT_INTERVAL, self(), heartbeat_ticker),
+    logger:warning("[iot_host] uuid: ~p, udp heartbeat lost but ssl channel is alive: ~p", [UUID, ChannelPid]),
+    NState = mark_host_online(UUID, HostStatus, State#state{heartbeat_counter = 0}),
+    {keep_state, NState};
+
+%% 没有收到 UDP 心跳且没有 SSL channel，主机下线, 设备状态不变
+handle_event(info, {timeout, _, heartbeat_ticker}, _, State = #state{uuid = UUID, heartbeat_counter = 0, host_status = HostStatus}) ->
+    logger:warning("[iot_host] uuid: ~p, heartbeat lost, devices will unknown", [UUID]),
+    erlang:start_timer(?HEARTBEAT_INTERVAL, self(), heartbeat_ticker),
+    NState = mark_host_offline(UUID, HostStatus, State#state{channel_pid = undefined, has_session = false, heartbeat_counter = 0}),
+    {keep_state, NState};
+
+%% 其他情况下需要重置系统计数器
+handle_event(info, {timeout, _, heartbeat_ticker}, _, State = #state{}) ->
+    erlang:start_timer(?HEARTBEAT_INTERVAL, self(), heartbeat_ticker),
+    {keep_state, State#state{heartbeat_counter = 0}};
+
+%% 当websocket断开的时候，主机的状态不一定改变；主机的状态改变通过心跳机制，会话状态需要改变
+handle_event(info, {'DOWN', _Ref, process, ChannelPid, Reason}, _, State = #state{uuid = UUID, channel_pid = ChannelPid, has_session = HasSession}) ->
+    logger:warning("[iot_host] uuid: ~p, channel: ~p, down with reason: ~p, has_session: ~p, state: ~p", [UUID, ChannelPid, Reason, HasSession, State]),
+    {keep_state, State#state{channel_pid = undefined, has_session = false}};
+
+handle_event(info, {'DOWN', _Ref, process, Pid, Reason}, _, State = #state{uuid = UUID}) ->
+    logger:debug("[iot_host] uuid: ~p, process_pid: ~p, down with reason: ~p, state: ~p", [UUID, Pid, Reason, State]),
+    {keep_state, State};
+
+handle_event(Event, Info, StateName, State = #state{uuid = UUID}) ->
+    logger:warning("[iot_host] host: ~p, event: ~p, unknown message: ~p, state_name: ~p, state: ~p", [UUID, Event, Info, StateName, state_map(State)]),
+    {keep_state, State}.
+
+%% @private
+%% @doc This function is called by a gen_statem when it is about to
+%% terminate. It should be the opposite of Module:init/1 and do any
+%% necessary cleaning up. When it returns, the gen_statem terminates with
+%% Reason. The return value is ignored.
+terminate(Reason, _StateName, _State = #state{uuid = UUID, has_session = HasSession}) ->
+    logger:debug("[iot_host] host: ~p, terminate with reason: ~p, has_session: ~p", [UUID, Reason, HasSession]),
+    ok.
+
+%% @private
+%% @doc Convert process state when code is changed
+code_change(_OldVsn, StateName, State = #state{}, _Extra) ->
+    {ok, StateName, State}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec container_call(Pid :: pid(), Request :: term()) ->
+    {ok, Ref :: request_ref()} | {error, Reason :: any()}.
+container_call(Pid, Request) when is_pid(Pid) ->
+    gen_statem:call(Pid, {container_call, self(), Request}).
+
+-spec get_route_key(binary()) -> binary().
+get_route_key(<<"">>) ->
+    <<"/">>;
+get_route_key(RouteKey) when is_binary(RouteKey) ->
+    RouteKey.
+
+%% 将当前的state转换成map
+state_map(#state{host_id = HostId, uuid = UUID, host_status = HostStatus, has_session = HasSession, heartbeat_counter = HeartbeatCounter, channel_pid = ChannelPid, metrics = Metrics}) ->
+    #{
+        host_id => HostId,
+        uuid => UUID,
+        host_status => HostStatus,
+        has_session => HasSession,
+        heartbeat_counter => HeartbeatCounter,
+        channel_pid => ChannelPid,
+        metrics => Metrics
+    }.
+
+flush_reply(Ref) ->
+    receive
+        {command_reply, Ref, _Reply} ->
+            ok
+    after 0 ->
+        ok
+    end.
+
+-spec request_ref() -> request_ref().
+request_ref() ->
+    crypto:strong_rand_bytes(16).
+
+-spec mark_host_online(binary(), integer(), #state{}) -> #state{}.
+mark_host_online(UUID, HostStatus, State) ->
+    case maybe_mark_host_online(UUID, HostStatus) of
+        keep_status ->
+            State;
+        {next_status, NStatus} ->
+            State#state{host_status = NStatus}
+    end.
+
+-spec mark_host_offline(binary(), integer(), #state{}) -> #state{}.
+mark_host_offline(UUID, HostStatus, State) ->
+    case maybe_mark_host_offline(UUID, HostStatus) of
+        keep_status ->
+            State;
+        {next_status, NStatus} ->
+            State#state{host_status = NStatus}
+    end.
+
+-spec maybe_mark_host_online(binary(), integer()) -> {next_status, NStatus :: integer()} | keep_status.
+maybe_mark_host_online(UUID, HostStatus) ->
+    case HostStatus of
+        ?HOST_OFFLINE ->
+            case iot_api_client:change_host_status(UUID, ?HOST_ONLINE) of
+                {ok, Result} ->
+                    logger:debug("[iot_host] uuid: ~p, change host status online success: ~p", [UUID, Result]),
+                    {next_status, ?HOST_ONLINE};
+                {error, Reason} ->
+                    logger:warning("[iot_host] uuid: ~p, change host status online failed: ~p", [UUID, Reason]),
+                    keep_status
+            end;
+        _ ->
+            keep_status
+    end.
+
+-spec maybe_mark_host_offline(binary(), integer()) -> {next_status, NStatus :: integer()} | keep_status.
+maybe_mark_host_offline(UUID, HostStatus) ->
+    case HostStatus of
+        ?HOST_ONLINE ->
+            case iot_api_client:change_host_status(UUID, ?HOST_OFFLINE) of
+                {ok, Result} ->
+                    logger:debug("[iot_host] uuid: ~p, change host status offline success: ~p", [UUID, Result]),
+                    {next_status, ?HOST_OFFLINE};
+                {error, Reason} ->
+                    logger:warning("[iot_host] uuid: ~p, change host status offline failed: ~p", [UUID, Reason]),
+                    keep_status
+            end;
+        _ ->
+            keep_status
+    end.

apps/iot/src/host/iot_host_sup.erl

@@ -0,0 +1,85 @@
+%%%-------------------------------------------------------------------
+%%% @author aresei
+%%% @copyright (C) 2023, <COMPANY>
+%%% @doc
+%%% @end
+%%%-------------------------------------------------------------------
+-module(iot_host_sup).
+-include("iot.hrl").
+
+-behaviour(supervisor).
+
+-export([start_link/0, init/1, delete_host/1, ensured_host_started/1]).
+
+start_link() ->
+    supervisor:start_link({local, ?MODULE}, ?MODULE, []).
+
+init([]) ->
+    Specs = lists:map(fun child_spec/1, iot_api_client:get_all_hosts()),
+
+    {ok, {#{strategy => one_for_one, intensity => 1000, period => 3600}, Specs}}.
+
+-spec ensured_host_started(UUID :: binary()) -> {ok, Pid :: pid()} | {error, Reason :: any()}.
+ensured_host_started(UUID) when is_binary(UUID) ->
+    case iot_host:get_pid(UUID) of
+        undefined ->
+            %% 尝试删除下host对应的信息
+            case delete_host(UUID) of
+                ok ->
+                    case supervisor:start_child(?MODULE, child_spec(UUID)) of
+                        {ok, Pid} when is_pid(Pid) ->
+                            {ok, Pid};
+                        {error, {'already_started', Pid}} when is_pid(Pid) ->
+                            {ok, Pid};
+                        {error, Error} ->
+                            {error, Error}
+                    end;
+                {error, Reason} ->
+                    {error, Reason}
+            end;
+        Pid when is_pid(Pid) ->
+            {ok, Pid}
+    end.
+
+delete_host(UUID) when is_binary(UUID) ->
+    Id = iot_host:get_name(UUID),
+    case supervisor:terminate_child(?MODULE, Id) of
+        ok ->
+            delete_host_child(Id, UUID);
+        {error, not_found} ->
+            delete_host_child(Id, UUID);
+        {error, Reason} ->
+            logger:warning("[iot_host_sup] terminate host: ~p failed: ~p", [UUID, Reason]),
+            {error, Reason}
+    end.
+
+delete_host_child(Id, UUID) ->
+    case supervisor:delete_child(?MODULE, Id) of
+        ok ->
+            ok;
+        {error, not_found} ->
+            ok;
+        {error, running} ->
+            %% ensure killed then delete again
+            iot_host:kill(UUID),
+            case supervisor:delete_child(?MODULE, Id) of
+                ok ->
+                    ok;
+                {error, not_found} ->
+                    ok;
+                {error, Reason} ->
+                    {error, Reason}
+            end;
+        {error, Reason} ->
+            {error, Reason}
+    end.
+
+-spec child_spec(UUID :: binary()) -> map().
+child_spec(UUID) when is_binary(UUID) ->
+    Id = iot_host:get_name(UUID),
+    #{id => Id,
+        start => {iot_host, start_link, [Id, UUID]},
+        restart => permanent,
+        shutdown => 2000,
+        type => worker,
+        modules => ['iot_host']}.

apps/iot/src/http/container_handler.erl

@@ -0,0 +1,239 @@
+%%%-------------------------------------------------------------------
+%%% @author licheng5
+%%% @copyright (C) 2020, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 26. 4月 2020 3:36 下午
+%%%-------------------------------------------------------------------
+-module(container_handler).
+-author("licheng5").
+-include("iot.hrl").
+
+-define(REQ_TIMEOUT, 10000).
+
+%% API
+-export([handle_request/4]).
+
+handle_request("GET", "/container/get_all", #{<<"uuid">> := UUID}, _) when is_binary(UUID) ->
+    %% 检查ConfigJson是否是合法的json字符串
+    case iot_host:get_pid(UUID) of
+        undefined ->
+            {ok, 200, iot_util:json_error(-1, <<"host not found">>)};
+        Pid when is_pid(Pid) ->
+                    case iot_host:get_containers(Pid) of
+                {ok, Ref} ->
+                    case iot_host:await_reply(Pid, Ref, ?REQ_TIMEOUT) of
+                        ok ->
+                            {ok, 200, request_success_response(<<"ok">>)};
+                        {ok, Result} ->
+                            {ok, 200, request_success_response(Result)};
+                        {error, Reason} ->
+                            request_error_http_response(Reason)
+                    end;
+                {error, Reason} when is_binary(Reason) ->
+                    {ok, 200, iot_util:json_error(-1, Reason)}
+            end
+    end;
+
+%% 下发config.json, 微服务接受后，保存服务配置
+handle_request("POST", "/container/push_config", _,
+    #{<<"uuid">> := UUID, <<"container_name">> := ContainerName, <<"config">> := Config, <<"timeout">> := Timeout0})
+    when is_binary(UUID), is_binary(ContainerName), is_binary(Config), is_integer(Timeout0) ->
+
+    %% 检查ConfigJson是否是合法的json字符串
+    true = iot_util:is_json(Config),
+    case iot_host:get_pid(UUID) of
+        undefined ->
+            {ok, 200, iot_util:json_error(-1, <<"host not found">>)};
+        Pid when is_pid(Pid) ->
+            Timeout = Timeout0 * 1000,
+            case iot_host:config_container(Pid, ContainerName, Config) of
+                {ok, Ref} ->
+                    case iot_host:await_reply(Pid, Ref, Timeout) of
+                        ok ->
+                            {ok, 200, request_success_response(<<"ok">>)};
+                        {ok, Result} ->
+                            {ok, 200, request_success_response(Result)};
+                        {error, Reason} ->
+                            request_error_http_response(Reason)
+                    end;
+                {error, Reason} when is_binary(Reason) ->
+                    {ok, 200, iot_util:json_error(-1, Reason)}
+            end
+    end;
+
+%% 部署微服务
+handle_request("POST", "/container/deploy", _, #{<<"uuid">> := UUID, <<"task_id">> := TaskId, <<"config">> := Config})
+    when is_binary(UUID), is_integer(TaskId), is_map(Config) ->
+    case iot_host:get_pid(UUID) of
+        undefined ->
+            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
+        Pid when is_pid(Pid) ->
+            case iot_container_task_sup:ensure_started(UUID, TaskId) of
+                {ok, _TaskPid} ->
+                    handle_deploy_container(Pid, UUID, TaskId, Config);
+                {error, Reason} ->
+                    {ok, 500, iot_util:json_error(500, reason_to_binary(Reason))}
+            end
+    end;
+
+%% 启动服务
+handle_request("POST", "/container/start", _, #{<<"uuid">> := UUID, <<"container_name">> := ContainerName}) when is_binary(UUID), is_binary(ContainerName) ->
+    case iot_host:get_pid(UUID) of
+        undefined ->
+            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
+        Pid when is_pid(Pid) ->
+            case iot_host:start_container(Pid, ContainerName) of
+                {ok, Ref} ->
+                    case iot_host:await_reply(Pid, Ref, ?REQ_TIMEOUT) of
+                        ok ->
+                            {ok, 200, request_success_response(<<"ok">>)};
+                        {ok, Result} ->
+                            {ok, 200, request_success_response(Result)};
+                        {error, Reason} ->
+                            request_error_http_response(Reason)
+                    end;
+                {error, Reason} when is_binary(Reason) ->
+                    {ok, 200, iot_util:json_error(400, Reason)}
+            end
+    end;
+
+%% 停止服务
+handle_request("POST", "/container/stop", _, #{<<"uuid">> := UUID, <<"container_name">> := ContainerName}) when is_binary(UUID), is_binary(ContainerName) ->
+    case iot_host:get_pid(UUID) of
+        undefined ->
+            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
+        Pid when is_pid(Pid) ->
+            case iot_host:stop_container(Pid, ContainerName) of
+                {ok, Ref} ->
+                    case iot_host:await_reply(Pid, Ref, ?REQ_TIMEOUT) of
+                        ok ->
+                            {ok, 200, request_success_response(<<"ok">>)};
+                        {ok, Result} ->
+                            {ok, 200, request_success_response(Result)};
+                        {error, Reason} ->
+                            request_error_http_response(Reason)
+                    end;
+                {error, Reason} when is_binary(Reason) ->
+                    {ok, 200, iot_util:json_error(400, Reason)}
+            end
+    end;
+
+handle_request("POST", "/container/kill", _, #{<<"uuid">> := UUID, <<"container_name">> := ContainerName}) when is_binary(UUID), is_binary(ContainerName) ->
+    case iot_host:get_pid(UUID) of
+        undefined ->
+            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
+        Pid when is_pid(Pid) ->
+            case iot_host:kill_container(Pid, ContainerName) of
+                {ok, Ref} ->
+                    case iot_host:await_reply(Pid, Ref, ?REQ_TIMEOUT) of
+                        ok ->
+                            {ok, 200, request_success_response(<<"ok">>)};
+                        {ok, Result} ->
+                            {ok, 200, request_success_response(Result)};
+                        {error, Reason} ->
+                            request_error_http_response(Reason)
+                    end;
+                {error, Reason} when is_binary(Reason) ->
+                    {ok, 200, iot_util:json_error(400, Reason)}
+            end
+    end;
+
+%% 删除容器
+handle_request("POST", "/container/remove", _, #{<<"uuid">> := UUID, <<"container_name">> := ContainerName}) when is_binary(UUID), is_binary(ContainerName) ->
+    case iot_host:get_pid(UUID) of
+        undefined ->
+            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
+        Pid when is_pid(Pid) ->
+            case iot_host:remove_container(Pid, ContainerName) of
+                {ok, Ref} ->
+                    case iot_host:await_reply(Pid, Ref, ?REQ_TIMEOUT) of
+                        ok ->
+                            {ok, 200, request_success_response(<<"ok">>)};
+                        {ok, Result} ->
+                            {ok, 200, request_success_response(Result)};
+                        {error, Reason} ->
+                            request_error_http_response(Reason)
+                    end;
+                {error, Reason} when is_binary(Reason) ->
+                    {ok, 200, iot_util:json_error(400, Reason)}
+            end
+    end;
+
+handle_request(_, Path, _, _) ->
+    Path1 = list_to_binary(Path),
+    {ok, 200, iot_util:json_error(-1, <<"url: ", Path1/binary, " not found">>)}.
+
+request_success_response(Result) when is_binary(Result) ->
+    case decode_json_bytes(Result) of
+        {ok, Data} ->
+            iot_util:json_data(Data);
+        error ->
+            iot_util:json_data(Result)
+    end;
+request_success_response(Result) ->
+    iot_util:json_data(Result).
+
+request_error_response(Code, Reason) when is_integer(Code), is_binary(Reason) ->
+    case decode_json_bytes(Reason) of
+        {ok, #{<<"message">> := Message}} when is_binary(Message) ->
+            iot_util:json_error(Code, Message);
+        {ok, Message} when is_binary(Message) ->
+            iot_util:json_error(Code, Message);
+        _ ->
+            iot_util:json_error(Code, Reason)
+    end.
+
+-spec request_error_http_response(Reason :: term()) ->
+    {ok, HttpStatus :: 400 | 504, Body :: binary()}.
+request_error_http_response(Reason) ->
+    HttpStatus = request_error_status(Reason),
+    {ok, HttpStatus, request_error_response(HttpStatus, reason_to_binary(Reason))}.
+
+-spec request_error_status(Reason :: term()) -> integer().
+request_error_status(timeout) ->
+    504;
+request_error_status(<<"timeout">>) ->
+    504;
+request_error_status(_) ->
+    400.
+
+-spec reason_to_binary(term()) -> binary().
+reason_to_binary(Reason) when is_binary(Reason) ->
+    Reason;
+reason_to_binary(timeout) ->
+    <<"timeout">>;
+reason_to_binary(invalid_response) ->
+    <<"invalid response">>;
+reason_to_binary(Reason) ->
+    unicode:characters_to_binary(io_lib:format("~p", [Reason])).
+
+decode_json_bytes(Data) when is_binary(Data) ->
+    case catch json:decode(Data) of
+        {'EXIT', _} ->
+            error;
+        {error, _} ->
+            error;
+        Decoded ->
+            {ok, Decoded}
+    end.
+
+-spec handle_deploy_container(pid(), binary(), integer(), map()) ->
+    {ok, 200 | 400 | 504, binary()}.
+handle_deploy_container(Pid, UUID, TaskId, Config) ->
+    case iot_host:deploy_container(Pid, TaskId, Config) of
+        {ok, Ref} ->
+            case iot_host:await_reply(Pid, Ref, ?REQ_TIMEOUT) of
+                ok ->
+                    {ok, 200, request_success_response(<<"ok">>)};
+                {ok, Result} ->
+                    {ok, 200, request_success_response(Result)};
+                {error, Reason} ->
+                    ok = iot_container_task_sup:fail(UUID, TaskId, Reason),
+                    request_error_http_response(Reason)
+            end;
+        {error, Reason} when is_binary(Reason) ->
+            ok = iot_container_task_sup:fail(UUID, TaskId, Reason),
+            {ok, 200, iot_util:json_error(400, Reason)}
+    end.

apps/iot/src/http/endpoint_handler.erl

@@ -0,0 +1,155 @@
+%%%-------------------------------------------------------------------
+%%% @author licheng5
+%%% @copyright (C) 2020, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 26. 4月 2020 3:36 下午
+%%%-------------------------------------------------------------------
+-module(endpoint_handler).
+-author("licheng5").
+-include_lib("endpoint/include/endpoint.hrl").
+
+%% API
+-export([handle_request/4]).
+
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%% helper methods
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+
+%% 获取endpoint的运行状态
+handle_request("POST", "/endpoint/run_statuses", _, Ids) when is_list(Ids) ->
+    Statuses = lists:map(fun(Id) ->
+        case endpoint:get_pid(Id) of
+            undefined ->
+                0;
+            Pid when is_pid(Pid) ->
+                1
+        end
+                         end, Ids),
+    {ok, 200, iot_util:json_data(Statuses)};
+
+handle_request("POST", "/endpoint/start", _, #{<<"id">> := Id}) when is_integer(Id) ->
+    case iot_api_client:get_endpoint(Id) of
+        undefined ->
+            {ok, 200, iot_util:json_error(404, <<"endpoint not found">>)};
+        {ok, EndpointInfo} ->
+            case endpoint:endpoint_record(EndpointInfo) of
+                {ok, Endpoint = #endpoint{title = Title}} ->
+                    case endpoint_adapter_sup:ensured_endpoint_started(Endpoint) of
+                        {ok, Pid} when is_pid(Pid) ->
+                            {ok, 200, iot_util:json_data(<<"success">>)};
+                        {error, Reason} ->
+                            logger:warning("[endpoint_handler] start endpoint: ~p, get error: ~p", [Title, Reason]),
+                            {ok, 200, iot_util:json_error(404, <<"start endpoint error">>)}
+                    end;
+                error ->
+                    {ok, 200, iot_util:json_error(404, <<"endpoint invalid">>)}
+            end
+    end;
+
+handle_request("POST", "/endpoint/stop", _, #{<<"id">> := Id}) when is_integer(Id) ->
+    case iot_api_client:get_endpoint(Id) of
+        undefined ->
+            {ok, 200, iot_util:json_error(404, <<"endpoint not found">>)};
+        {ok, _} ->
+            case endpoint_adapter_sup:delete_endpoint(Id) of
+                ok ->
+                    {ok, 200, iot_util:json_data(<<"success">>)};
+                {error, Reason} ->
+                    logger:warning("[endpoint_handler] stop endpoint id: ~p, get error: ~p", [Id, Reason]),
+                    {ok, 200, iot_util:json_error(404, <<"stop endpoint error">>)}
+            end
+    end;
+
+handle_request("POST", "/endpoint/restart", _, #{<<"id">> := Id}) when is_integer(Id) ->
+    case iot_api_client:get_endpoint(Id) of
+        undefined ->
+            {ok, 200, iot_util:json_error(404, <<"endpoint not found">>)};
+        {ok, EndpointInfo} ->
+            case endpoint:endpoint_record(EndpointInfo) of
+                {ok, Endpoint = #endpoint{title = Title}} ->
+                    case endpoint:get_pid(Id) of
+                        undefined ->
+                            case endpoint_adapter_sup:ensured_endpoint_started(Endpoint) of
+                                {ok, Pid} when is_pid(Pid) ->
+                                    {ok, 200, iot_util:json_data(<<"success">>)};
+                                {error, Reason} ->
+                                    logger:warning("[endpoint_handler] start endpoint: ~p, get error: ~p", [Title, Reason]),
+                                    {ok, 200, iot_util:json_error(404, <<"restart endpoint error">>)}
+                            end;
+                        Pid when is_pid(Pid) ->
+                            case endpoint_adapter_sup:delete_endpoint(Id) of
+                                ok ->
+                                    case endpoint_adapter_sup:ensured_endpoint_started(Endpoint) of
+                                        {ok, Pid0} when is_pid(Pid0) ->
+                                            {ok, 200, iot_util:json_data(<<"success">>)};
+                                        {error, Reason} ->
+                                            logger:warning("[endpoint_handler] start endpoint: ~p, get error: ~p", [Title, Reason]),
+                                            {ok, 200, iot_util:json_error(404, <<"restart endpoint error">>)}
+                                    end;
+                                {error, Reason} ->
+                                    logger:warning("[endpoint_handler] start endpoint: ~p, get error: ~p", [Title, Reason]),
+                                    {ok, 200, iot_util:json_error(404, <<"stop endpoint error">>)}
+                            end
+                    end;
+                error ->
+                    {ok, 200, iot_util:json_error(404, <<"endpoint invalid">>)}
+            end
+    end;
+
+%% http接口的测试
+handle_request("POST", "/endpoint/test", _, #{<<"protocol">> := <<"http">>, <<"config">> := #{<<"url">> := Url, <<"pool_size">> := PoolSize}}) when is_integer(PoolSize), PoolSize > 0 ->
+    case endpoint_tester:test(#http_endpoint{url = Url, pool_size = PoolSize}) of
+        ok ->
+            {ok, 200, iot_util:json_data(<<"ok">>)};
+        {error, Reason} ->
+            logger:debug("[endpint_handler] test http: ~p, error: ~p", [Url, Reason]),
+            {ok, 200, iot_util:json_error(-1, <<"url failed">>)}
+    end;
+
+%% 测试mqtt
+handle_request("POST", "/endpoint/test", _, #{<<"protocol">> := <<"mqtt">>, <<"config">> := Config}) ->
+    case endpoint:parse_config(<<"mqtt">>, Config) of
+        {ok, MqttEndpoint = #mqtt_endpoint{}} ->
+            case endpoint_tester:test(MqttEndpoint) of
+                ok ->
+                    {ok, 200, iot_util:json_data(<<"ok">>)};
+                {error, Reason} ->
+                    {ok, 200, iot_util:json_error(-1, Reason)}
+            end;
+        {error, Errors} ->
+            {ok, 200, iot_util:json_error(-1, Errors)}
+    end;
+
+%% 测试kafka
+handle_request("POST", "/endpoint/test", _, #{<<"protocol">> := <<"kafka">>, <<"config">> := Config}) ->
+    case endpoint:parse_config(<<"kafka">>, Config) of
+        {ok, KafkaEndpoint = #kafka_endpoint{}} ->
+            case endpoint_tester:test(KafkaEndpoint) of
+                ok ->
+                    {ok, 200, iot_util:json_data(<<"ok">>)};
+                {error, Reason} ->
+                    {ok, 200, iot_util:json_error(-1, Reason)}
+            end;
+        {error, Errors} ->
+            {ok, 200, iot_util:json_error(-1, Errors)}
+    end;
+
+%% 发布消息
+handle_request("POST", "/endpoint/publish_metric", _, #{<<"route_key">> := RouteKey, <<"metric">> := Metric0}) when is_binary(RouteKey) ->
+    if
+        is_map(Metric0) orelse is_list(Metric0) ->
+            Metric = iolist_to_binary(json:encode(Metric0)),
+            endpoint_subscription:publish(RouteKey, Metric),
+            {ok, 200, iot_util:json_data(<<"ok">>)};
+        is_binary(Metric0) ->
+            endpoint_subscription:publish(RouteKey, Metric0),
+            {ok, 200, iot_util:json_data(<<"ok">>)};
+        true ->
+            {ok, 200, iot_util:json_error(-1, <<"invalid metric">>)}
+    end;
+
+handle_request(_, Path, _, _) ->
+    Path1 = list_to_binary(Path),
+    {ok, 200, iot_util:json_error(-1, <<"url: ", Path1/binary, " not found">>)}.

apps/iot/src/http/event_stream_handler.erl

@@ -0,0 +1,73 @@
+%%%-------------------------------------------------------------------
+%%% @author anlicheng
+%%% @copyright (C) 2025, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 08. 5月 2025 13:00
+%%%-------------------------------------------------------------------
+-module(event_stream_handler).
+-author("anlicheng").
+
+%% API
+-export([init/2]).
+
+init(Req0, Opts) ->
+    ok = iot_log:set_metadata(),
+    Method = binary_to_list(cowboy_req:method(Req0)),
+    Path = binary_to_list(cowboy_req:path(Req0)),
+    GetParams0 = cowboy_req:parse_qs(Req0),
+    GetParams = maps:from_list(GetParams0),
+
+    logger:debug("method: ~p, path: ~p,  get: ~p", [Method, Path, GetParams]),
+    case parse_stream_params(GetParams) of
+        {ok, UUID, TaskId} ->
+            Req1 = cowboy_req:stream_reply(200, #{
+                <<"Content-Type">> => <<"text/event-stream">>,
+                <<"Cache-Control">> => <<"no-cache">>,
+                <<"Connection">> => <<"keep-alive">>
+            }, Req0),
+
+            ok = iot_container_task:subscribe(UUID, TaskId, self()),
+            receiver_events(TaskId, Req1),
+
+            {ok, Req1, Opts};
+        {error, Reason} ->
+            Req1 = cowboy_req:reply(400, #{
+                <<"Content-Type">> => <<"application/json">>
+            }, iot_util:json_error(400, Reason), Req0),
+
+            {ok, Req1, Opts}
+    end.
+
+receiver_events(TaskId, Req) ->
+    receive
+        {stream_data, TaskId, Type, Stream} ->
+            Data = iolist_to_binary(json:encode(#{<<"type">> => Type, <<"stream">> => Stream})),
+            Body = iolist_to_binary([<<"event: message\n">>, <<"data: ", Data/binary, "\n">>, <<"\n">>]),
+            ok = cowboy_req:stream_body(Body, nofin, Req),
+
+            receiver_events(TaskId, Req);
+        {stream_close, TaskId, Reason} ->
+            CloseFrame = iolist_to_binary([<<"event: close\n">>, <<"data: ", Reason/binary, "\n">>, <<"\n">>]),
+            ok = cowboy_req:stream_body(CloseFrame, fin, Req)
+    after 30000 ->
+            ok = cowboy_req:stream_body(<<": heartbeat\n\n">>, nofin, Req),
+            receiver_events(TaskId, Req)
+    end.
+
+-spec parse_stream_params(map()) -> {ok, binary(), integer()} | {error, binary()}.
+parse_stream_params(#{<<"uuid">> := UUID, <<"task_id">> := TaskId0}) when is_binary(UUID), is_binary(TaskId0) ->
+    try binary_to_integer(TaskId0) of
+        TaskId when TaskId >= 0 ->
+            {ok, UUID, TaskId};
+        _ ->
+            {error, <<"task_id must be non-negative">>}
+    catch
+        error:badarg ->
+            {error, <<"task_id must be integer">>}
+    end;
+parse_stream_params(#{<<"task_id">> := _TaskId0}) ->
+    {error, <<"uuid required">>};
+parse_stream_params(_) ->
+    {error, <<"uuid and task_id required">>}.

apps/iot/src/http/host_handler.erl

@@ -18,7 +18,7 @@
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 
 handle_request("GET", "/host/metric", #{<<"uuid">> := UUID}, _) ->
-    lager:debug("[host_handler] get host metric uuid is: ~p", [UUID]),
+    logger:debug("[host_handler] get host metric uuid is: ~p", [UUID]),
     case iot_host:get_pid(UUID) of
         undefined ->
             {ok, 200, iot_util:json_error(404, <<"host not found">>)};
@@ -42,28 +42,13 @@ handle_request("GET", "/host/status", #{<<"uuid">> := UUID}, _) when is_binary(U
             {ok, 200, iot_util:json_data(StatusInfo)}
     end;
 
-%% 重新加载对应的主机信息
-handle_request("POST", "/host/reload", _, #{<<"uuid">> := UUID}) when is_binary(UUID) ->
-    lager:debug("[host_handler] will reload host uuid: ~p", [UUID]),
-    case iot_host_sup:ensured_host_started(UUID) of
-        {ok, Pid} when is_pid(Pid) ->
-            {ok, #{<<"authorize_status">> := AuthorizeStatus}} = host_bo:get_host_by_uuid(UUID),
-            ok = iot_host:activate(Pid, AuthorizeStatus =:= 1),
-            lager:debug("[host_handler] already_started reload host uuid: ~p, success", [UUID]),
-            {ok, 200, iot_util:json_data(<<"success">>)};
-        Error ->
-            lager:debug("[host_handler] reload host uuid: ~p, error: ~p", [UUID, Error]),
-            {ok, 200, iot_util:json_error(404, <<"reload error">>)}
-    end;
-
 %% 删除对应的主机信息
 handle_request("POST", "/host/delete", _, #{<<"uuid">> := UUID}) when is_binary(UUID) ->
-    case iot_host_sup:delete_host(UUID) of
-        ok ->
-            lager:debug("[host_handler] will delete host uuid: ~p", [UUID]),
+    case iot_host:get_pid(UUID) of
+        Pid when is_pid(Pid) ->
+            ok = iot_host_sup:delete_host(UUID),
             {ok, 200, iot_util:json_data(<<"success">>)};
-        {error, Reason} ->
-            lager:debug("[host_handler] delete host uuid: ~p, get error is: ~p", [UUID, Reason]),
+        undefined ->
             {ok, 200, iot_util:json_error(404, <<"error">>)}
     end;
 
@@ -71,39 +56,40 @@ handle_request("POST", "/host/delete", _, #{<<"uuid">> := UUID}) when is_binary(
 handle_request("POST", "/host/activate", _, #{<<"uuid">> := UUID, <<"auth">> := true}) when is_binary(UUID) ->
     case iot_host_sup:ensured_host_started(UUID) of
         {error, Reason} ->
-            lager:debug("[host_handler] activate host_id: ~p, failed with reason: ~p", [UUID, Reason]),
+            logger:debug("[host_handler] activate host_id: ~p, failed with reason: ~p", [UUID, Reason]),
             {ok, 200, iot_util:json_error(400, <<"host not found">>)};
         {ok, Pid} when is_pid(Pid) ->
-            lager:debug("[host_handler] activate host_id: ~p, start", [UUID]),
-            ok = iot_host:activate(Pid, true),
-
-            {ok, 200, iot_util:json_data(<<"success">>)}
+            logger:debug("[host_handler] activate host_id: ~p, start", [UUID]),
+            auth_response(iot_host:activate(Pid, true))
     end;
 
 %% 处理主机的授权的 取消激活
 handle_request("POST", "/host/activate", _, #{<<"uuid">> := UUID, <<"auth">> := false}) when is_binary(UUID) ->
     case iot_host_sup:ensured_host_started(UUID) of
         {error, Reason} ->
-            lager:debug("[host_handler] activate host_id: ~p, failed with reason: ~p", [UUID, Reason]),
+            logger:debug("[host_handler] activate host_id: ~p, failed with reason: ~p", [UUID, Reason]),
             {ok, 200, iot_util:json_error(400, <<"host not found">>)};
         {ok, Pid} when is_pid(Pid) ->
-            lager:debug("[host_handler] activate host_id: ~p, start", [UUID]),
-            ok = iot_host:activate(Pid, false),
-
-            {ok, 200, iot_util:json_data(<<"success">>)}
+            logger:debug("[host_handler] activate host_id: ~p, start", [UUID]),
+            auth_response(iot_host:activate(Pid, false))
     end;
 
 %% 主机事件发布
-handle_request("POST", "/host/pub", _, #{<<"uuid">> := UUID, <<"topic">> := Topic, <<"content">> := Content})
-    when is_binary(UUID), is_binary(Topic), is_binary(Content) ->
+handle_request("POST", "/host/pub", _, #{<<"uuid">> := UUID, <<"topic">> := Topic, <<"qos">> := Qos0, <<"content">> := Content})
+    when is_binary(UUID), is_binary(Topic), is_binary(Content), is_integer(Qos0) ->
 
+    Qos = case Qos0 > 0 of true -> 1; false -> 0 end,
     case iot_host_sup:ensured_host_started(UUID) of
         {error, Reason} ->
-            lager:debug("[host_handler] pub host_id: ~p, topic: ~p, failed with reason: ~p", [UUID, Topic, Reason]),
+            logger:debug("[host_handler] pub host_id: ~p, topic: ~p, failed with reason: ~p", [UUID, Topic, Reason]),
             {ok, 200, iot_util:json_error(400, <<"host not found">>)};
         {ok, Pid} when is_pid(Pid) ->
-            ok = iot_host:pub(Pid, Topic, Content),
-            {ok, 200, iot_util:json_data(<<"success">>)}
+            case iot_host:pub(Pid, Topic, Qos, Content) of
+                ok ->
+                    {ok, 200, iot_util:json_data(<<"success">>)};
+                {error, Reason} ->
+                    {ok, 200, iot_util:json_error(400, Reason)}
+            end
     end;
 
 handle_request(_, Path, _, _) ->
@@ -113,3 +99,28 @@ handle_request(_, Path, _, _) ->
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 %% helper methods
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+
+auth_response(ok) ->
+    {ok, 200, iot_util:json_data(<<"success">>)};
+auth_response({error, Reason}) ->
+    {ok, auth_error_status(Reason), iot_util:json_error(auth_error_status(Reason), reason_to_binary(Reason))}.
+
+auth_error_status(timeout) ->
+    504;
+auth_error_status(<<"timeout">>) ->
+    504;
+auth_error_status(_) ->
+    400.
+
+reason_to_binary(Reason) when is_binary(Reason) ->
+    Reason;
+reason_to_binary(timeout) ->
+    <<"timeout">>;
+reason_to_binary(invalid_response) ->
+    <<"invalid response">>;
+reason_to_binary({failed, Reason}) ->
+    reason_to_binary(Reason);
+reason_to_binary({channel_closed, Reason}) ->
+    <<"channel closed: ", (reason_to_binary(Reason))/binary>>;
+reason_to_binary(Reason) ->
+    unicode:characters_to_binary(io_lib:format("~p", [Reason])).

apps/iot/src/http/http_protocol.erl

@@ -6,49 +6,34 @@
 %%% @end
 %%% Created : 08. 5月 2025 13:00
 %%%-------------------------------------------------------------------
--module(http_server).
+-module(http_protocol).
 -author("anlicheng").
 
 %% API
--export([start/0]).
 -export([init/2]).
 
-%% 启动http服务
-start() ->
-    {ok, Props} = application:get_env(iot, http_server),
-    Acceptors = proplists:get_value(acceptors, Props, 50),
-    MaxConnections = proplists:get_value(max_connections, Props, 10240),
-    Backlog = proplists:get_value(backlog, Props, 1024),
-    Port = proplists:get_value(port, Props),
-
-    Dispatcher = cowboy_router:compile([
-        {'_', [
-            {"/host/[...]", ?MODULE, [host_handler]},
-            {"/service/[...]", ?MODULE, [service_handler]},
-            {"/device/[...]", ?MODULE, [device_handler]}
-        ]}
-    ]),
-
-    TransOpts = [
-        {port, Port},
-        {num_acceptors, Acceptors},
-        {backlog, Backlog},
-        {max_connections, MaxConnections}
-    ],
-    {ok, Pid} = cowboy:start_clear(http_listener, TransOpts, #{env => #{dispatch => Dispatcher}}),
-
-    lager:debug("[http_server] the http server start at: ~p, pid is: ~p", [Port, Pid]).
+-define(MAX_BODY_BYTES, 10 * 1024 * 1024).
 
 init(Req0, Opts = [Mod|_]) ->
+    ok = iot_log:set_metadata(),
     Method = binary_to_list(cowboy_req:method(Req0)),
     Path = binary_to_list(cowboy_req:path(Req0)),
     GetParams0 = cowboy_req:parse_qs(Req0),
     GetParams = maps:from_list(GetParams0),
-    {ok, PostParams, Req1} = parse_body(Req0),
+    case parse_body(Req0) of
+        {ok, PostParams, Req1} ->
+            handle_request(Mod, Method, Path, GetParams, PostParams, Req1, Opts);
+        {error, StatusCode, Resp, Req1} ->
+            Req2 = cowboy_req:reply(StatusCode, #{
+                <<"Content-Type">> => <<"application/json;charset=utf-8">>
+            }, Resp, Req1),
+            {ok, Req2, Opts}
+    end.
 
+handle_request(Mod, Method, Path, GetParams, PostParams, Req1, Opts) ->
     try Mod:handle_request(Method, Path, GetParams, PostParams) of
         {ok, StatusCode, Resp} ->
-            lager:debug("[http_protocol] request path: ~p, get_params: ~p, post_params: ~p, response: ~ts",
+            logger:debug("[http_protocol] request path: ~p, get_params: ~p, post_params: ~p, response: ~ts",
                 [Path, GetParams, PostParams, Resp]),
             AcceptEncoding = cowboy_req:header(<<"accept-encoding">>, Req1, <<>>),
             Req2 = case iolist_size(Resp) >= 1024 andalso supported_gzip(AcceptEncoding) of
@@ -72,7 +57,7 @@ init(Req0, Opts = [Mod|_]) ->
             }, ErrResp, Req1),
             {ok, Req2, Opts};
         _:Error:Stack ->
-            lager:warning("[http_handler] get error: ~p, stack: ~p", [Error, Stack]),
+            logger:warning("[http_handler] get error: ~p, stack: ~p", [Error, Stack]),
             Req2 = cowboy_req:reply(500, #{
                 <<"Content-Type">> => <<"text/html;charset=utf-8">>
             }, <<"Internal Server Error">>, Req1),
@@ -87,17 +72,18 @@ parse_body(Req0) ->
     ContentType = cowboy_req:header(<<"content-type">>, Req0),
     case ContentType of
         <<"application/json", _/binary>> ->
-            {ok, Body, Req1} = read_body(Req0),
-            case Body =/= <<"">> of
-                 true ->
-                    {ok, catch jiffy:decode(Body, [return_maps]), Req1};
-                false ->
-                    {ok, #{}, Req1}
+            case read_body(Req0) of
+                {ok, Body, Req1} ->
+                    decode_json_body(Body, Req1);
+                {error, payload_too_large, Req1} ->
+                    {error, 413, iot_util:json_error(413, <<"payload too large">>), Req1}
             end;
         <<"application/x-www-form-urlencoded">> ->
-            {ok, PostParams0, Req1} = cowboy_req:read_urlencoded_body(Req0),
-            PostParams = maps:from_list(PostParams0),
-            {ok, PostParams, Req1};
+            case cowboy_req:read_urlencoded_body(Req0) of
+                {ok, PostParams0, Req1} ->
+                    PostParams = maps:from_list(PostParams0),
+                    {ok, PostParams, Req1}
+            end;
         _ ->
             {ok, #{}, Req0}
     end.
@@ -108,7 +94,33 @@ read_body(Req) ->
 read_body(Req, AccData) ->
     case cowboy_req:read_body(Req) of
         {ok, Data, Req1} ->
-            {ok, <<AccData/binary, Data/binary>>, Req1};
+            NAccData = <<AccData/binary, Data/binary>>,
+            case byte_size(NAccData) =< ?MAX_BODY_BYTES of
+                true ->
+                    {ok, NAccData, Req1};
+                false ->
+                    {error, payload_too_large, Req1}
+            end;
         {more, Data, Req1} ->
-            read_body(Req1, <<AccData/binary, Data/binary>>)
+            NAccData = <<AccData/binary, Data/binary>>,
+            case byte_size(NAccData) =< ?MAX_BODY_BYTES of
+                true ->
+                    read_body(Req1, NAccData);
+                false ->
+                    {error, payload_too_large, Req1}
+            end
+    end.
+
+decode_json_body(<<"">>, Req) ->
+    {ok, #{}, Req};
+decode_json_body(Body, Req) ->
+    case catch json:decode(Body) of
+        {'EXIT', _} ->
+            {error, 400, iot_util:json_error(400, <<"invalid json">>), Req};
+        {error, _} ->
+            {error, 400, iot_util:json_error(400, <<"invalid json">>), Req};
+        Decoded when is_map(Decoded); is_list(Decoded) ->
+            {ok, Decoded, Req};
+        _ ->
+            {error, 400, iot_util:json_error(400, <<"invalid json">>), Req}
     end.

apps/iot/src/http_handlers/device_handler.erl

@@ -1,68 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author licheng5
-%%% @copyright (C) 2020, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 26. 4月 2020 3:36 下午
-%%%-------------------------------------------------------------------
--module(device_handler).
--author("licheng5").
--include("iot.hrl").
-
-%% API
--export([handle_request/4]).
-
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-%% helper methods
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-
-%% 重新加载对应的主机信息
-handle_request("POST", "/device/reload", _, #{<<"host_id">> := HostId, <<"device_uuid">> := DeviceUUID}) when is_integer(HostId), is_binary(DeviceUUID) ->
-    lager:debug("[device_handler] host_id: ~p, will reload device uuid: ~p", [HostId, DeviceUUID]),
-    AliasName = iot_host:get_alias_name(HostId),
-    case global:whereis_name(AliasName) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"reload device failed">>)};
-        HostPid when is_pid(HostPid) ->
-            case iot_host:reload_device(HostPid, DeviceUUID) of
-                ok ->
-                    {ok, 200, iot_util:json_data(<<"success">>)};
-                {error, Reason} ->
-                    lager:debug("[device_handler] reload device: ~p, get error: ~p", [DeviceUUID, Reason]),
-                    {ok, 200, iot_util:json_error(404, <<"reload device failed">>)}
-            end
-    end;
-
-%% 删除对应的主机信息
-handle_request("POST", "/device/delete", _, #{<<"host_id">> := HostId, <<"device_uuid">> := DeviceUUID}) when is_integer(HostId), is_binary(DeviceUUID) ->
-    AliasName = iot_host:get_alias_name(HostId),
-    case global:whereis_name(AliasName) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"delete device failed">>)};
-        HostPid when is_pid(HostPid) ->
-            ok = iot_host:delete_device(HostPid, DeviceUUID),
-            {ok, 200, iot_util:json_data(<<"success">>)}
-    end;
-
-%% 处理主机的授权的激活
-handle_request("POST", "/device/activate", _, #{<<"host_id">> := HostId, <<"device_uuid">> := DeviceUUID, <<"auth">> := Auth})
-    when is_integer(HostId), is_binary(DeviceUUID), is_boolean(Auth) ->
-
-    AliasName = iot_host:get_alias_name(HostId),
-    case global:whereis_name(AliasName) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"activate device failed">>)};
-        HostPid when is_pid(HostPid) ->
-            case iot_host:activate_device(HostPid, DeviceUUID, Auth) of
-                ok ->
-                    {ok, 200, iot_util:json_data(<<"success">>)};
-                {error, Reason} ->
-                    lager:debug("[device_handler] activate device: ~p, get error: ~p", [DeviceUUID, Reason]),
-                    {ok, 200, iot_util:json_error(404, <<"activate device failed">>)}
-            end
-    end;
-
-handle_request(_, Path, _, _) ->
-    Path1 = list_to_binary(Path),
-    {ok, 200, iot_util:json_error(-1, <<"url: ", Path1/binary, " not found">>)}.

apps/iot/src/http_handlers/endpoint_handler.erl

@@ -1,103 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author licheng5
-%%% @copyright (C) 2020, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 26. 4月 2020 3:36 下午
-%%%-------------------------------------------------------------------
--module(endpoint_handler).
--author("licheng5").
--include("endpoint.hrl").
-
-%% API
--export([handle_request/4]).
-
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-%% helper methods
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-
-%% 获取endpoint的运行状态
-handle_request("POST", "/endpoint/run_statuses", _, Ids) when is_list(Ids) ->
-    Statuses = lists:map(fun(Id) ->
-        case endpoint:get_pid(Id) of
-            undefined ->
-                0;
-            Pid when is_pid(Pid) ->
-                1
-        end
-                         end, Ids),
-    {ok, 200, iot_util:json_data(Statuses)};
-
-handle_request("POST", "/endpoint/start", _, #{<<"id">> := Id}) when is_integer(Id) ->
-    case endpoint_bo:get_endpoint(Id) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"endpoint not found">>)};
-        {ok, EndpointInfo} ->
-            case endpoint_bo:endpoint_record(EndpointInfo) of
-                {ok, Endpoint = #endpoint{name = Name}} ->
-                    case endpoint_sup:ensured_endpoint_started(Endpoint) of
-                        {ok, Pid} when is_pid(Pid) ->
-                            {ok, 200, iot_util:json_data(<<"success">>)};
-                        {error, Reason} ->
-                            lager:warning("[endpoint_handler] start endpoint: ~p, get error: ~p", [Name, Reason]),
-                            {ok, 200, iot_util:json_error(404, <<"start endpoint error">>)}
-                    end;
-                error ->
-                    {ok, 200, iot_util:json_error(404, <<"endpoint invalid">>)}
-            end
-    end;
-
-handle_request("POST", "/endpoint/stop", _, #{<<"id">> := Id}) when is_integer(Id) ->
-    case endpoint_bo:get_endpoint(Id) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"endpoint not found">>)};
-        {ok, _} ->
-            case endpoint_sup:delete_endpoint(Id) of
-                ok ->
-                    {ok, 200, iot_util:json_data(<<"success">>)};
-                {error, Reason} ->
-                    lager:warning("[endpoint_handler] stop endpoint id: ~p, get error: ~p", [Id, Reason]),
-                    {ok, 200, iot_util:json_error(404, <<"stop endpoint error">>)}
-            end
-    end;
-
-handle_request("POST", "/endpoint/restart", _, #{<<"id">> := Id}) when is_integer(Id) ->
-    case endpoint_bo:get_endpoint(Id) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"endpoint not found">>)};
-        {ok, EndpointInfo} ->
-            case endpoint_bo:endpoint_record(EndpointInfo) of
-                {ok, Endpoint = #endpoint{name = Name}} ->
-                    case endpoint:get_pid(Id) of
-                        undefined ->
-                            case endpoint_sup:ensured_endpoint_started(Endpoint) of
-                                {ok, Pid} when is_pid(Pid) ->
-                                    {ok, 200, iot_util:json_data(<<"success">>)};
-                                {error, Reason} ->
-                                    lager:warning("[endpoint_handler] start endpoint: ~p, get error: ~p", [Name, Reason]),
-                                    {ok, 200, iot_util:json_error(404, <<"restart endpoint error">>)}
-                            end;
-                        Pid ->
-                            case endpoint_sup:delete_endpoint(Id) of
-                                ok ->
-                                    case endpoint_sup:ensured_endpoint_started(Endpoint) of
-                                        {ok, Pid} when is_pid(Pid) ->
-                                            {ok, 200, iot_util:json_data(<<"success">>)};
-                                        {error, Reason} ->
-                                            lager:warning("[endpoint_handler] start endpoint: ~p, get error: ~p", [Name, Reason]),
-                                            {ok, 200, iot_util:json_error(404, <<"restart endpoint error">>)}
-                                    end;
-                                {error, Reason} ->
-                                    lager:warning("[endpoint_handler] start endpoint: ~p, get error: ~p", [Name, Reason]),
-                                    {ok, 200, iot_util:json_error(404, <<"stop endpoint error">>)}
-                            end
-                    end;
-                error ->
-                    {ok, 200, iot_util:json_error(404, <<"endpoint invalid">>)}
-            end
-    end;
-
-handle_request(_, Path, _, _) ->
-    Path1 = list_to_binary(Path),
-    {ok, 200, iot_util:json_error(-1, <<"url: ", Path1/binary, " not found">>)}.

apps/iot/src/http_handlers/service_handler.erl

@@ -1,146 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author licheng5
-%%% @copyright (C) 2020, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 26. 4月 2020 3:36 下午
-%%%-------------------------------------------------------------------
--module(service_handler).
--author("licheng5").
--include("iot.hrl").
-
-%% API
--export([handle_request/4]).
-
-%% 下发config.json, 微服务接受后，保存服务配置
-handle_request("POST", "/service/push_config", _,
-    #{<<"uuid">> := UUID, <<"service_id">> := ServiceId, <<"config_json">> := ConfigJson, <<"timeout">> := Timeout0})
-    when is_binary(UUID), is_binary(ServiceId), is_binary(ConfigJson), is_integer(Timeout0) ->
-
-    %% 检查ConfigJson是否是合法的json字符串
-    true = iot_util:is_json(ConfigJson),
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(-1, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            Timeout = Timeout0 * 1000,
-            case iot_host:async_service_config(Pid, ServiceId, ConfigJson, Timeout) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, Timeout) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(-1, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(-1, Reason)}
-            end
-    end;
-
-%% 部署微服务
-handle_request("POST", "/service/deploy", _, #{<<"uuid">> := UUID, <<"task_id">> := TaskId, <<"service_id">> := ServiceId, <<"tar_url">> := TarUrl})
-    when is_binary(UUID), is_integer(TaskId), is_binary(ServiceId), is_binary(TarUrl) ->
-
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            case iot_host:deploy_service(Pid, TaskId, ServiceId, TarUrl) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, 5000) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-%% 启动服务
-handle_request("POST", "/service/start", _, #{<<"uuid">> := UUID, <<"service_id">> := ServiceId}) when is_binary(UUID), is_binary(ServiceId) ->
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            case iot_host:start_service(Pid, ServiceId) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, 5000) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-%% 停止服务
-handle_request("POST", "/service/stop", _, #{<<"uuid">> := UUID, <<"service_id">> := ServiceId}) when is_binary(UUID), is_binary(ServiceId) ->
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            case iot_host:stop_service(Pid, ServiceId) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, 5000) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-%% 远程调用微服务, 返回值的格式为json
-handle_request("POST", "/service/invoke", _, #{<<"uuid">> := UUID, <<"service_id">> := ServiceId, <<"payload">> := Payload, <<"timeout">> := Timeout0})
-    when is_binary(UUID), is_binary(ServiceId), is_binary(Payload), is_integer(Timeout0) ->
-
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            Timeout = Timeout0 * 1000,
-            case iot_host:invoke_service(Pid, ServiceId, Payload, Timeout) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, Timeout) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-handle_request("POST", "/service/task_log", _, #{<<"uuid">> := UUID, <<"task_id">> := TaskId}) when is_binary(UUID), is_integer(TaskId) ->
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            case iot_host:task_log(Pid, TaskId) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, 5000) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-handle_request(_, Path, _, _) ->
-    Path1 = list_to_binary(Path),
-    {ok, 200, iot_util:json_error(-1, <<"url: ", Path1/binary, " not found">>)}.
-
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-%% helper methods
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

apps/iot/src/iot.app.src

@@ -6,20 +6,16 @@
   {applications,
    [
     sync,
+    endpoint,
     emqtt,
     eredis,
     ranch,
     cowboy,
-    lager,
-    jiffy,
     brod,
-    parse_trans,
     hackney,
     poolboy,
-    mysql,
     gproc,
     % gpb,
-    esockd,
     mnesia,
     crypto,
     public_key,
@@ -27,6 +23,7 @@
     erts,
     runtime_tools,
     observer,
+    inets,
     kernel,
     stdlib
    ]},

apps/iot/src/iot_app.erl

@@ -9,47 +9,100 @@
 -export([start/2, stop/1]).
 
 start(_StartType, _StartArgs) ->
+    ok = iot_log:set_metadata(),
     io:setopts([{encoding, unicode}]),
     %% 加速内存的回收
     erlang:system_flag(fullsweep_after, 16),
-    %% 启动mnesia数据库
-    start_mnesia(),
 
-    %% 启动http服务
-    http_server:start(),
+    iot_mnesia:safe_start(),
 
-    %% 启动tcp服务
-    tcp_server:start(),
+    %% 启动http服务。当前 supervisor 初始化依赖本机 simulator API，因此保留此顺序。
+    start_http_server(),
+
+    %% 启动endpoints的相关依赖
+    start_endpoints(),
+
+    %% 启动ssl服务
+    start_ssl_server(),
 
     iot_sup:start_link().
 
 stop(_State) ->
+    stop_started_services(),
     ok.
 
 %% internal functions
 
-%% 启动内存数据库
-start_mnesia() ->
-    ok = ensure_mnesia_schema(),
-    %% 启动数据库
-    ok = mnesia:start(),
-    Tables = mnesia:system_info(tables),
-    lager:debug("[iot_app] tables: ~p", [Tables]),
-    %% 创建数据库表
-    ok.
+start_endpoints() ->
+    EndpointInfos = iot_api_client:get_all_endpoints(),
+    Endpoints = lists:filtermap(fun(EndpointInfo) ->
+        case endpoint:endpoint_record(EndpointInfo) of
+            error ->
+                false;
+            {ok, Endpoint} ->
+                {true, Endpoint}
+        end
+                                end, EndpointInfos),
+    ok = endpoint_adapter_sup:start_endpoints(Endpoints).
 
--spec ensure_mnesia_schema() -> any().
-ensure_mnesia_schema() ->
-    case mnesia:system_info(use_dir) of
-        true ->
-            ok;
-        false ->
-            mnesia:stop(),
-            case mnesia:create_schema([node()]) of
-                ok -> ok;
-                {error, {_, {already_exists, _}}} -> ok;
-                Error ->
-                    lager:debug("[iot_app] create mnesia schema failed with error: ~p", [Error]),
-                    throw({init_schema, Error})
-            end
-    end.
+start_http_server() ->
+    {ok, Props} = application:get_env(iot, http_server),
+    Acceptors = proplists:get_value(acceptors, Props, 50),
+    MaxConnections = proplists:get_value(max_connections, Props, 10240),
+    Backlog = proplists:get_value(backlog, Props, 1024),
+    Port = proplists:get_value(port, Props),
+
+    Dispatcher = cowboy_router:compile([
+        {'_', [
+            {"/host/[...]", http_protocol, [host_handler]},
+            {"/container/[...]", http_protocol, [container_handler]},
+            {"/endpoint/[...]", http_protocol, [endpoint_handler]},
+            {"/simulator/[...]", http_protocol, [simulator_api_handler]},
+            {"/event_stream", event_stream_handler, []}
+        ]}
+    ]),
+
+    TransOpts = #{
+        max_connections => MaxConnections,
+        num_acceptors => Acceptors,
+        shutdown => brutal_kill,
+        socket_opts => [
+            {backlog, Backlog},
+            {port, Port}
+        ]
+    },
+    {ok, Pid} = cowboy:start_clear(http_listener, TransOpts, #{env => #{dispatch => Dispatcher}}),
+
+    logger:debug("[http_server] the http server start at: ~p, pid is: ~p", [Port, Pid]).
+
+%% 启动ssl服务
+start_ssl_server() ->
+    {ok, Props} = application:get_env(iot, ssl_server),
+    Acceptors = proplists:get_value(acceptors, Props, 50),
+    MaxConnections = proplists:get_value(max_connections, Props, 10240),
+    Backlog = proplists:get_value(backlog, Props, 1024),
+    Port = proplists:get_value(port, Props),
+    PrivDir = code:priv_dir(iot),
+    CertFile = filename:join([PrivDir, "ssl", "server.crt"]),
+    KeyFile = filename:join([PrivDir, "ssl", "server.key"]),
+
+    TransOpts = #{
+        max_connections => MaxConnections,
+        num_acceptors => Acceptors,
+        shutdown => brutal_kill,
+        socket_opts => [
+            {nodelay, true},
+            {backlog, Backlog},
+            {port, Port},
+            {certfile, CertFile},
+            {keyfile, KeyFile}
+        ]
+    },
+    {ok, _} = ranch:start_listener(ssl_server, ranch_ssl, TransOpts, ssl_channel, []),
+    logger:debug("[iot_app] the ssl server start at: ~p", [Port]).
+
+stop_started_services() ->
+    _ = cowboy:stop_listener(http_listener),
+    _ = ranch:stop_listener(ssl_server),
+    _ = iot_mnesia:stop(),
+    ok.

apps/iot/src/iot_auth.erl

@@ -1,33 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%% 授权校验模块
-%%% @end
-%%% Created : 27. 6月 2023 09:48
-%%%-------------------------------------------------------------------
--module(iot_auth).
--author("aresei").
-
-%% API
--export([check/5]).
-
-%% 检测token是否是合法值
--spec check(Username :: binary(), Token :: binary(), UUID :: binary(), Salt :: binary(), Timestamp :: integer()) -> boolean().
-check(Username, Token, UUID, Salt, Timestamp) when is_binary(Username), is_binary(Token), is_binary(UUID), is_binary(Salt), is_integer(Timestamp) ->
-    true;
-check(Username, Token, UUID, Salt, Timestamp) when is_binary(Username), is_binary(Token), is_binary(UUID), is_binary(Salt), is_integer(Timestamp) ->
-    BinTimestamp = integer_to_binary(Timestamp),
-    %% 1分钟内有效
-    case iot_util:current_time() - Timestamp =< 60 of
-        true ->
-            {ok, PreTokens} = application:get_env(iot, pre_tokens),
-            case proplists:get_value(Username, PreTokens) of
-                undefined ->
-                    false;
-                PreToken when is_binary(PreToken) ->
-                    iot_util:md5(<<Salt/binary, "!", PreToken/binary, "!", UUID/binary, "!", BinTimestamp/binary>>) =:= Token
-            end;
-        false ->
-            false
-    end.

apps/iot/src/iot_device.erl

@@ -1,126 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 14. 8月 2023 11:40
-%%%-------------------------------------------------------------------
--module(iot_device).
--author("aresei").
--include("iot.hrl").
-
-%% API
--export([new/1, is_activated/1, change_status/2, reload/1, auth/2]).
-
-%% 终端是否授权
--define(DEVICE_AUTH_DENIED, 0).
--define(DEVICE_AUTH_AUTHED, 1).
-
-%% 状态
--define(STATE_DENIED, denied).
--define(STATE_ACTIVATED, activated).
-
--record(device, {
-    device_uuid :: binary(),
-    auth_state  = ?STATE_DENIED,
-    status = ?DEVICE_OFFLINE
-}).
-
-%%%===================================================================
-%%% API
-%%%===================================================================
-
--spec new(DeviceInfo :: binary() | map()) -> error | {ok, Device :: #device{}}.
-new(DeviceUUID) when is_binary(DeviceUUID) ->
-    case device_bo:get_device_by_uuid(DeviceUUID) of
-        {ok, #{<<"device_uuid">> := DeviceUUID, <<"authorize_status">> := AuthorizeStatus, <<"status">> := Status}} ->
-            {ok, #device{device_uuid = DeviceUUID, status = Status, auth_state = auth_state(AuthorizeStatus)}};
-        undefined ->
-            lager:warning("[iot_device] device uuid: ~p, loaded from mysql failed", [DeviceUUID]),
-            error
-    end;
-new(#{<<"device_uuid">> := DeviceUUID, <<"authorize_status">> := AuthorizeStatus, <<"status">> := Status}) ->
-    {ok, #device{device_uuid = DeviceUUID, status = Status, auth_state = auth_state(AuthorizeStatus)}}.
-
--spec is_activated(Device :: #device{}) -> boolean().
-is_activated(#device{auth_state = AuthState}) ->
-    AuthState =:= ?STATE_ACTIVATED.
-
--spec change_status(Device :: #device{}, NewStatus :: integer()) -> NDevice :: #device{}.
-change_status(Device = #device{status = Status}, NewStatus) when is_integer(NewStatus), Status =:= NewStatus ->
-    Device;
-change_status(Device = #device{device_uuid = DeviceUUID}, ?DEVICE_ONLINE) ->
-    {ok, _} = device_bo:change_status(DeviceUUID, ?DEVICE_ONLINE),
-    report_event(DeviceUUID, ?DEVICE_ONLINE),
-    Device#device{status = ?DEVICE_ONLINE};
-change_status(Device = #device{device_uuid = DeviceUUID}, ?DEVICE_OFFLINE) ->
-    {ok, #{<<"status">> := Status}} = device_bo:get_device_by_uuid(DeviceUUID),
-    case Status of
-        ?DEVICE_NOT_JOINED ->
-            lager:debug("[iot_device] device: ~p, device_maybe_offline, not joined, can not change to offline", [DeviceUUID]),
-            Device#device{status = ?DEVICE_NOT_JOINED};
-        ?DEVICE_OFFLINE ->
-            lager:debug("[iot_device] device: ~p, device_maybe_offline, is offline, do nothing", [DeviceUUID]),
-            Device#device{status = ?DEVICE_OFFLINE};
-        ?DEVICE_ONLINE ->
-            {ok, _} = device_bo:change_status(DeviceUUID, ?DEVICE_OFFLINE),
-            report_event(DeviceUUID, ?DEVICE_OFFLINE),
-            Device#device{status = ?DEVICE_OFFLINE}
-    end.
-
--spec reload(Device :: #device{}) -> error | {ok, NDevice :: #device{}}.
-reload(Device = #device{device_uuid = DeviceUUID}) ->
-    lager:debug("[iot_device] will reload: ~p", [DeviceUUID]),
-    case device_bo:get_device_by_uuid(DeviceUUID) of
-        {ok, #{<<"authorize_status">> := AuthorizeStatus, <<"status">> := Status}} ->
-            {ok, Device#device{device_uuid = DeviceUUID, status = Status, auth_state = auth_state(AuthorizeStatus)}};
-        undefined ->
-            lager:warning("[iot_device] device uuid: ~p, loaded from mysql failed", [DeviceUUID]),
-            error
-    end.
-
--spec auth(Device :: #device{}, Auth :: boolean()) -> NDevice :: #device{}.
-auth(Device = #device{auth_state = StateName, device_uuid = DeviceUUID}, Auth) when is_boolean(Auth) ->
-    case {StateName, Auth} of
-        {?STATE_DENIED, false} ->
-            lager:debug("[iot_device] device_uuid: ~p, auth: false, will keep state_name: ~p", [DeviceUUID, ?STATE_DENIED]),
-            Device;
-        {?STATE_DENIED, true} ->
-            Device#device{auth_state = ?STATE_ACTIVATED};
-        {?STATE_ACTIVATED, false} ->
-            lager:debug("[iot_device] device_uuid: ~p, auth: false, state_name from: ~p, to: ~p", [DeviceUUID, ?STATE_ACTIVATED, ?STATE_DENIED]),
-            Device#device{auth_state = ?STATE_DENIED};
-        {?STATE_ACTIVATED, true} ->
-            lager:debug("[iot_device] device_uuid: ~p, auth: true, will keep state_name: ~p", [DeviceUUID, ?STATE_ACTIVATED]),
-            Device
-    end.
-
-%%%===================================================================
-%%% Internal functions
-%%%===================================================================
-
--spec auth_state(integer()) -> atom().
-auth_state(?DEVICE_AUTH_AUTHED) ->
-    ?STATE_ACTIVATED;
-auth_state(?DEVICE_AUTH_DENIED) ->
-    ?STATE_DENIED.
-
--spec report_event(DeviceUUID :: binary(), NewStatus :: integer()) -> no_return().
-report_event(DeviceUUID, NewStatus) when is_binary(DeviceUUID), is_integer(NewStatus) ->
-    TextMap = #{
-        0 => <<"离线"/utf8>>,
-        1 => <<"在线"/utf8>>
-    },
-    %% 设备的状态信息上报给中电
-    Timestamp = iot_util:timestamp_of_seconds(),
-    FieldsList = [#{
-        <<"key">> => <<"device_status">>,
-        <<"value">> => NewStatus,
-        <<"value_text">> => maps:get(NewStatus, TextMap),
-        <<"unit">> => 0,
-        <<"type">> => <<"DI">>,
-        <<"name">> => <<"设备状态"/utf8>>,
-        <<"timestamp">> => Timestamp
-    }],
-    iot_router:route_uuid(DeviceUUID, FieldsList, Timestamp),
-    lager:debug("[iot_device] device_uuid: ~p, route fields: ~p", [DeviceUUID, FieldsList]).

apps/iot/src/iot_host.erl

@@ -1,482 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 22. 9月 2023 16:38
-%%%-------------------------------------------------------------------
--module(iot_host).
--author("aresei").
--include("iot.hrl").
--include("message_pb.hrl").
-
--behaviour(gen_statem).
-
-%% 心跳包检测时间间隔, 15分钟检测一次
--define(HEARTBEAT_INTERVAL, 900 * 1000).
-
-%% 状态
--define(STATE_DENIED, denied).
--define(STATE_ACTIVATED, activated).
-
-%% API
--export([start_link/2, get_name/1, get_alias_name/1, get_pid/1, handle/2, activate/2]).
--export([get_metric/1, get_status/1]).
-%% 通讯相关
--export([pub/3, attach_channel/2, command/3]).
--export([deploy_service/4, start_service/2, stop_service/2, invoke_service/4, async_service_config/4, task_log/2, await_reply/2]).
-%% 设备管理
--export([reload_device/2, delete_device/2, activate_device/3]).
--export([heartbeat/1]).
-
-%% gen_statem callbacks
--export([init/1, handle_event/4, terminate/3, code_change/4, callback_mode/0]).
-
--record(state, {
-    host_id :: integer(),
-    %% 从数据库里面读取到的数据
-    uuid :: binary(),
-    has_session = false :: boolean(),
-    %% 心跳计数器
-    heartbeat_counter = 0 :: integer(),
-    %% websocket相关
-    channel_pid :: undefined | pid(),
-
-    %% 设备的关系, #{device_uuid => Device}
-    device_map = #{},
-
-    %% 主机的相关信息
-    metrics = #{} :: map()
-}).
-
-%%%===================================================================
-%%% API
-%%%===================================================================
--spec get_pid(UUID :: binary()) -> undefined | pid().
-get_pid(UUID) when is_binary(UUID) ->
-    Name = get_name(UUID),
-    whereis(Name).
-
--spec get_name(UUID :: binary()) -> atom().
-get_name(UUID) when is_binary(UUID) ->
-    binary_to_atom(<<"iot_host:", UUID/binary>>).
-
--spec get_alias_name(HostId :: integer()) -> atom().
-get_alias_name(HostId0) when is_integer(HostId0) ->
-    HostId = integer_to_binary(HostId0),
-    binary_to_atom(<<"iot_host_id:", HostId/binary>>).
-
-%% 处理消息
--spec handle(Pid :: pid(), Packet :: {atom(), binary()} | {atom(), {binary(), binary()}}) -> no_return().
-handle(Pid, Packet) when is_pid(Pid) ->
-    gen_statem:cast(Pid, {handle, Packet}).
-
--spec get_status(Pid :: pid()) -> {ok, Status :: map()}.
-get_status(Pid) when is_pid(Pid) ->
-    gen_statem:call(Pid, get_status).
-
-%% 激活主机, true 表示激活； false表示关闭激活
--spec activate(Pid :: pid(), Auth :: boolean()) -> ok.
-activate(Pid, Auth) when is_pid(Pid), is_boolean(Auth) ->
-    gen_statem:call(Pid, {activate, Auth}).
-
--spec get_metric(Pid :: pid()) -> {ok, MetricInfo :: map()}.
-get_metric(Pid) when is_pid(Pid) ->
-    gen_statem:call(Pid, get_metric).
-
--spec attach_channel(pid(), pid()) -> ok | {error, Reason :: binary()} | {denied, Reason :: binary()}.
-attach_channel(Pid, ChannelPid) when is_pid(Pid), is_pid(ChannelPid) ->
-    gen_statem:call(Pid, {attach_channel, ChannelPid}).
-
--spec async_service_config(Pid :: pid(), ServiceId :: binary(), ConfigJson :: binary(), Timeout :: integer()) -> {ok, Ref :: reference()} | {error, Reason :: any()}.
-async_service_config(Pid, ServiceId, ConfigJson, Timeout) when is_pid(Pid), is_binary(ServiceId), is_binary(ConfigJson), is_integer(Timeout) ->
-    ConfigBin = message_pb:encode_msg(#push_service_config{service_id = ServiceId, config_json = ConfigJson, timeout = Timeout}),
-    gen_statem:call(Pid, {async_call, self(), ?PUSH_SERVICE_CONFIG, ConfigBin}).
-
--spec deploy_service(Pid :: pid(), TaskId :: integer(), ServiceId :: binary(), TarUrl :: binary()) -> {ok, Ref :: reference()} | {error, Reason :: any()}.
-deploy_service(Pid, TaskId, ServiceId, TarUrl) when is_pid(Pid), is_integer(TaskId), is_binary(ServiceId), is_binary(TarUrl) ->
-    PushBin = message_pb:encode_msg(#deploy{task_id = TaskId, service_id = ServiceId, tar_url = TarUrl}),
-    gen_statem:call(Pid, {async_call, self(), ?PUSH_DEPLOY, PushBin}).
-
--spec start_service(Pid :: pid(), ServiceId :: binary()) -> {ok, Ref :: reference()} | {error, Reason :: any()}.
-start_service(Pid, ServiceId) when is_pid(Pid), is_binary(ServiceId) ->
-    gen_statem:call(Pid, {async_call, self(), ?PUSH_START_SERVICE, ServiceId}).
-
--spec stop_service(Pid :: pid(), ServiceId :: binary()) -> {ok, Ref :: reference()} | {error, Reason :: any()}.
-stop_service(Pid, ServiceId) when is_pid(Pid), is_binary(ServiceId) ->
-    gen_statem:call(Pid, {async_call, self(), ?PUSH_STOP_SERVICE, ServiceId}).
-
--spec invoke_service(Pid :: pid(), ServiceId :: binary(), Payload :: binary(), Timeout :: integer()) -> {ok, Ref :: reference()} | {error, Reason :: any()}.
-invoke_service(Pid, ServiceId, Payload, Timeout) when is_pid(Pid), is_binary(ServiceId), is_binary(Payload), is_integer(Timeout) ->
-    InvokeBin = message_pb:encode_msg(#invoke{service_id = ServiceId, payload = Payload, timeout = Timeout}),
-    gen_statem:call(Pid, {async_call, self(), ?PUSH_INVOKE, InvokeBin}).
-
--spec task_log(Pid :: pid(), TaskId :: integer()) -> {ok, Ref :: reference()} | {error, Reason :: any()}.
-task_log(Pid, TaskId) when is_pid(Pid), is_integer(TaskId) ->
-    TaskLogBin = message_pb:encode_msg(#fetch_task_log{task_id = TaskId}),
-    gen_statem:call(Pid, {async_call, self(), ?PUSH_TASK_LOG, TaskLogBin}).
-
--spec await_reply(Ref :: reference(), Timeout :: integer()) -> {ok, Result :: binary()} | {error, Reason :: binary()}.
-await_reply(Ref, Timeout) when is_reference(Ref), is_integer(Timeout) ->
-    receive
-        {async_call_reply, Ref, #async_call_reply{code = 1, result = Result}} ->
-            {ok, Result};
-        {async_call_reply, Ref, #async_call_reply{code = 0, message = Message}} ->
-            {error, Message}
-    after Timeout ->
-        {error, <<"timeout">>}
-    end.
-
--spec pub(Pid :: pid(), Topic :: binary(),  Content :: binary()) -> ok | {error, Reason :: any()}.
-pub(Pid, Topic, Content) when is_pid(Pid), is_binary(Topic), is_binary(Content) ->
-    gen_statem:call(Pid, {pub, Topic, Content}).
-
--spec command(Pid :: pid(), CommandType :: integer(),  Command :: binary()) -> ok | {error, Reason :: any()}.
-command(Pid, CommandType, Command) when is_pid(Pid), is_integer(CommandType), is_binary(Command) ->
-    gen_statem:call(Pid, {command, CommandType, Command}).
-
-%% 设备管理相关
--spec reload_device(Pid :: pid(), DeviceUUID :: binary()) -> ok | {error, Reason :: any()}.
-reload_device(Pid, DeviceUUID) when is_pid(Pid), is_binary(DeviceUUID) ->
-    gen_statem:call(Pid, {reload_device, DeviceUUID}).
-
--spec delete_device(Pid :: pid(), DeviceUUID :: binary()) -> ok.
-delete_device(Pid, DeviceUUID) when is_pid(Pid), is_binary(DeviceUUID) ->
-    gen_statem:call(Pid, {delete_device, DeviceUUID}).
-
--spec activate_device(Pid :: pid(), DeviceUUID :: binary(), Auth :: boolean()) -> ok | {error, Reason :: any()}.
-activate_device(Pid, DeviceUUID, Auth) when is_pid(Pid), is_binary(DeviceUUID), is_boolean(Auth) ->
-    gen_statem:call(Pid, {activate_device, DeviceUUID, Auth}).
-
--spec heartbeat(Pid :: pid()) -> no_return().
-heartbeat(undefined) ->
-    ok;
-heartbeat(Pid) when is_pid(Pid) ->
-    gen_statem:cast(Pid, heartbeat).
-
-%% @doc Creates a gen_statem process which calls Module:init/1 to
-%% initialize. To ensure a synchronized start-up procedure, this
-%% function does not return until Module:init/1 has returned.
-start_link(Name, UUID) when is_atom(Name), is_binary(UUID) ->
-    gen_statem:start_link({local, Name}, ?MODULE, [UUID], []).
-
-%%%===================================================================
-%%% gen_statem callbacks
-%%%===================================================================
-
-%% @private
-%% @doc Whenever a gen_statem is started using gen_statem:start/[3,4] or
-%% gen_statem:start_link/[3,4], this function is called by the new
-%% process to initialize.
-init([UUID]) ->
-    case host_bo:get_host_by_uuid(UUID) of
-        {ok, #{<<"id">> := HostId, <<"authorize_status">> := AuthorizeStatus}} ->
-            %% 通过host_id注册别名, 可以避免通过查询数据库获取HostPid
-            AliasName = get_alias_name(HostId),
-            global:register_name(AliasName, self()),
-
-            %% 心跳检测机制
-            erlang:start_timer(?HEARTBEAT_INTERVAL, self(), heartbeat_ticker),
-
-            StateName = case AuthorizeStatus =:= 1 of
-                            true  -> ?STATE_ACTIVATED;
-                            false -> ?STATE_DENIED
-                        end,
-
-            %% 加载设备信息
-            {ok, DeviceInfos} = device_bo:get_host_devices(HostId),
-            Devices = lists:filtermap(fun(DeviceInfo = #{<<"device_uuid">> := DeviceUUID}) ->
-                case iot_device:new(DeviceInfo) of
-                    error ->
-                        false;
-                    {ok, Device} ->
-                        {true, {DeviceUUID, Device}}
-                end
-                                      end, DeviceInfos),
-
-            {ok, StateName, #state{host_id = HostId, uuid = UUID, device_map = maps:from_list(Devices), has_session = false}};
-        undefined ->
-            lager:warning("[iot_host] host uuid: ~p, loaded from mysql failed", [UUID]),
-            ignore
-    end.
-
-%% @private
-%% @doc This function is called by a gen_statem when it needs to find out
-%% the callback mode of the callback module.
-callback_mode() ->
-    handle_event_function.
-
-%% @private
-%% @doc If callback_mode is handle_event_function, then whenever a
-%% gen_statem receives an event from call/2, cast/2, or as a normal
-%% process message, this function is called.
-
-handle_event({call, From}, get_metric, _, State = #state{metrics = Metrics}) ->
-    {keep_state, State, [{reply, From, {ok, Metrics}}]};
-
-%% 获取主机的状态
-handle_event({call, From}, get_status, _, State = #state{channel_pid = ChannelPid, heartbeat_counter = HeartbeatCounter, metrics = Metrics, has_session = HasSession}) ->
-    HasChannel = (ChannelPid /= undefined),
-    Reply = #{
-        <<"has_channel">> => HasChannel,
-        <<"has_session">> => HasSession,
-        <<"heartbeat_counter">> => HeartbeatCounter,
-        <<"metrics">> => Metrics
-    },
-    {keep_state, State, [{reply, From, {ok, Reply}}]};
-
-%% 只要channel存在，就负责将消息推送到边缘端主机
-handle_event({call, From}, {async_call, ReceiverPid, PushType, PushBin}, _, State = #state{uuid = UUID, channel_pid = ChannelPid, has_session = HasSession}) ->
-    case HasSession andalso is_pid(ChannelPid) of
-        true ->
-            %% 通过websocket发送请求
-            Ref = tcp_channel:async_call(ChannelPid, ReceiverPid, PushType, PushBin),
-            {keep_state, State, [{reply, From, {ok, Ref}}]};
-        false ->
-            lager:debug("[iot_host] uuid: ~p, publish_type: ~p, invalid state: ~p", [UUID, PushType, state_map(State)]),
-            {keep_state, State, [{reply, From, {error, <<"主机离线，发送请求失败"/utf8>>}}]}
-    end;
-
-%% 发送指令时, pub/sub
-handle_event({call, From}, {pub, Topic, Content}, ?STATE_ACTIVATED, State = #state{uuid = UUID, channel_pid = ChannelPid, has_session = HasSession}) ->
-    case HasSession andalso is_pid(ChannelPid) of
-        true ->
-            lager:debug("[iot_host] host: ~p, publish to topic: ~p, content: ~p", [UUID, Topic, Content]),
-            %% 通过websocket发送请求
-            tcp_channel:pub(ChannelPid, Topic, Content),
-
-            {keep_state, State, [{reply, From, ok}]};
-        false ->
-            lager:debug("[iot_host] uuid: ~p, publish to topic: ~p, content: ~p, invalid state: ~p", [UUID, Topic, Content, state_map(State)]),
-            {keep_state, State, [{reply, From, {error, <<"主机离线，发送失败"/utf8>>}}]}
-    end;
-
-%% 发送指令时
-handle_event({call, From}, {command, CommandType, Command}, ?STATE_ACTIVATED, State = #state{uuid = UUID, channel_pid = ChannelPid, has_session = HasSession}) ->
-    case HasSession andalso is_pid(ChannelPid) of
-        true ->
-            lager:debug("[iot_host] host: ~p, command_type: ~p, command: ~p", [UUID, CommandType, Command]),
-            %% 通过websocket发送请求
-            tcp_channel:command(ChannelPid, CommandType, Command),
-            {keep_state, State, [{reply, From, ok}]};
-        false ->
-            lager:debug("[iot_host] host: ~p, command_type: ~p, command: ~p, invalid state: ~p", [UUID, CommandType, Command, state_map(State)]),
-            {keep_state, State, [{reply, From, {error, <<"主机离线，发送指令失败"/utf8>>}}]}
-    end;
-
-%% 激活主机
-handle_event({call, From}, {activate, true}, _, State = #state{uuid = UUID, channel_pid = ChannelPid}) ->
-    case is_pid(ChannelPid) of
-        true ->
-            lager:debug("[iot_host] uuid: ~p, activate: true", [UUID]),
-            tcp_channel:command(ChannelPid, ?COMMAND_AUTH, <<1:8>>);
-        false ->
-            lager:debug("[iot_host] uuid: ~p, activate: true, no channel", [UUID])
-    end,
-    {next_state, ?STATE_ACTIVATED, State, [{reply, From, ok}]};
-
-%% 关闭授权
-handle_event({call, From}, {activate, false}, _, State = #state{uuid = UUID, channel_pid = ChannelPid}) ->
-    case is_pid(ChannelPid) of
-        true ->
-            tcp_channel:command(ChannelPid, ?COMMAND_AUTH, <<0:8>>),
-            lager:debug("[iot_host] uuid: ~p, activate: false", [UUID]),
-            tcp_channel:stop(ChannelPid, closed);
-        false ->
-            lager:debug("[iot_host] uuid: ~p, activate: false, no channel", [UUID])
-    end,
-    {next_state, ?STATE_DENIED, State#state{channel_pid = undefined, has_session = false}, [{reply, From, ok}]};
-
-%% 绑定channel
-handle_event({call, From}, {attach_channel, ChannelPid}, StateName, State = #state{uuid = UUID, channel_pid = undefined}) ->
-    case StateName of
-        ?STATE_ACTIVATED ->
-            erlang:monitor(process, ChannelPid),
-            %% 更新主机为在线状态
-            {ok, AffectedRow} = host_bo:change_status(UUID, ?HOST_ONLINE),
-            report_event(UUID, ?HOST_ONLINE),
-            lager:debug("[iot_host] host_id(attach_channel) uuid: ~p, will change status, affected_row: ~p", [UUID, AffectedRow]),
-            {keep_state, State#state{channel_pid = ChannelPid, has_session = true}, [{reply, From, ok}]};
-        %% 主机未激活
-        ?STATE_DENIED ->
-            lager:notice("[iot_host] attach_channel host_id uuid: ~p, channel: ~p, host inactivated", [UUID, ChannelPid]),
-            erlang:monitor(process, ChannelPid),
-            {keep_state, State#state{channel_pid = ChannelPid}, [{reply, From, {denied, <<"host inactivated">>}}]}
-    end;
-%% 已经绑定了channel
-handle_event({call, From}, {attach_channel, _}, _, State = #state{uuid = UUID, channel_pid = OldChannelPid}) ->
-    lager:notice("[iot_host] attach_channel host_id uuid: ~p, old channel exists: ~p", [UUID, OldChannelPid]),
-    {keep_state, State, [{reply, From, {error, <<"channel existed">>}}]};
-
-%% 重新加载设备信息
-handle_event({call, From}, {reload_device, DeviceUUID}, _, State = #state{device_map = DeviceMap}) ->
-    case maps:find(DeviceUUID, DeviceMap) of
-        error ->
-            {keep_state, State, [{reply, From, {error, <<"device not found">>}}]};
-        {ok, Device} ->
-            case iot_device:reload(Device) of
-                error ->
-                    {keep_state, State#state{device_map = maps:remove(Device, DeviceMap)}, [{reply, From, {error, <<"reload device error">>}}]};
-                {ok, NDevice} ->
-                    {keep_state, State#state{device_map = maps:put(DeviceUUID, NDevice, DeviceMap)}, [{reply, From, ok}]}
-            end
-    end;
-
-%% 删除设备
-handle_event({call, From}, {delete_device, DeviceUUID}, _, State = #state{device_map = DeviceMap}) ->
-    {keep_state, State#state{device_map = maps:remove(DeviceUUID, DeviceMap)}, [{reply, From, ok}]};
-
-%% 激活设备
-handle_event({call, From}, {activate_device, DeviceUUID, Auth}, _, State = #state{device_map = DeviceMap}) ->
-    case maps:find(DeviceUUID, DeviceMap) of
-        error ->
-            {keep_state, State, [{reply, From, {error, <<"device not found">>}}]};
-        {ok, Device} ->
-            NDevice = iot_device:auth(Device, Auth),
-            {keep_state, State#state{device_map = maps:put(DeviceUUID, NDevice, DeviceMap)}, [{reply, From, ok}]}
-    end;
-
-%% todo
-handle_event(cast, {handle, {data, #data{service_id = ServiceId, device_uuid = DeviceUUID, route_key = RouteKey0, metric = Metric}}}, ?STATE_ACTIVATED, State = #state{uuid = UUID, has_session = true, device_map = DeviceMap}) ->
-    lager:debug("[iot_host] metric_data host: ~p, service_id: ~p, device_uuid: ~p, route_key: ~p, metric: ~p", [UUID, ServiceId, DeviceUUID, RouteKey0, Metric]),
-    case DeviceUUID =/= <<"">> of
-        true ->
-            case maps:find(DeviceUUID, DeviceMap) of
-                error ->
-                    lager:warning("[iot_host] host uuid: ~p, device uuid: ~p not found, metric: ~p", [UUID, DeviceUUID, Metric]),
-                    {keep_state, State};
-                {ok, Device} ->
-                    case iot_device:is_activated(Device) of
-                        true ->
-                            RouteKey = get_route_key(RouteKey0),
-                            case endpoint:get_alias_pid(RouteKey) of
-                                undefined ->
-                                    ok;
-                                EndpointPid ->
-                                    endpoint:forward(EndpointPid, ServiceId, Metric)
-                            end,
-                            NDevice = iot_device:change_status(Device, ?DEVICE_ONLINE),
-                            {keep_state, State#state{device_map = maps:put(DeviceUUID, NDevice, DeviceMap)}};
-                        false ->
-                            lager:warning("[iot_host] host uuid: ~p, device_uuid: ~p not activated, metric: ~p", [UUID, DeviceUUID, Metric]),
-                            {keep_state, State}
-                    end
-            end;
-        false ->
-            {keep_state, State}
-    end;
-
-%% ping的数据是通过aes加密后的，因此需要在有会话的情况下才行
-handle_event(cast, {handle, {ping, Metrics}}, ?STATE_ACTIVATED, State = #state{uuid = UUID, has_session = true}) ->
-    lager:debug("[iot_host] ping host_id uuid: ~p, get ping: ~p", [UUID, Metrics]),
-    {keep_state, State#state{metrics = Metrics}};
-
-handle_event(cast, {handle, {inform,  #service_inform{service_id = ServiceId, status = Status, timestamp = Timestamp}}}, ?STATE_ACTIVATED, State = #state{uuid = UUID, has_session = true}) ->
-    lager:debug("[iot_host] inform host: ~p, service_id: ~p, status: ~p, timestamp: ~p", [UUID, ServiceId, Status, Timestamp]),
-    {keep_state, State};
-
-handle_event(cast, {handle, {event, #event{service_id = ServiceId, event_type = EventType, params = Params}}}, ?STATE_ACTIVATED, State = #state{uuid = UUID, has_session = true}) ->
-    lager:debug("[iot_host] event uuid: ~p, service_id: ~p, event_type: ~p, params: ~p", [UUID, ServiceId, EventType, Params]),
-    %DevicePid = iot_device:get_pid(DeviceUUID),
-    %iot_device:change_status(DevicePid, Status),
-
-    {keep_state, State};
-
-%% 心跳机制
-handle_event(cast, heartbeat, _, State = #state{heartbeat_counter = HeartbeatCounter}) ->
-    {keep_state, State#state{heartbeat_counter = HeartbeatCounter + 1}};
-
-%% 没有收到心跳包，主机下线, 设备状态不变
-handle_event(info, {timeout, _, heartbeat_ticker}, _, State = #state{uuid = UUID, heartbeat_counter = 0, channel_pid = ChannelPid}) ->
-    lager:warning("[iot_host] uuid: ~p, heartbeat lost, devices will unknown", [UUID]),
-    {ok, #{<<"status">> := Status}} = host_bo:get_host_by_uuid(UUID),
-    case Status of
-        ?HOST_NOT_JOINED ->
-            lager:debug("[iot_host] host: ~p, host_maybe_offline, host not joined, can not change to offline", [UUID]);
-        ?HOST_OFFLINE ->
-            lager:debug("[iot_host] host: ~p, host_maybe_offline, host now is offline, do nothing", [UUID]);
-        ?HOST_ONLINE ->
-            {ok, _} = host_bo:change_status(UUID, ?HOST_OFFLINE),
-            report_event(UUID, ?HOST_OFFLINE)
-    end,
-
-    %% 关闭channel，主机需要重新连接，才能保存状态的一致
-    is_pid(ChannelPid) andalso tcp_channel:stop(ChannelPid, closed),
-    erlang:start_timer(?HEARTBEAT_INTERVAL, self(), heartbeat_ticker),
-
-    {keep_state, State#state{channel_pid = undefined, has_session = false, heartbeat_counter = 0}};
-
-%% 其他情况下需要重置系统计数器
-handle_event(info, {timeout, _, heartbeat_ticker}, _, State = #state{}) ->
-    erlang:start_timer(?HEARTBEAT_INTERVAL, self(), heartbeat_ticker),
-    {keep_state, State#state{heartbeat_counter = 0}};
-
-%% 当websocket断开的时候，主机的状态不一定改变；主机的状态改变通过心跳机制，会话状态需要改变
-handle_event(info, {'DOWN', _Ref, process, ChannelPid, Reason}, _, State = #state{uuid = UUID, channel_pid = ChannelPid, has_session = HasSession}) ->
-    lager:warning("[iot_host] uuid: ~p, channel: ~p, down with reason: ~p, has_session: ~p, state: ~p", [UUID, ChannelPid, Reason, HasSession, State]),
-    {keep_state, State#state{channel_pid = undefined, has_session = false}};
-
-handle_event(info, {'DOWN', _Ref, process, Pid, Reason}, _, State = #state{uuid = UUID}) ->
-    lager:debug("[iot_host] uuid: ~p, process_pid: ~p, down with reason: ~p, state: ~p", [UUID, Pid, Reason, State]),
-    {keep_state, State};
-
-handle_event(Event, Info, StateName, State = #state{uuid = UUID}) ->
-    lager:warning("[iot_host] host: ~p, event: ~p, unknown message: ~p, state_name: ~p, state: ~p", [UUID, Event, Info, StateName, state_map(State)]),
-    {keep_state, State}.
-
-%% @private
-%% @doc This function is called by a gen_statem when it is about to
-%% terminate. It should be the opposite of Module:init/1 and do any
-%% necessary cleaning up. When it returns, the gen_statem terminates with
-%% Reason. The return value is ignored.
-terminate(Reason, _StateName, _State = #state{uuid = UUID, has_session = HasSession}) ->
-    lager:debug("[iot_host] host: ~p, terminate with reason: ~p, has_session: ~p", [UUID, Reason, HasSession]),
-    ok.
-
-%% @private
-%% @doc Convert process state when code is changed
-code_change(_OldVsn, StateName, State = #state{}, _Extra) ->
-    {ok, StateName, State}.
-
-%%%===================================================================
-%%% Internal functions
-%%%===================================================================
-
-get_route_key(<<"">>) ->
-    <<"default">>;
-get_route_key(RouteKey) when is_binary(RouteKey) ->
-    RouteKey.
-
--spec report_event(UUID :: binary(), NewStatus :: integer()) -> no_return().
-report_event(UUID, NewStatus) when is_binary(UUID), is_integer(NewStatus) ->
-    TextMap = #{
-        0 => <<"离线"/utf8>>,
-        1 => <<"在线"/utf8>>
-    },
-
-    %% 设备的状态信息上报给中电
-    Timestamp = iot_util:timestamp_of_seconds(),
-    FieldsList = [#{
-        <<"key">> => <<"host_status">>,
-        <<"value">> => NewStatus,
-        <<"value_text">> => maps:get(NewStatus, TextMap),
-        <<"unit">> => 0,
-        <<"type">> => <<"DI">>,
-        <<"name">> => <<"主机状态"/utf8>>,
-        <<"timestamp">> => Timestamp
-    }],
-    %% todo 这里需要实现新的机制
-    % iot_router:route_uuid(UUID, FieldsList, Timestamp),
-    lager:debug("[iot_host] host_uuid: ~p, route fields: ~p", [UUID, FieldsList]).
-
-%% 将当前的state转换成map
-state_map(#state{host_id = HostId, uuid = UUID, has_session = HasSession, heartbeat_counter = HeartbeatCounter, channel_pid = ChannelPid, metrics = Metrics}) ->
-    #{
-        host_id => HostId,
-        uuid => UUID,
-        has_session => HasSession,
-        heartbeat_counter => HeartbeatCounter,
-        channel_pid => ChannelPid,
-        metrics => Metrics
-    }.

apps/iot/src/iot_host_sup.erl

@@ -1,50 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%% @end
-%%%-------------------------------------------------------------------
--module(iot_host_sup).
--include("iot.hrl").
-
--behaviour(supervisor).
-
--export([start_link/0, init/1, delete_host/1, ensured_host_started/1]).
-
-start_link() ->
-    supervisor:start_link({local, ?MODULE}, ?MODULE, []).
-
-init([]) ->
-    Specs = lists:map(fun child_spec/1, host_bo:get_all_hosts()),
-
-    {ok, {#{strategy => one_for_one, intensity => 1000, period => 3600}, Specs}}.
-
--spec ensured_host_started(UUID :: binary()) -> {ok, Pid :: pid()} | {error, Reason :: any()}.
-ensured_host_started(UUID) when is_binary(UUID) ->
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            case supervisor:start_child(?MODULE, child_spec(UUID)) of
-                {ok, Pid} when is_pid(Pid) ->
-                    {ok, Pid};
-                {error, {'already_started', Pid}} when is_pid(Pid) ->
-                    {ok, Pid};
-                {error, Error} ->
-                    {error, Error}
-            end;
-        Pid when is_pid(Pid) ->
-            {ok, Pid}
-    end.
-
-delete_host(UUID) when is_binary(UUID) ->
-    Id = iot_host:get_name(UUID),
-    ok = supervisor:terminate_child(?MODULE, Id),
-    supervisor:delete_child(?MODULE, Id).
-
-child_spec(UUID) ->
-    Id = iot_host:get_name(UUID),
-    #{id => Id,
-        start => {iot_host, start_link, [Id, UUID]},
-        restart => permanent,
-        shutdown => 2000,
-        type => worker,
-        modules => ['iot_host']}.

apps/iot/src/iot_http_client.erl

@@ -1,40 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author licheng5
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 03. 3月 2023 11:48
-%%%-------------------------------------------------------------------
--module(iot_http_client).
--author("licheng5").
-
-%% API
--export([post/2]).
-
-post(Url, Body) when is_list(Url), is_binary(Body) ->
-    case hackney:request(post, Url, [], Body) of
-        {ok, 200, _, ClientRef} ->
-            case hackney:body(ClientRef) of
-                {ok, RespBody} ->
-                    lager:debug("[iot_http_client] url: ~p, response is: ~p", [Url, RespBody]),
-                    ok;
-                {error, Reason} ->
-                    lager:warning("[iot_http_client] url: ~p, get error: ~p", [Url, Reason]),
-                    {error, Reason}
-            end;
-
-        {ok, HttpCode, _, ClientRef} ->
-            case hackney:body(ClientRef) of
-                {ok, RespBody} ->
-                    lager:debug("[iot_http_client] url: ~p, http_code: ~p, response is: ~p", [Url, HttpCode, RespBody]),
-                    ok;
-                {error, Reason} ->
-                    lager:warning("[iot_http_client] url: ~p, http_code: ~p, get error: ~p", [Url, HttpCode, Reason]),
-                    {error, Reason}
-            end;
-
-        {error, Reason} ->
-            lager:warning("[iot_http_client] url: ~p, get error: ~p", [Url, Reason]),
-            {error, Reason}
-    end.

apps/iot/src/iot_log.erl

@@ -0,0 +1,10 @@
+%%%-------------------------------------------------------------------
+%%% @doc Logger helper for project-scoped metadata.
+%%%-------------------------------------------------------------------
+-module(iot_log).
+
+-export([set_metadata/0]).
+
+-spec set_metadata() -> ok.
+set_metadata() ->
+    logger:set_process_metadata(#{domain => [iot]}).

apps/iot/src/iot_router.erl

@@ -1,26 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 04. 7月 2023 11:30
-%%%-------------------------------------------------------------------
--module(iot_router).
--author("aresei").
--include("iot.hrl").
-
-%% API
--export([route_uuid/3]).
-
--spec route_uuid(RouterUUID :: binary(), Fields :: list(), Timestamp :: integer()) -> no_return().
-route_uuid(RouterUUID, Fields, Timestamp) when is_binary(RouterUUID), is_list(Fields), is_integer(Timestamp) ->
-    %% 查找终端设备对应的点位信息
-    case redis_client:hget(RouterUUID, <<"location_code">>) of
-        {ok, undefined} ->
-            lager:warning("[iot_host] the north_data hget location_code, uuid: ~p, not found, fields: ~p", [RouterUUID, Fields]);
-        {ok, LocationCode} when is_binary(LocationCode) ->
-            iot_zd_endpoint:forward(LocationCode, Fields, Timestamp);
-        {error, Reason} ->
-            lager:warning("[iot_host] the north_data hget location_code uuid: ~p, get error: ~p, fields: ~p", [RouterUUID, Reason, Fields])
-    end.

apps/iot/src/iot_sup.erl

@@ -29,21 +29,30 @@ init([]) ->
 
     Specs = [
         #{
-            id => 'iot_name_server',
-            start => {'iot_name_server', start_link, []},
-            restart => permanent,
-            shutdown => 2000,
-            type => worker,
-            modules => ['iot_name_server']
-        },
-
-        #{
-            id => 'endpoint_sup',
-            start => {'endpoint_sup', start_link, []},
+            id => 'iot_container_task_sup',
+            start => {'iot_container_task_sup', start_link, []},
             restart => permanent,
             shutdown => 2000,
             type => supervisor,
-            modules => ['endpoint_sup']
+            modules => ['iot_container_task_sup']
+        },
+
+        #{
+            id => 'udp_server',
+            start => {'udp_server', start_link, []},
+            restart => permanent,
+            shutdown => 2000,
+            type => worker,
+            modules => ['udp_server']
+        },
+
+        #{
+            id => 'ctrl_server',
+            start => {'ctrl_server', start_link, []},
+            restart => permanent,
+            shutdown => 2000,
+            type => worker,
+            modules => ['ctrl_server']
         },
 
         #{
@@ -61,7 +70,11 @@ init([]) ->
 %% internal functions
 
 pools() ->
-    {ok, Pools} = application:get_env(iot, pools),
-    lists:map(fun({Name, PoolArgs, WorkerArgs}) ->
-        poolboy:child_spec(Name, [{name, {local, Name}}|PoolArgs], WorkerArgs)
-              end, Pools).
+    case application:get_env(iot, pools) of
+        undefined ->
+            [];
+        {ok, Pools} ->
+            lists:map(fun({Name, PoolArgs, WorkerArgs}) ->
+                poolboy:child_spec(Name, [{name, {local, Name}}|PoolArgs], WorkerArgs)
+                      end, Pools)
+    end.

apps/iot/src/iot_udp_handler.erl

@@ -1,26 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 31. 8月 2023 13:48
-%%%-------------------------------------------------------------------
--module(iot_udp_handler).
--author("aresei").
-
-%% API
--export([start_link/2, loop/2]).
-
-start_link(Transport, Peer) ->
-    {ok, spawn_link(?MODULE, loop, [Transport, Peer])}.
-
-loop(Transport = {udp, Server, _Sock}, Peer) ->
-    receive
-        {datagram, Server, <<Len:16, HostUUID:Len/binary>>} ->
-            Pid = iot_host:get_pid(HostUUID),
-            iot_host:heartbeat(Pid),
-            loop(Transport, Peer);
-        {datagram, Server, _} ->
-            exit(normal)
-    end.

apps/iot/src/iot_util.erl

@@ -10,14 +10,15 @@
 -author("licheng5").
 
 %% API
--export([timestamp/0, number_format/2, current_time/0, timestamp_of_seconds/0, float_to_binary/2, int_format/2, file_uri/1]).
--export([step/3, chunks/2, rand_bytes/1, uuid/0, md5/1, parse_mapper/1]).
+-export([timestamp/0, number_format/2, current_time/0, timestamp_of_seconds/0, float_to_binary/2, int_format/2]).
+-export([step/3, rand_bytes/1, uuid/0, md5/1, parse_mapper/1]).
 -export([json_data/1, json_error/2, is_json/1]).
 -export([queue_limited_in/3, assert_call/2, assert/2]).
+-export([sha256/1]).
 
 -spec is_json(Json :: term()) -> boolean().
 is_json(Json) when is_binary(Json) ->
-    case catch jiffy:decode(Json, [return_maps]) of
+    case catch json:decode(Json) of
         Result when is_list(Result) orelse is_map(Result) ->
             true;
         _ ->
@@ -60,33 +61,18 @@ step(Start, End, Step, Acc) when Start < End ->
 step(_, _, _, Acc) ->
     lists:reverse(Acc).
 
-%% 将数据分组
--spec chunks(list(), integer()) -> [list()].
-chunks(List, Size) when is_list(List), is_integer(Size), Size > 0, length(List) =< Size ->
-    [List];
-chunks(List, Size) when is_list(List), is_integer(Size), Size > 0 ->
-    chunks0(List, Size, Size, [], []).
-chunks0([], _, _, [], AccTarget) ->
-    lists:reverse(AccTarget);
-chunks0([], _, _, Target, AccTarget) ->
-    lists:reverse([lists:reverse(Target) | AccTarget]);
-chunks0(List, Size, 0, Target, AccTarget) ->
-    chunks0(List, Size, Size, [], [lists:reverse(Target) | AccTarget]);
-chunks0([Hd | Tail], Size, Num, Target, AccTarget) ->
-    chunks0(Tail, Size, Num - 1, [Hd | Target], AccTarget).
-
 json_data(Data) ->
-    jiffy:encode(#{
+    iolist_to_binary(json:encode(#{
         <<"result">> => Data
-    }, [force_utf8]).
+    })).
 
-json_error(ErrCode, ErrMessage) when is_integer(ErrCode), is_binary(ErrMessage) ->
-    jiffy:encode(#{
+json_error(ErrCode, ErrMessage) when is_integer(ErrCode) ->
+    iolist_to_binary(json:encode(#{
         <<"error">> => #{
             <<"code">> => ErrCode,
             <<"message">> => ErrMessage
         }
-    }, [force_utf8]).
+    })).
 
 uuid() ->
     rand_bytes(16).
@@ -112,6 +98,14 @@ assert_call(true, Fun) ->
 assert_call(false, _) ->
     ok.
 
+-spec sha256(Str :: string() | binary()) -> binary().
+sha256(Str) when is_list(Str) ->
+    sha256(unicode:characters_to_binary(Str));
+sha256(Bin) when is_binary(Bin) ->
+    HashBin = crypto:hash(sha256, Bin),
+    HexStr = lists:flatten([io_lib:format("~2.16.0B", [B]) || B <- binary:bin_to_list(HashBin)]),
+    list_to_binary(string:lowercase(HexStr)).
+
 -spec md5(Str :: binary()) -> binary().
 md5(Str) when is_binary(Str) ->
     list_to_binary(lists:flatten([hex(X) || <<X:4>> <= erlang:md5(Str)])).
@@ -123,18 +117,9 @@ hex(N) ->
 
 %% 转换映射器
 -spec parse_mapper(Mapper :: binary() | string()) -> error | {ok, F :: fun((binary(), any()) -> any())}.
-parse_mapper(Mapper) when is_binary(Mapper) ->
-    parse_mapper(binary_to_list(Mapper));
-parse_mapper(Mapper) when is_list(Mapper) ->
-    {ok, Tokens, _} = erl_scan:string(Mapper),
-    {ok, ExprList} = erl_parse:parse_exprs(Tokens),
-    {value, F, _} = erl_eval:exprs(ExprList, []),
-    case is_function(F, 2) orelse is_function(F, 3) of
-        true ->
-            {ok, F};
-        false ->
-            error
-    end.
+parse_mapper(_Mapper) ->
+    %% Dynamic Erlang eval is intentionally disabled in runtime code.
+    error.
 
 -spec float_to_binary(Num :: number(), integer()) -> binary().
 float_to_binary(V, _) when is_integer(V) ->
@@ -149,13 +134,3 @@ assert(false, F) when is_function(F) ->
     F();
 assert(false, Msg) ->
     throw(Msg).
-
-
--spec file_uri(Filename :: binary()) -> error | {ok, FileUri :: binary()}.
-file_uri(Filename) when is_binary(Filename) ->
-    case binary:split(Filename, <<"-">>, [global]) of
-        [Year, Month, Day | _] ->
-            {ok, <<"https://lgsiot.njau.edu.cn/upload/", Year/binary, $/, Month/binary, $/, Day/binary, $/, Filename/binary>>};
-        _ ->
-            error
-    end.

apps/iot/src/mnesia/efka_client_store.erl

@@ -0,0 +1,154 @@
+%%%-------------------------------------------------------------------
+%%% @doc efka_client table store based on mnesia.
+%%%-------------------------------------------------------------------
+-module(efka_client_store).
+
+-include("iot_tables.hrl").
+
+-export([ensure_table/0, register/2, get/1, update/1, delete/1, list/0]).
+-export([auth/2, verify_heartbeat/4]).
+
+-define(TOKEN_HASH_ITERATIONS, 32).
+-define(TOKEN_HASH_BYTES, 32).
+-define(HEARTBEAT_TIMESTAMP_WINDOW, 120).
+
+-spec ensure_table() -> ok | {error, term()}.
+ensure_table() ->
+    case mnesia:create_table(efka_client, [
+        {attributes, record_info(fields, efka_client)},
+        {type, set},
+        {record_name, efka_client},
+        {disc_copies, [node()]}
+    ]) of
+        {atomic, ok} ->
+            ok;
+        {aborted, {already_exists, efka_client}} ->
+            ok;
+        {aborted, Reason} ->
+            {error, Reason}
+    end.
+
+-spec auth(UUID :: binary(), Token :: binary()) -> boolean().
+auth(UUID, Token) when is_binary(UUID), UUID =/= <<>>, is_binary(Token) ->
+    case mnesia:dirty_read(efka_client, UUID) of
+        [] ->
+            false;
+        [#efka_client{token_hash = TokenHash, salt = Salt}] ->
+            hash_token(Token, Salt) =:= TokenHash
+    end.
+
+-spec verify_heartbeat(UUID :: binary(), Timestamp :: integer(), Payload :: binary(), Mac :: binary()) -> boolean().
+verify_heartbeat(UUID, Timestamp, Payload, Mac)
+  when is_binary(UUID), UUID =/= <<>>, is_integer(Timestamp), is_binary(Payload), is_binary(Mac) ->
+    case valid_heartbeat_timestamp(Timestamp) of
+        true ->
+            case mnesia:dirty_read(efka_client, UUID) of
+                [] ->
+                    false;
+                [#efka_client{heartbeat_secret = HeartbeatSecret}] ->
+                    ExpectedMac = crypto:mac(hmac, sha256, HeartbeatSecret, Payload),
+                    ExpectedMac =:= Mac
+            end;
+        false ->
+            false
+    end.
+
+-spec register(UUID :: binary(), Token :: binary()) -> ok | {error, term()}.
+register(UUID, Token) when is_binary(UUID), UUID =/= <<>>, is_binary(Token), byte_size(Token) >= 32 ->
+    case mnesia:transaction(fun() ->
+        case mnesia:read(efka_client, UUID, write) of
+            [] ->
+                %% 1. 生成 16 字节随机盐
+                Salt = crypto:strong_rand_bytes(16),
+                TokenHash = hash_token(Token, Salt),
+                HeartbeatSecret = heartbeat_secret(Token),
+                ok = mnesia:write(#efka_client{
+                    uuid = UUID,
+                    token_hash = TokenHash,
+                    salt = Salt,
+                    heartbeat_secret = HeartbeatSecret,
+                    timestamp = iot_util:timestamp()
+                }),
+                ok;
+            [_] ->
+                {error, already_exists}
+        end
+                            end) of
+        {atomic, ok} ->
+            ok;
+        {atomic, {error, Reason}} ->
+            {error, Reason};
+        {aborted, Reason} ->
+            {error, Reason}
+    end.
+
+-spec get(binary()) -> {ok, #efka_client{}} | not_found | {error, term()}.
+get(UUID) when is_binary(UUID), UUID =/= <<>> ->
+    case mnesia:transaction(fun() ->
+        case mnesia:read(efka_client, UUID, read) of
+            [Client = #efka_client{}] ->
+                {ok, Client};
+            [] ->
+                not_found
+        end
+    end) of
+        {atomic, Result} ->
+            Result;
+        {aborted, Reason} ->
+            {error, Reason}
+    end.
+
+-spec update(#efka_client{}) -> ok | {error, term()}.
+update(Client = #efka_client{uuid = UUID}) when is_binary(UUID), UUID =/= <<>> ->
+    case mnesia:transaction(fun() ->
+        case mnesia:read(efka_client, UUID, write) of
+            [_] ->
+                ok = mnesia:write(Client),
+                ok;
+            [] ->
+                {error, not_found}
+        end
+    end) of
+        {atomic, ok} ->
+            ok;
+        {atomic, {error, Reason}} ->
+            {error, Reason};
+        {aborted, Reason} ->
+            {error, Reason}
+    end.
+
+-spec delete(binary()) -> ok | {error, term()}.
+delete(UUID) when is_binary(UUID), UUID =/= <<>> ->
+    case mnesia:transaction(fun() ->
+        ok = mnesia:delete({efka_client, UUID}),
+        ok
+    end) of
+        {atomic, ok} ->
+            ok;
+        {aborted, Reason} ->
+            {error, Reason}
+    end.
+
+-spec list() -> {ok, [#efka_client{}]} | {error, term()}.
+list() ->
+    case mnesia:transaction(fun() ->
+        mnesia:foldl(fun(Client = #efka_client{}, Acc) -> [Client | Acc] end, [], efka_client)
+    end) of
+        {atomic, Clients} ->
+            {ok, lists:reverse(Clients)};
+        {aborted, Reason} ->
+            {error, Reason}
+    end.
+
+-spec hash_token(binary(), binary()) -> binary().
+hash_token(Token, Salt) when is_binary(Token), is_binary(Salt) ->
+    crypto:pbkdf2_hmac(sha256, Token, Salt, ?TOKEN_HASH_ITERATIONS, ?TOKEN_HASH_BYTES).
+
+-spec heartbeat_secret(binary()) -> binary().
+heartbeat_secret(Token) when is_binary(Token) ->
+    crypto:hash(sha256, Token).
+
+-spec valid_heartbeat_timestamp(integer()) -> boolean().
+valid_heartbeat_timestamp(Timestamp) ->
+    Now = iot_util:current_time(),
+    Timestamp =< Now andalso Now - Timestamp =< ?HEARTBEAT_TIMESTAMP_WINDOW.

apps/iot/src/mnesia/iot_mnesia.erl

@@ -0,0 +1,74 @@
+%%%-------------------------------------------------------------------
+%%% @doc Mnesia bootstrap and runtime startup helpers.
+%%%
+%%% `init/0` is a manual, one-time bootstrap operation. Application
+%%% startup must call `start/0`, which only starts an existing schema and
+%%% verifies required tables.
+%%%-------------------------------------------------------------------
+-module(iot_mnesia).
+
+-include("iot_tables.hrl").
+
+-export([safe_start/0, stop/0]).
+
+-define(TABLES, [efka_client]).
+-define(WAIT_TIMEOUT, 5000).
+
+safe_start() ->
+    case ensure_schema() of
+        create_and_start ->
+            ok = mnesia:start(),
+            ok = create_tables(),
+            ok = wait_for_tables();
+        only_start ->
+            ok = mnesia:start(),
+            ok = wait_for_tables()
+    end,
+    log_runtime_info().
+
+ensure_schema() ->
+    stopped = mnesia:stop(),
+    case mnesia:create_schema([node()]) of
+        ok ->
+            create_and_start;
+        {error, {_, {already_exists, _}}} ->
+            only_start;
+        {error, {_, already_exists}} ->
+            only_start;
+        {error, _Reason} ->
+            only_start
+    end.
+
+-spec stop() -> ok.
+stop() ->
+    _ = mnesia:stop(),
+    ok.
+
+-spec create_tables() -> ok.
+create_tables() ->
+    case efka_client_store:ensure_table() of
+        ok ->
+            ok;
+        {error, Reason} ->
+            {error, {create_table_failed, {efka_client, Reason}}}
+    end.
+
+-spec wait_for_tables() -> ok.
+wait_for_tables() ->
+    case mnesia:wait_for_tables(?TABLES, ?WAIT_TIMEOUT) of
+        ok ->
+            ok;
+        {timeout, Tables} ->
+            {error, {mnesia_tables_timeout, Tables}};
+        {error, Reason} ->
+            {error, {mnesia_tables_error, Reason}}
+    end.
+
+-spec log_runtime_info() -> ok.
+log_runtime_info() ->
+    logger:debug("[iot_mnesia] node: ~p, dir: ~p, tables: ~p", [
+        node(),
+        mnesia:system_info(directory),
+        mnesia:system_info(tables)
+    ]),
+    ok.

apps/iot/src/mocker/iot_mqtt_consumer.erl

@@ -1,281 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%% 1. 需要考虑集群部署的相关问题，上行的数据可能在集群中共享
-%%% 2. host进程不能直接去监听topic，这样涉及到新增和下线的很多问题
-%%% @end
-%%% Created : 12. 3月 2023 21:27
-%%%-------------------------------------------------------------------
--module(iot_mqtt_consumer).
--author("aresei").
--include("iot.hrl").
-
--behaviour(gen_server).
-
-%% API
--export([start_link/0]).
--export([mock/5]).
-
-%% gen_server callbacks
--export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
-
--define(SERVER, ?MODULE).
--define(RETRY_INTERVAL, 5000).
-
-%% 执行超时时间
--define(EXECUTE_TIMEOUT, 10 * 1000).
-
-%% 需要订阅的主题信息
--define(Topics,[
-    {<<"CET/NX/download">>, 2}
-]).
-
--record(state, {
-    conn_pid :: undefined | pid(),
-    logger_pid :: pid(),
-    mqtt_props :: list(),
-    %% 执行中的任务数
-    flight_num = 0
-}).
-
-%%%===================================================================
-%%% API
-%%%===================================================================
-
-mock(LocationCode, Para, SType, CType, Value) when is_binary(LocationCode), is_integer(SType), is_integer(CType), is_integer(Para) ->
-    Req = #{
-        <<"version">> => <<"1.0">>,
-        <<"ts">> => iot_util:current_time(),
-        <<"properties">> => #{
-            <<"type">> => <<"ctrl">>,
-            <<"para">> => Para,
-            <<"stype">> => SType,
-            <<"ctype">> => CType,
-            <<"value">> => Value,
-            <<"timestamp">> => iot_util:current_time()
-        },
-        <<"location_code">> => LocationCode
-    },
-    gen_server:call(?MODULE, {mock, Req}).
-
-%% @doc Spawns the server and registers the local name (unique)
--spec(start_link() ->
-    {ok, Pid :: pid()} | ignore | {error, Reason :: term()}).
-start_link() ->
-    gen_server:start_link({local, ?MODULE}, ?MODULE, [], []).
-
-%%%===================================================================
-%%% gen_server callbacks
-%%%===================================================================
-
-%% @private
-%% @doc Initializes the server
--spec(init(Args :: term()) ->
-    {ok, State :: #state{}} | {ok, State :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term()} | ignore).
-init([]) ->
-    erlang:process_flag(trap_exit, true),
-
-    {ok, Props} = application:get_env(iot, zhongdian),
-    %% 创建转发器, 避免阻塞当前进程的创建，因此采用了延时初始化的机制
-    erlang:start_timer(0, self(), create_consumer),
-    %% 启动日志记录器
-    {ok, LoggerPid} = iot_logger:start_link("zd_directive_data"),
-
-    {ok, #state{mqtt_props = Props, conn_pid = undefined, logger_pid = LoggerPid}}.
-
-%% @private
-%% @doc Handling call messages
--spec(handle_call(Request :: term(), From :: {pid(), Tag :: term()},
-    State :: #state{}) ->
-    {reply, Reply :: term(), NewState :: #state{}} |
-    {reply, Reply :: term(), NewState :: #state{}, timeout() | hibernate} |
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), Reply :: term(), NewState :: #state{}} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_call({mock, Request}, _From, State = #state{conn_pid = ConnPid, flight_num = FlightNum}) when is_pid(ConnPid) ->
-    publish_directive(Request, jiffy:encode(Request, [force_utf8])),
-    {reply, ok, State#state{flight_num = FlightNum + 1}}.
-
-%% @private
-%% @doc Handling cast messages
--spec(handle_cast(Request :: term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_cast(_Request, State = #state{}) ->
-    {noreply, State}.
-
-%% @private
-%% @doc Handling all non call/cast messages
--spec(handle_info(Info :: timeout() | term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_info({disconnect, ReasonCode, Properties}, State) ->
-    lager:debug("[iot_zd_consumer] Recv a DISONNECT packet - ReasonCode: ~p, Properties: ~p", [ReasonCode, Properties]),
-    {stop, disconnected, State};
-%% 必须要做到消息的快速分发，数据的json反序列需要在host进程进行
-handle_info({publish, #{packet_id := _PacketId, payload := Payload, qos := 2, topic := Topic}}, State = #state{flight_num = FlightNum}) ->
-    lager:debug("[iot_zd_consumer] Recv a topic: ~p, publish packet: ~ts, qos: 2", [Topic, Payload]),
-
-    Request = catch jiffy:decode(Payload, [return_maps]),
-    publish_directive(Request, Payload),
-
-    {noreply, State#state{flight_num = FlightNum + 1}};
-
-handle_info({publish, #{packet_id := _PacketId, payload := Payload, qos := Qos, topic := Topic}}, State) ->
-    lager:notice("[iot_zd_consumer] Recv a topic: ~p, publish packet: ~ts, qos: ~p, qos is error", [Topic, Payload, Qos]),
-    {noreply, State};
-
-handle_info({puback, Packet = #{packet_id := _PacketId}}, State = #state{}) ->
-    lager:debug("[iot_zd_consumer] receive puback packet: ~p", [Packet]),
-    {noreply, State};
-
-handle_info({timeout, _, create_consumer}, State = #state{mqtt_props = Props, conn_pid = undefined}) ->
-    try
-        {ok, ConnPid} = create_consumer(Props),
-        {noreply, State#state{conn_pid = ConnPid}}
-    catch _:Error:Stack ->
-        lager:warning("[iot_zd_consumer] config: ~p, create consumer get error: ~p, stack: ~p", [Props, Error, Stack]),
-        erlang:start_timer(?RETRY_INTERVAL, self(), create_consumer),
-        {noreply, State#state{conn_pid = undefined}}
-    end;
-
-%% postman进程挂掉时，重新建立新的
-handle_info({'EXIT', ConnPid, Reason}, State = #state{conn_pid = ConnPid}) ->
-    lager:warning("[iot_zd_consumer] consumer exited with reason: ~p", [Reason]),
-    erlang:start_timer(?RETRY_INTERVAL, self(), create_consumer),
-
-    {noreply, State#state{conn_pid = undefined}};
-
-handle_info({'EXIT', LoggerPid, Reason}, State = #state{logger_pid = LoggerPid}) ->
-    lager:warning("[iot_zd_consumer] logger exited with reason: ~p", [Reason]),
-    {ok, LoggerPid} = iot_logger:start_link("zd_directive_data"),
-
-    {noreply, State#state{logger_pid = LoggerPid}};
-
-handle_info({directive_reply, Reply}, State = #state{logger_pid = LoggerPid, flight_num = FlightNum}) ->
-    FlightInfo = <<"flight_num: ", (integer_to_binary(FlightNum - 1))/binary>>,
-    case Reply of
-        {ok, RawReq, DirectiveResult} ->
-            case DirectiveResult of
-                ok ->
-                    iot_logger:write(LoggerPid, [<<"[success]">>, RawReq, <<"OK">>, FlightInfo]);
-                {ok, Response} when is_binary(Response) ->
-                    iot_logger:write(LoggerPid, [<<"[success]">>, RawReq, Response, FlightInfo]);
-                {error, Reason0} ->
-                    Reason = if
-                                 is_atom(Reason0) -> atom_to_binary(Reason0);
-                                 is_binary(Reason0) -> Reason0;
-                                 true ->  <<"Unknow error">>
-                             end,
-                    iot_logger:write(LoggerPid, [<<"[error]">>, RawReq, Reason, FlightInfo])
-            end;
-        {error, RawReq, Error} when is_binary(Error) ->
-            iot_logger:write(LoggerPid, [<<"[error]">>, RawReq, Error, FlightInfo])
-    end,
-    {noreply, State#state{flight_num = FlightNum - 1}};
-
-handle_info(Info, State = #state{}) ->
-    lager:notice("[iot_zd_consumer] get a unknown info: ~p", [Info]),
-    {noreply, State}.
-
-%% @private
-%% @doc This function is called by a gen_server when it is about to
-%% terminate. It should be the opposite of Module:init/1 and do any
-%% necessary cleaning up. When it returns, the gen_server terminates
-%% with Reason. The return value is ignored.
--spec(terminate(Reason :: (normal | shutdown | {shutdown, term()} | term()),
-    State :: #state{}) -> term()).
-terminate(Reason, _State = #state{conn_pid = ConnPid}) when is_pid(ConnPid) ->
-    %% 取消topic的订阅
-    TopicNames = lists:map(fun({Name, _}) -> Name end, ?Topics),
-    {ok, _Props, _ReasonCode} = emqtt:unsubscribe(ConnPid, #{}, TopicNames),
-
-    ok = emqtt:disconnect(ConnPid),
-    lager:debug("[iot_zd_consumer] terminate with reason: ~p", [Reason]),
-    ok;
-terminate(Reason, _State) ->
-    lager:debug("[iot_zd_consumer] terminate with reason: ~p", [Reason]),
-    ok.
-
-%% @private
-%% @doc Convert process state when code is changed
--spec(code_change(OldVsn :: term() | {down, term()}, State :: #state{},
-    Extra :: term()) ->
-    {ok, NewState :: #state{}} | {error, Reason :: term()}).
-code_change(_OldVsn, State = #state{}, _Extra) ->
-    {ok, State}.
-
-%%%===================================================================
-%%% Internal functions
-%%%===================================================================
-
-publish_directive(#{<<"version">> := Version, <<"location_code">> := LocationCode, <<"properties">> := DirectiveParams}, RawReq) ->
-    %% 通过LocationCode查找到主机和Device_uuid
-    ReceiverPid = self(),
-    case redis_client:hgetall(LocationCode) of
-        {ok, #{<<"host_uuid">> := HostUUID, <<"device_uuid">> := DeviceUUID}} ->
-            case iot_host:get_pid(HostUUID) of
-                undefined ->
-                    ReceiverPid ! {directive_reply, {error, RawReq, <<"host uuid: ", HostUUID/binary, " not found">>}};
-                Pid when is_pid(Pid) ->
-                    ok
-            end;
-        {ok, Map} when is_map(Map) ->
-            RedisData = iolist_to_binary(jiffy:encode(Map, [force_utf8])),
-            ReceiverPid ! {directive_reply, {error, RawReq, <<"invalid redis data: ", RedisData/binary>>}};
-        _ ->
-            ReceiverPid ! {directive_reply, {error, RawReq, <<"location_code: ", LocationCode/binary, " not found in redis">>}}
-    end;
-publish_directive(Other, RawReq) ->
-    lager:warning("[iot_zd_consumer] get a error message: ~p", [Other]),
-    self() ! {directive_reply, {error, RawReq, <<"unknown directive">>}}.
-
--spec create_consumer(Props :: list()) -> {ok, ConnPid :: pid()} | {error, Reason :: any()}.
-create_consumer(Props) when is_list(Props) ->
-    Node = atom_to_binary(node()),
-    ClientId = <<"mqtt-client-", Node/binary, "-zhongdian_mqtt_consumer">>,
-
-    %% 建立到emqx服务器的连接
-    Host = proplists:get_value(host, Props),
-    Port = proplists:get_value(port, Props, 18080),
-    Username = proplists:get_value(username, Props),
-    Password = proplists:get_value(password, Props),
-    Keepalive = proplists:get_value(keepalive, Props, 86400),
-
-    Opts = [
-        {clientid, ClientId},
-        {host, Host},
-        {port, Port},
-        {owner, self()},
-        {tcp_opts, []},
-        {username, Username},
-        {password, Password},
-        {keepalive, Keepalive},
-        {auto_ack, true},
-        {connect_timeout, 5000},
-        {proto_ver, v5},
-        {retry_interval, 5000}
-    ],
-
-    %% 建立到emqx服务器的连接
-    lager:debug("[iot_zd_consumer] opts is: ~p", [Opts]),
-    case emqtt:start_link(Opts) of
-        {ok, ConnPid} ->
-            %% 监听和host相关的全部事件
-            lager:debug("[iot_zd_consumer] start conntecting, pid: ~p", [ConnPid]),
-            {ok, _} = emqtt:connect(ConnPid),
-            lager:debug("[iot_zd_consumer] connect success, pid: ~p", [ConnPid]),
-            SubscribeResult = emqtt:subscribe(ConnPid, ?Topics),
-            lager:debug("[iot_zd_consumer] subscribe topics: ~p, result is: ~p", [?Topics, SubscribeResult]),
-
-            {ok, ConnPid};
-        ignore ->
-            {error, ignore};
-        {error, Reason} ->
-            {error, Reason}
-    end.

apps/iot/src/mysql/mysql_pool.erl

@@ -1,48 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2018, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 29. 九月 2018 17:01
-%%%-------------------------------------------------------------------
--module(mysql_pool).
--author("aresei").
-
-%% API
--export([get_row/2, get_row/3, get_all/2, get_all/3]).
--export([update/4, update_by/2, update_by/3, insert/4]).
-
-%% 从数据库中查找一行记录
--spec get_row(Pool :: atom(), Sql::binary()) -> {ok, Record::map()} | undefined.
-get_row(Pool, Sql) when is_atom(Pool), is_binary(Sql) ->
-    poolboy:transaction(Pool, fun(ConnPid) -> mysql_provider:get_row(ConnPid, Sql) end).
-
--spec get_row(Pool :: atom(), Sql::binary(), Params::list()) -> {ok, Record::map()} | undefined.
-get_row(Pool, Sql, Params) when is_atom(Pool), is_binary(Sql), is_list(Params) ->
-    poolboy:transaction(Pool, fun(ConnPid) -> mysql_provider:get_row(ConnPid, Sql, Params) end).
-
--spec get_all(Pool :: atom(), Sql::binary()) -> {ok, Rows::list()} | {error, Reason :: any()}.
-get_all(Pool, Sql) when is_atom(Pool), is_binary(Sql) ->
-    poolboy:transaction(Pool, fun(ConnPid) -> mysql_provider:get_all(ConnPid, Sql) end).
-
--spec get_all(Pool :: atom(), Sql::binary(), Params::list()) -> {ok, Rows::list()} | {error, Reason::any()}.
-get_all(Pool, Sql, Params) when is_atom(Pool), is_binary(Sql), is_list(Params) ->
-    poolboy:transaction(Pool, fun(ConnPid) -> mysql_provider:get_all(ConnPid, Sql, Params) end).
-
--spec insert(Pool :: atom(), Table :: binary(), Fields :: map() | list(), boolean()) ->
-    ok | {ok, InsertId :: integer()} | {error, Reason :: any()}.
-insert(Pool, Table, Fields, FetchInsertId) when is_atom(Pool), is_binary(Table), is_list(Fields); is_map(Fields), is_boolean(FetchInsertId) ->
-    poolboy:transaction(Pool, fun(ConnPid) -> mysql_provider:insert(ConnPid, Table, Fields, FetchInsertId) end).
-
--spec update_by(Pool :: atom(), UpdateSql :: binary()) -> {ok, AffectedRows :: integer()} | {error, Reason :: any()}.
-update_by(Pool, UpdateSql) when is_atom(Pool), is_binary(UpdateSql) ->
-    poolboy:transaction(Pool, fun(ConnPid) -> mysql_provider:update_by(ConnPid, UpdateSql) end).
-
--spec update_by(Pool :: atom(), UpdateSql :: binary(), Params :: list()) -> {ok, AffectedRows :: integer()} | {error, Reason :: any()}.
-update_by(Pool, UpdateSql, Params) when is_atom(Pool), is_binary(UpdateSql) ->
-    poolboy:transaction(Pool, fun(ConnPid) -> mysql_provider:update_by(ConnPid, UpdateSql, Params) end).
-
--spec update(Pool :: atom(), Table :: binary(), Fields :: map(), WhereFields :: map()) -> {ok, AffectedRows::integer()} | {error, Reason::any()}.
-update(Pool, Table, Fields, WhereFields) when is_atom(Pool), is_binary(Table), is_map(Fields), is_map(WhereFields) ->
-    poolboy:transaction(Pool, fun(ConnPid) -> mysql_provider:update(ConnPid, Table, Fields, WhereFields) end).

apps/iot/src/mysql/mysql_provider.erl

@@ -1,144 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2018, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 29. 九月 2018 17:01
-%%%-------------------------------------------------------------------
--module(mysql_provider).
--author("aresei").
-
-%% API
--export([get_row/2, get_row/3, get_all/2, get_all/3]).
--export([update/4, update_by/2, update_by/3, insert/4]).
-
-%% 从数据库中查找一行记录
--spec get_row(ConnPid :: pid(), Sql::binary()) -> {ok, Record::map()} | undefined.
-get_row(ConnPid, Sql) when is_pid(ConnPid), is_binary(Sql) ->
-    lager:debug("[mysql_client] get_row sql is: ~p", [Sql]),
-    case mysql:query(ConnPid, Sql) of
-        {ok, Names, [Row | _]} ->
-            {ok, maps:from_list(lists:zip(Names, Row))};
-        {ok, _, []} ->
-            undefined;
-        Error ->
-            lager:warning("[mysql_client] get error: ~p", [Error]),
-            undefined
-    end.
-
--spec get_row(ConnPid :: pid(), Sql::binary(), Params::list()) -> {ok, Record::map()} | undefined.
-get_row(ConnPid, Sql, Params) when is_pid(ConnPid), is_binary(Sql), is_list(Params) ->
-    lager:debug("[mysql_client] get_row sql is: ~p, params: ~p", [Sql, Params]),
-    case mysql:query(ConnPid, Sql, Params) of
-        {ok, Names, [Row | _]} ->
-            {ok, maps:from_list(lists:zip(Names, Row))};
-        {ok, _, []} ->
-            undefined;
-        Error ->
-            lager:warning("[mysql_client] get error: ~p", [Error]),
-            undefined
-    end.
-
--spec get_all(ConnPid :: pid(), Sql::binary()) -> {ok, Rows::list()} | {error, Reason :: any()}.
-get_all(ConnPid, Sql) when is_pid(ConnPid), is_binary(Sql) ->
-    lager:debug("[mysql_client] get_all sql is: ~p", [Sql]),
-    case mysql:query(ConnPid, Sql) of
-        {ok, Names, Rows} ->
-            {ok, lists:map(fun(Row) -> maps:from_list(lists:zip(Names, Row)) end, Rows)};
-        {error, Reason} ->
-            lager:warning("[mysql_client] get error: ~p", [Reason]),
-            {error, Reason}
-    end.
-
--spec get_all(ConnPid :: pid(), Sql::binary(), Params::list()) -> {ok, Rows::list()} | {error, Reason::any()}.
-get_all(ConnPid, Sql, Params) when is_pid(ConnPid), is_binary(Sql), is_list(Params) ->
-    lager:debug("[mysql_client] get_all sql is: ~p, params: ~p", [Sql, Params]),
-    case mysql:query(ConnPid, Sql, Params) of
-        {ok, Names, Rows} ->
-            {ok, lists:map(fun(Row) -> maps:from_list(lists:zip(Names, Row)) end, Rows)};
-        {error, Reason} ->
-            lager:warning("[mysql_client] get error: ~p", [Reason]),
-            {error, Reason}
-    end.
-
--spec insert(ConnPid :: pid(), Table :: binary(), Fields :: map() | list(), boolean()) ->
-    ok | {ok, InsertId :: integer()} | {error, Reason :: any()}.
-insert(ConnPid, Table, Fields, FetchInsertId) when is_pid(ConnPid), is_binary(Table), is_map(Fields), is_boolean(FetchInsertId) ->
-    insert(ConnPid, Table, maps:to_list(Fields), FetchInsertId);
-insert(ConnPid, Table, Fields, FetchInsertId) when is_pid(ConnPid), is_binary(Table), is_list(Fields), is_boolean(FetchInsertId) ->
-    {Keys, Values} = kvs(Fields),
-
-    FieldSql = iolist_to_binary(lists:join(<<", ">>, Keys)),
-    Placeholders = lists:duplicate(length(Keys), <<"?">>),
-    ValuesPlaceholder = iolist_to_binary(lists:join(<<", ">>, Placeholders)),
-
-    Sql = <<"INSERT INTO ", Table/binary, "(", FieldSql/binary, ") VALUES(", ValuesPlaceholder/binary, ")">>,
-    lager:debug("[mysql_client] insert sql is: ~p, params: ~p", [Sql, Values]),
-    case mysql:query(ConnPid, Sql, Values) of
-        ok ->
-            case FetchInsertId of
-                true ->
-                    InsertId = mysql:insert_id(ConnPid),
-                    {ok, InsertId};
-                false ->
-                    ok
-            end;
-        Error ->
-            Error
-    end.
-
--spec update_by(ConnPid :: pid(), UpdateSql :: binary()) -> {ok, AffectedRows :: integer()} | {error, Reason :: any()}.
-update_by(ConnPid, UpdateSql) when is_pid(ConnPid), is_binary(UpdateSql) ->
-    lager:debug("[mysql_client] updateBySql sql: ~p", [UpdateSql]),
-    case mysql:query(ConnPid, UpdateSql) of
-        ok ->
-            AffectedRows = mysql:affected_rows(ConnPid),
-            {ok, AffectedRows};
-        Error ->
-            Error
-    end.
-
--spec update_by(ConnPid :: pid(), UpdateSql :: binary(), Params :: list()) -> {ok, AffectedRows :: integer()} | {error, Reason :: any()}.
-update_by(ConnPid, UpdateSql, Params) when is_pid(ConnPid), is_binary(UpdateSql) ->
-    lager:debug("[mysql_client] updateBySql sql: ~p, params: ~p", [UpdateSql, Params]),
-    case mysql:query(ConnPid, UpdateSql, Params) of
-        ok ->
-            AffectedRows = mysql:affected_rows(ConnPid),
-            {ok, AffectedRows};
-        Error ->
-            Error
-    end.
-
--spec update(ConnPid :: pid(), Sql :: binary(), Fields :: map(), WhereFields :: map()) ->
-    {ok, AffectedRows::integer()} | {error, Reason::any()}.
-update(ConnPid, Table, Fields, WhereFields) when is_pid(ConnPid), is_binary(Table), is_map(Fields), is_map(WhereFields) ->
-    %% 拼接set
-    {SetKeys, SetVals} = kvs(Fields),
-    SetKeys1 = lists:map(fun(K) when is_binary(K) -> <<"`", K/binary, "` = ?">> end, SetKeys),
-    SetSql = iolist_to_binary(lists:join(<<", ">>, SetKeys1)),
-
-    %% 拼接where
-    {WhereKeys, WhereVals} = kvs(WhereFields),
-    WhereKeys1 = lists:map(fun(K) when is_binary(K) -> <<"`", K/binary, "` = ?">> end, WhereKeys),
-    WhereSql = iolist_to_binary(lists:join(<<" AND ">>, WhereKeys1)),
-
-    Params = SetVals ++ WhereVals,
-
-    Sql = <<"UPDATE ", Table/binary, " SET ",  SetSql/binary, " WHERE ", WhereSql/binary>>,
-    lager:debug("[mysql_client] update sql is: ~p, params: ~p", [Sql, Params]),
-    case mysql:query(ConnPid, Sql, Params) of
-        ok ->
-            AffectedRows = mysql:affected_rows(ConnPid),
-            {ok, AffectedRows};
-        Error ->
-            lager:error("[mysql_client] update sql: ~p, params: ~p, get a error: ~p", [Sql, Params, Error]),
-            Error
-    end.
-
--spec kvs(Fields :: map() | list()) -> {Keys :: list(), Values :: list()}.
-kvs(Fields) when is_map(Fields) ->
-    kvs(maps:to_list(Fields));
-kvs(Fields) when is_list(Fields) ->
-    {Keys0, Values0} = lists:foldl(fun({K, V}, {Acc0, Acc1}) -> {[K|Acc0], [V|Acc1]} end, {[], []}, Fields),
-    {lists:reverse(Keys0), lists:reverse(Values0)}.

apps/iot/src/simulator/simulator_api_handler.erl

@@ -0,0 +1,156 @@
+%%%-------------------------------------------------------------------
+%%% @author Codex
+%%% @doc
+%%% 本地模拟 control api，用于替代 iot_api_client 依赖的远端接口。
+%%% @end
+%%%-------------------------------------------------------------------
+-module(simulator_api_handler).
+
+-include_lib("endpoint/include/endpoint.hrl").
+
+-export([handle_request/4]).
+
+-define(HOST_UUID, <<"qbxmjyzrkpntfgswaevodhluicqzxplkm">>).
+-define(HOST_ID, 1).
+-define(DEVICE_UUID, <<"sim-device-001">>).
+-define(DEVICE_ID, 1).
+-define(ENDPOINT_ID, 1).
+-define(TABLE, ?MODULE).
+
+-spec handle_request(string(), string(), map(), map()) ->
+    {ok, non_neg_integer(), iodata()}.
+handle_request("GET", "/simulator/get_all_hosts", _, _) ->
+    {ok, 200, iot_util:json_data([?HOST_UUID])};
+handle_request("GET", "/simulator/get_host_by_uuid", #{<<"uuid">> := UUID}, _) when is_binary(UUID) ->
+    reply_host_by_uuid(UUID);
+handle_request("GET", "/simulator/get_host_by_id", #{<<"host_id">> := HostIdBin}, _) when is_binary(HostIdBin) ->
+    case binary_to_integer(HostIdBin) of
+        ?HOST_ID ->
+            {ok, 200, iot_util:json_data(host_info())};
+        _ ->
+            {ok, 200, iot_util:json_error(404, <<"host not found">>)}
+    end;
+handle_request("POST", "/simulator/change_host_status", _, #{<<"uuid">> := UUID, <<"new_status">> := Status})
+        when is_binary(UUID), is_integer(Status) ->
+    reply_change_host_status(UUID, Status);
+handle_request("GET", "/simulator/get_host_devices", #{<<"host_id">> := HostIdBin}, _) when is_binary(HostIdBin) ->
+    case binary_to_integer(HostIdBin) of
+        ?HOST_ID ->
+            {ok, 200, iot_util:json_data([device_info()])};
+        _ ->
+            {ok, 200, iot_util:json_error(404, <<"host not found">>)}
+    end;
+handle_request("GET", "/simulator/get_device_by_uuid", #{<<"device_uuid">> := DeviceUUID}, _) when is_binary(DeviceUUID) ->
+    case DeviceUUID of
+        ?DEVICE_UUID ->
+            {ok, 200, iot_util:json_data(device_info())};
+        _ ->
+            {ok, 200, iot_util:json_error(404, <<"device not found">>)}
+    end;
+handle_request("POST", "/simulator/change_device_status", _, #{<<"device_uuid">> := DeviceUUID, <<"new_status">> := Status})
+        when is_binary(DeviceUUID), is_integer(Status) ->
+    reply_change_device_status(DeviceUUID, Status);
+handle_request("GET", "/simulator/get_all_endpoints", _, _) ->
+    {ok, 200, iot_util:json_data([endpoint_info()])};
+handle_request("GET", "/simulator/get_endpoint", #{<<"id">> := IdBin}, _) when is_binary(IdBin) ->
+    case binary_to_integer(IdBin) of
+        ?ENDPOINT_ID ->
+            {ok, 200, iot_util:json_data(endpoint_info())};
+        _ ->
+            {ok, 200, iot_util:json_error(404, <<"endpoint not found">>)}
+    end;
+handle_request("POST", "/simulator/endpoint_sink", _, Body) ->
+    logger:debug("[simulator_api_handler] receive endpoint sink body: ~p", [Body]),
+    {ok, 200, iot_util:json_data(<<"ok">>)};
+handle_request(_, Path, _, _) ->
+    PathBin = list_to_binary(Path),
+    {ok, 200, iot_util:json_error(-1, <<"url: ", PathBin/binary, " not found">>)}.
+
+%%%===================================================================
+%%% Internal functions
+%%%===================================================================
+
+-spec reply_host_by_uuid(binary()) -> {ok, non_neg_integer(), iodata()}.
+reply_host_by_uuid(?HOST_UUID) ->
+    {ok, 200, iot_util:json_data(host_info())};
+reply_host_by_uuid(_) ->
+    {ok, 200, iot_util:json_error(404, <<"host not found">>)}.
+
+-spec reply_change_host_status(binary(), integer()) -> {ok, non_neg_integer(), iodata()}.
+reply_change_host_status(?HOST_UUID, Status) ->
+    ok = put_state({host_status, ?HOST_UUID}, Status),
+    {ok, 200, iot_util:json_data(1)};
+reply_change_host_status(_, _) ->
+    {ok, 200, iot_util:json_error(404, <<"host not found">>)}.
+
+-spec reply_change_device_status(binary(), integer()) -> {ok, non_neg_integer(), iodata()}.
+reply_change_device_status(?DEVICE_UUID, Status) ->
+    ok = put_state({device_status, ?DEVICE_UUID}, Status),
+    {ok, 200, iot_util:json_data(1)};
+reply_change_device_status(_, _) ->
+    {ok, 200, iot_util:json_error(404, <<"device not found">>)}.
+
+-spec host_info() -> map().
+host_info() ->
+    #{
+        <<"id">> => ?HOST_ID,
+        <<"uuid">> => ?HOST_UUID,
+        <<"authorize_status">> => 1,
+        <<"status">> => get_state({host_status, ?HOST_UUID}, 1)
+    }.
+
+-spec device_info() -> map().
+device_info() ->
+    #{
+        <<"id">> => ?DEVICE_ID,
+        <<"host_id">> => ?HOST_ID,
+        <<"device_uuid">> => ?DEVICE_UUID,
+        <<"status">> => get_state({device_status, ?DEVICE_UUID}, 1)
+    }.
+
+-spec endpoint_info() -> map().
+endpoint_info() ->
+    #{
+        <<"id">> => ?ENDPOINT_ID,
+        <<"matcher">> => <<"simulator.metric">>,
+        <<"title">> => <<"Simulator HTTP Endpoint">>,
+        <<"type">> => <<"http">>,
+        <<"config">> => #{
+            <<"url">> => <<"http://127.0.0.1:18090/simulator/endpoint_sink">>,
+            <<"pool_size">> => 4
+        },
+        <<"status">> => 1,
+        <<"updated_at">> => iot_util:timestamp(),
+        <<"created_at">> => iot_util:timestamp()
+    }.
+
+-spec ensure_table() -> ets:tid().
+ensure_table() ->
+    case ets:info(?TABLE) of
+        undefined ->
+            try ets:new(?TABLE, [named_table, public, set]) of
+                Tid ->
+                    Tid
+            catch
+                error:badarg ->
+                    ?TABLE
+            end;
+        _ ->
+            ?TABLE
+    end.
+
+-spec get_state(term(), term()) -> term().
+get_state(Key, Default) ->
+    _ = ensure_table(),
+    case ets:lookup(?TABLE, Key) of
+        [{_, Value}] ->
+            Value;
+        [] ->
+            Default
+    end.
+
+-spec put_state(term(), term()) -> ok.
+put_state(Key, Value) ->
+    _ = ensure_table(),
+    true = ets:insert(?TABLE, {Key, Value}),
+    ok.

apps/iot/src/tcp/tcp_channel.erl

@@ -1,218 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author licheng5
-%%% @copyright (C) 2021, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 11. 1月 2021 上午12:17
-%%%-------------------------------------------------------------------
--module(tcp_channel).
--author("licheng5").
--include("iot.hrl").
--include("message_pb.hrl").
-
-%% API
--export([pub/3, async_call/4, command/3]).
--export([stop/2]).
-
--export([start_link/2]).
-%% gen_server callbacks
--export([init/1, handle_call/3, handle_cast/2, handle_info/2, code_change/3, terminate/2]).
-
--record(state, {
-    transport,
-    socket,
-
-    uuid :: undefined | binary(),
-    %% 用户进程id
-    host_pid = undefined,
-    %% 发送消息对应的id
-    packet_id = 1 :: integer(),
-
-    %% 请求响应的对应关系
-    inflight = #{}
-}).
-
-%% 向通道中写入消息
--spec pub(Pid :: pid(), Topic :: binary(), Content :: binary()) -> no_return().
-pub(Pid, Topic, Content) when is_pid(Pid), is_binary(Topic), is_binary(Content) ->
-    gen_server:cast(Pid, {pub, Topic, Content}).
-
-%% 向通道中写入消息
--spec command(Pid :: pid(), CommandType :: integer(), Command :: binary()) -> no_return().
-command(Pid, CommandType, Command) when is_pid(Pid), is_integer(CommandType), is_binary(Command) ->
-    gen_server:cast(Pid, {command, CommandType, Command}).
-
-%% 向通道中写入消息
--spec async_call(Pid :: pid(), ReceiverPid :: pid(), CallType :: integer(), CallBin :: binary()) -> Ref :: reference().
-async_call(Pid, ReceiverPid, CallType, CallBin) when is_pid(Pid), is_pid(ReceiverPid), is_integer(CallType), is_binary(CallBin) ->
-    Ref = make_ref(),
-    gen_server:cast(Pid, {async_call, ReceiverPid, Ref, CallType, CallBin}),
-    Ref.
-
-%% 关闭方法
--spec stop(Pid :: pid(), Reason :: any()) -> no_return().
-stop(undefined, _Reason) ->
-    ok;
-stop(Pid, Reason) when is_pid(Pid) ->
-    gen_server:stop(Pid, Reason, 5000).
-
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-%% 逻辑处理方法
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-
-start_link(Transport, Sock) ->
-    {ok, proc_lib:spawn_link(?MODULE, init, [[Transport, Sock]])}.
-
-init([Transport, Sock]) ->
-    lager:debug("[sdlan_channel] get a new connection: ~p", [Sock]),
-    case Transport:wait(Sock) of
-        {ok, NewSock} ->
-            Transport:setopts(Sock, [{active, true}]),
-            % erlang:start_timer(?PING_TICKER, self(), ping_ticker),
-            gen_server:enter_loop(?MODULE, [], #state{transport = Transport, socket = NewSock});
-        {error, Reason} ->
-            {stop, Reason}
-    end.
-
-handle_call(_Request, _From, State) ->
-    {reply, ok, State}.
-
-%% 发送消息, 基于pub/sub机制
-handle_cast({pub, Topic, Content}, State = #state{transport = Transport, socket = Socket}) ->
-    PubBin = message_pb:encode_msg(#pub{topic = Topic, content = Content}),
-    Transport:send(Socket, <<?PACKET_PUB, PubBin/binary>>),
-    {noreply, State};
-
-%% 发送Command消息
-handle_cast({command, CommandType, Command}, State = #state{transport = Transport, socket = Socket}) ->
-    Transport:send(Socket, <<?PACKET_COMMAND, CommandType:8, Command/binary>>),
-    {noreply, State};
-
-%% 推送消息
-handle_cast({async_call, ReceiverPid, Ref, CallType, CallBin}, State = #state{transport = Transport, socket = Socket, packet_id = PacketId, inflight = Inflight}) ->
-    Transport:send(Socket, <<?PACKET_ASYNC_CALL, PacketId:32, CallType:8, CallBin/binary>>),
-    {noreply, State#state{packet_id = PacketId + 1, inflight = maps:put(PacketId, {ReceiverPid, Ref}, Inflight)}}.
-
-%% auth验证
-handle_info({tcp, Socket, <<?PACKET_REQUEST, PacketId:32, ?METHOD_AUTH:8, AuthRequestBin/binary>>}, State = #state{transport = Transport, socket = Socket}) ->
-    #auth_request{ uuid = UUID, username = Username, token = Token, salt = Salt, timestamp = Timestamp } = message_pb:decode_msg(AuthRequestBin, auth_request),
-    lager:debug("[ws_channel] auth uuid: ~p", [UUID]),
-    case iot_auth:check(Username, Token, UUID, Salt, Timestamp) of
-        true ->
-            case host_bo:get_host_by_uuid(UUID) of
-                undefined ->
-                    lager:warning("[ws_channel] uuid: ~p, user: ~p, host not found", [UUID, Username]),
-                    {stop, State};
-                {ok, _} ->
-                    %% 尝试启动主机的服务进程
-                    {ok, HostPid} = iot_host_sup:ensured_host_started(UUID),
-                    case iot_host:attach_channel(HostPid, self()) of
-                        ok ->
-                            %% 建立到host的monitor
-                            erlang:monitor(process, HostPid),
-                            AuthReplyBin = message_pb:encode_msg(#auth_reply{code = 0, message = <<"ok">>}),
-                            Transport:send(Socket, <<?PACKET_RESPONSE, PacketId:32, AuthReplyBin/binary>>),
-
-                            {noreply, State#state{uuid = UUID, host_pid = HostPid}};
-                        {denied, Reason} when is_binary(Reason) ->
-                            erlang:monitor(process, HostPid),
-                            AuthReplyBin = message_pb:encode_msg(#auth_reply{code = 1, message = Reason}),
-                            Transport:send(Socket, <<?PACKET_RESPONSE, PacketId:32, AuthReplyBin/binary>>),
-                            lager:debug("[ws_channel] uuid: ~p, attach channel get error: ~p, stop channel", [UUID, Reason]),
-
-                            {noreply, State#state{uuid = UUID, host_pid = HostPid}};
-
-                        {error, Reason} when is_binary(Reason) ->
-                            AuthReplyBin = message_pb:encode_msg(#auth_reply{code = 2, message = Reason}),
-                            Transport:send(Socket, <<?PACKET_RESPONSE, PacketId:32, AuthReplyBin/binary>>),
-                            lager:debug("[ws_channel] uuid: ~p, attach channel get error: ~p, stop channel", [UUID, Reason]),
-
-                            {stop, State}
-                    end
-            end;
-        false ->
-            lager:warning("[ws_channel] uuid: ~p, user: ~p, auth failed", [UUID, Username]),
-            {stop, State}
-    end;
-
-%% 请求微服务配置
-handle_info({tcp, Socket, <<?PACKET_REQUEST, PacketId:32, ?METHOD_REQUEST_SERVICE_CONFIG:8, ServiceId/binary>>}, State = #state{transport = Transport, socket = Socket}) ->
-    lager:debug("[ws_channel] service_config request service_id: ~p", [ServiceId]),
-    case micro_service_bo:get_service_config(ServiceId) of
-        error ->
-            Transport:send(Socket, <<?PACKET_RESPONSE, PacketId:32>>);
-        {ok, ConfigJson} when is_binary(ConfigJson) ->
-            Transport:send(Socket, <<?PACKET_RESPONSE, PacketId:32, ConfigJson/binary>>)
-    end,
-    {noreply, State};
-
-handle_info({tcp, Socket, <<?PACKET_REQUEST, ?METHOD_DATA:8, Data0/binary>>}, State = #state{socket = Socket, host_pid = HostPid}) when is_pid(HostPid) ->
-    Data = message_pb:decode_msg(Data0, data),
-    iot_host:handle(HostPid, {data, Data}),
-    {noreply, State};
-
-handle_info({tcp, Socket, <<?PACKET_REQUEST, ?METHOD_PING:8, PingData/binary>>}, State = #state{socket = Socket, host_pid = HostPid}) when is_pid(HostPid) ->
-    Ping = message_pb:decode_msg(PingData, ping),
-    iot_host:handle(HostPid, {ping, Ping}),
-    {noreply, State};
-
-handle_info({tcp, Socket, <<?PACKET_REQUEST, ?METHOD_INFORM:8, InformData/binary>>}, State = #state{socket = Socket, host_pid = HostPid}) when is_pid(HostPid) ->
-    ServiceInform = message_pb:decode_msg(InformData, service_inform),
-    iot_host:handle(HostPid, {inform, ServiceInform}),
-    {noreply, State};
-
-handle_info({tcp, Socket, <<?PACKET_REQUEST, ?METHOD_EVENT:8, EventData/binary>>}, State = #state{socket = Socket, host_pid = HostPid}) when is_pid(HostPid) ->
-    Event = message_pb:decode_msg(EventData, event),
-    iot_host:handle(HostPid, {event, Event}),
-    {noreply, State};
-
-%% 主机端的消息响应
-handle_info({tcp, Socket, <<?PACKET_ASYNC_CALL_REPLY, PacketId:32, ResponseBin/binary>>}, State = #state{socket = Socket, uuid = UUID, inflight = Inflight}) when PacketId > 0 ->
-    AsyncCallReply = message_pb:decode_msg(ResponseBin, async_call_reply),
-    lager:debug("[ws_channel] uuid: ~p, get async_call_reply: ~p, packet_id: ~p", [UUID, AsyncCallReply, PacketId]),
-    case maps:take(PacketId, Inflight) of
-        error ->
-            lager:warning("[ws_channel] get unknown async_call_reply message: ~p, packet_id: ~p", [AsyncCallReply, PacketId]),
-            {noreply, State};
-        {{ReceiverPid, Ref}, NInflight} ->
-            case is_pid(ReceiverPid) andalso is_process_alive(ReceiverPid) of
-                true ->
-                    ReceiverPid ! {async_call_reply, Ref, AsyncCallReply};
-                false ->
-                    lager:warning("[ws_channel] get async_call_reply message: ~p, packet_id: ~p, but receiver_pid is deaded", [AsyncCallReply, PacketId])
-            end,
-            {noreply, State#state{inflight = NInflight}}
-    end;
-
-%% 来自efka的ping包
-handle_info({tcp, Socket, <<?PACKET_PING>>}, State = #state{socket = Socket}) ->
-    {noreply, State};
-
-handle_info({tcp_error, Sock, Reason}, State = #state{socket = Sock}) ->
-    lager:notice("[sdlan_channel] tcp_error: ~p", [Reason]),
-    {stop, normal, State};
-
-handle_info({tcp_closed, Sock}, State = #state{socket = Sock}) ->
-    lager:notice("[sdlan_channel] tcp_closed"),
-    {stop, normal, State};
-
-%% 关闭当前通道
-handle_info({stop, Reason}, State) ->
-    {stop, Reason, State};
-
-%% 主机进程挂掉的时候，需要关闭掉链接
-handle_info({'DOWN', _, process, HostPid, Reason}, State = #state{uuid = UUID, host_pid = HostPid}) ->
-    lager:debug("[ws_channel] uuid: ~p, channel will close because host exited with reason: ~p", [UUID, Reason]),
-    {stop, State};
-
-handle_info(Info, State) ->
-    lager:warning("[sdlan_channel] get a unknown message: ~p, channel will closed", [Info]),
-    {noreply, State}.
-
-terminate(Reason, #state{}) ->
-    lager:warning("[sdlan_channel] stop with reason: ~p", [Reason]),
-    ok.
-
-code_change(_OldVsn, State, _Extra) ->
-    {ok, State}.

apps/iot/src/tcp/tcp_server.erl

@@ -1,37 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author anlicheng
-%%% @copyright (C) 2025, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 08. 5月 2025 12:58
-%%%-------------------------------------------------------------------
--module(tcp_server).
--author("anlicheng").
-
-%% API
--export([start/0]).
-
-%% 启动tcp服务
-start() ->
-    {ok, Props} = application:get_env(iot, tcp_server),
-    Acceptors = proplists:get_value(acceptors, Props, 50),
-    MaxConnections = proplists:get_value(max_connections, Props, 10240),
-    Backlog = proplists:get_value(backlog, Props, 1024),
-    Port = proplists:get_value(port, Props),
-
-    TransOpts = [
-        {tcp_options, [
-            binary,
-            {reuseaddr, true},
-            {active, false},
-            {packet, 4},
-            {nodelay, false},
-            {backlog, Backlog}
-        ]},
-        {acceptors, Acceptors},
-        {max_connections, MaxConnections}
-    ],
-    {ok, _} = esockd:open('iot/tcp_server', Port, TransOpts, {tcp_channel, start_link, []}),
-
-    lager:debug("[iot_app] the tcp server start at: ~p", [Port]).

apps/iot/src/transport/ssl_channel.erl

@@ -0,0 +1,369 @@
+%%%-------------------------------------------------------------------
+%%% @author licheng5
+%%% @copyright (C) 2021, <COMPANY>
+%%% @doc
+%%%
+%%% @end
+%%% Created : 11. 1月 2021 上午12:17
+%%%-------------------------------------------------------------------
+-module(ssl_channel).
+-author("licheng5").
+-behaviour(ranch_protocol).
+
+-define(INFLIGHT_TIMEOUT, 60000).
+-define(SSL_IDLE_TIMEOUT, 120000).
+
+%% API
+-export([pub/4, container_call/4, cancel_command_call/2]).
+
+-export([start_link/3, stop/2]).
+%% gen_server callbacks
+-export([init/3, handle_call/3, handle_cast/2, handle_info/2, code_change/3, terminate/2]).
+
+-record(state, {
+    transport,
+    socket,
+
+    uuid :: undefined | binary(),
+    is_authed = false :: boolean(),
+    %% 用户进程id
+    host_pid = undefined,
+
+    %% iot 发起的 command 与 command_response 的对应关系
+    inflight = #{},
+
+    idle_timer_ref :: undefined | reference()
+}).
+
+-record(inflight_command, {
+    receiver_pid :: undefined | pid(),
+    timer_ref :: reference()
+}).
+
+-type request_ref() :: binary().
+
+%% 向通道中写入消息
+-spec pub(Pid :: pid(), Topic :: binary(), Qos :: integer(), Content :: binary()) -> ok.
+pub(Pid, Topic, Qos, Content) when is_pid(Pid), is_binary(Topic), is_integer(Qos), is_binary(Content) ->
+    gen_server:cast(Pid, {pub, Topic, Qos, Content}).
+
+-spec container_call(Pid :: pid(), ReceiverPid :: pid(), Ref :: request_ref(), Request :: map()) -> ok.
+container_call(Pid, ReceiverPid, Ref, Request) when is_pid(Pid), is_pid(ReceiverPid), is_binary(Ref), is_map(Request) ->
+    gen_server:cast(Pid, {command_call, ReceiverPid, Ref,  {container, Request}}),
+    ok.
+
+-spec cancel_command_call(Pid :: pid(), Ref :: request_ref()) -> ok.
+cancel_command_call(Pid, Ref) when is_pid(Pid), is_binary(Ref) ->
+    gen_server:call(Pid, {cancel_command_call, Ref}).
+
+%% 关闭方法
+-spec stop(Pid :: undefined | pid(), Reason :: any()) -> ok.
+stop(undefined, _Reason) ->
+    ok;
+stop(Pid, Reason) when is_pid(Pid) ->
+    gen_server:stop(Pid, Reason, 5000).
+
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%% 逻辑处理方法
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+
+start_link(Ref, Transport, Opts) ->
+    {ok, proc_lib:spawn_link(?MODULE, init, [Ref, Transport, Opts])}.
+
+init(Ref, Transport, _Opts = []) ->
+    ok = iot_log:set_metadata(),
+    {ok, Socket} = ranch:handshake(Ref),
+    logger:debug("[ssl_channel] get a new connection: ~p", [Socket]),
+    Transport:setopts(Socket, [binary, {active, true}, {packet, 4}]),
+    IdleTimerRef = start_idle_timer(),
+    gen_server:enter_loop(?MODULE, [], #state{transport = Transport, socket = Socket, idle_timer_ref = IdleTimerRef}).
+
+handle_call({cancel_command_call, Ref}, _From, State = #state{inflight = Inflight}) ->
+    case maps:take(Ref, Inflight) of
+        {#inflight_command{timer_ref = TimerRef}, NInflight} ->
+            erlang:cancel_timer(TimerRef),
+            {reply, ok, State#state{inflight = NInflight}};
+        error ->
+            {reply, ok, State}
+    end;
+handle_call(_Request, _From, State) ->
+    {reply, ok, State}.
+
+%% 发送消息, 基于pub/sub机制
+handle_cast({pub, Topic, Qos, Content}, State = #state{transport = Transport, socket = Socket}) ->
+    Packet = term_to_binary({<<"message">>, {<<"pub">>, #{<<"topic">> => Topic, <<"qos">> => Qos, <<"content">> => Content}}}),
+    case Transport:send(Socket, Packet) of
+        ok ->
+            {noreply, State};
+        {error, Reason} ->
+            logger:warning("[ssl_channel] send pub failed, reason: ~p", [Reason]),
+            {stop, {send_failed, Reason}, State}
+    end;
+
+%% iot 请求 efka 时使用 command/command_response 语义。
+handle_cast({command_call, ReceiverPid, Ref, Body}, State = #state{transport = Transport, socket = Socket, inflight = Inflight}) ->
+    Packet = term_to_binary({<<"command">>, Ref, encode_command_body(Body)}),
+    case Transport:send(Socket, Packet) of
+        ok ->
+            TimerRef = erlang:start_timer(?INFLIGHT_TIMEOUT, self(), {command_timeout, Ref}),
+            CommandInfo = #inflight_command{receiver_pid = ReceiverPid, timer_ref = TimerRef},
+            {noreply, State#state{inflight = maps:put(Ref, CommandInfo, Inflight)}};
+        {error, Reason} ->
+            logger:warning("[ssl_channel] send command failed, ref: ~p, reason: ~p", [Ref, Reason]),
+            deliver_command_error(ReceiverPid, Ref, {send_failed, Reason}),
+            {stop, {send_failed, Reason}, State}
+    end.
+
+handle_info({timeout, TimerRef, {command_timeout, Ref}}, State = #state{inflight = Inflight}) ->
+    case maps:get(Ref, Inflight, undefined) of
+        #inflight_command{receiver_pid = ReceiverPid, timer_ref = TimerRef} ->
+            logger:warning("[ws_channel] command timeout, ref: ~p", [Ref]),
+            reply_command_timeout(ReceiverPid, Ref),
+            {noreply, State#state{inflight = maps:remove(Ref, Inflight)}};
+        _ ->
+            {noreply, State}
+    end;
+
+handle_info({timeout, TimerRef, ssl_idle_timeout}, State = #state{idle_timer_ref = TimerRef}) ->
+    logger:notice("[ssl_channel] ssl channel idle timeout"),
+    {stop, ssl_idle_timeout, State#state{idle_timer_ref = undefined}};
+handle_info({timeout, _TimerRef, ssl_idle_timeout}, State) ->
+    {noreply, State};
+
+%% 关闭当前通道
+handle_info({stop, Reason}, State) ->
+    {stop, Reason, State};
+
+%% 主机进程挂掉的时候，需要关闭掉链接
+handle_info({'DOWN', _, process, HostPid, Reason}, State = #state{uuid = UUID, host_pid = HostPid}) ->
+    logger:debug("[ws_channel] uuid: ~p, channel will close because host exited with reason: ~p", [UUID, Reason]),
+    {stop, Reason, State};
+
+handle_info({ssl, Socket, PacketBin}, State = #state{socket = Socket}) when is_binary(PacketBin) ->
+    State1 = reset_idle_timer(State),
+    try binary_to_term(PacketBin, [safe]) of
+        {<<"request">>, Ref, Body} ->
+            handle_request_frame(Ref, Body, State1);
+        {<<"message">>, Body} ->
+            handle_message_frame(Body, State1);
+        {<<"command_response">>, Ref, Response} ->
+            handle_command_response_frame(Ref, Response, State1);
+        Other ->
+            logger:warning("[ssl_channel] unsupported packet: ~p", [Other]),
+            {stop, bad_packet, State1}
+    catch error:Error ->
+        logger:warning("[ssl_channel] binary_to_term get error: ~p, packet_size: ~p", [Error, byte_size(PacketBin)]),
+        {stop, bad_packet, State1}
+    end;
+
+handle_info({ssl_closed, Socket}, State = #state{socket = Socket}) ->
+    logger:notice("[ssl_channel] ssl socket closed"),
+    {stop, normal, State};
+
+handle_info({ssl_error, Socket, Reason}, State = #state{socket = Socket}) ->
+    logger:notice("[ssl_channel] ssl socket error: ~p", [Reason]),
+    {stop, normal, State};
+
+handle_info({ssl_passive, Socket}, State = #state{transport = Transport, socket = Socket}) ->
+    ok = Transport:setopts(Socket, [{active, true}]),
+    {noreply, State};
+
+handle_info(Info, State) ->
+    logger:warning("[ssl_channel] get a unknown message: ~p, state: ~p", [Info, State]),
+    {noreply, State}.
+
+terminate(Reason, #state{inflight = Inflight, transport = Transport, socket = Socket, idle_timer_ref = IdleTimerRef}) ->
+    cancel_timer(IdleTimerRef),
+    maps:foreach(fun(Ref, CommandInfo) -> reply_command_closed(Ref, CommandInfo, Reason) end, Inflight),
+    Transport:close(Socket),
+    logger:warning("[ssl_channel] stop with reason: ~p", [Reason]),
+    ok.
+
+code_change(_OldVsn, State, _Extra) ->
+    {ok, State}.
+
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%%%% helper methods
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+
+-spec handle_request_frame(request_ref(), tuple(), #state{}) -> {noreply, #state{}} | {stop, term(), #state{}}.
+handle_request_frame(Ref, {<<"auth_request">>, #{<<"uuid">> := UUID}}, State = #state{is_authed = true}) ->
+    logger:warning("[ws_channel] repeated auth request, ref: ~p, uuid: ~p, close channel", [Ref, UUID]),
+    {stop, repeated_auth, State};
+handle_request_frame(Ref, {<<"auth_request">>, #{<<"uuid">> := UUID, <<"token">> := Token, <<"timestamp">> := Timestamp}}, State = #state{transport = Transport, socket = Socket}) ->
+    maybe
+        ok ?= auth(Token, UUID, Timestamp),
+        {ok, HostPid} ?= iot_host:lookup_pid(UUID),
+        ok ?= iot_host:attach_channel(HostPid, self()),
+        erlang:monitor(process, HostPid),
+        ok = send_reply_frame(Transport, Socket, Ref, {<<"auth_response">>, <<"ok">>}),
+        logger:debug("[ws_channel] auth uuid: ~p", [UUID]),
+        {noreply, State#state{uuid = UUID, is_authed = true, host_pid = HostPid}}
+    else
+        {error, Reason} ->
+            logger:warning("[ws_channel] uuid: ~p, auth failed with reason: ~p", [UUID, Reason]),
+            case send_reply_frame(Transport, Socket, Ref, {<<"auth_response">>, {<<"error">>, {<<"failed">>, Reason}}}) of
+                ok ->
+                    {stop, Reason, State};
+                {error, SendReason} ->
+                    logger:warning("[ws_channel] uuid: ~p, send auth failed response failed: ~p", [UUID, SendReason]),
+                    {stop, {send_failed, SendReason}, State}
+            end
+    end;
+handle_request_frame(Ref, {<<"container">>, ContainerCommand}, State) ->
+    logger:warning("[ws_channel] unsupported request message type: container, ref: ~p, command: ~p", [Ref, ContainerCommand]),
+    {stop, normal, State};
+handle_request_frame(Ref, Body, State) ->
+    logger:warning("[ws_channel] unsupported request body, ref: ~p, body: ~p", [Ref, Body]),
+    {stop, normal, State}.
+
+-spec handle_message_frame(term(), #state{}) ->
+    {noreply, #state{}} | {stop, term(), #state{}}.
+handle_message_frame(<<"ping">>, State = #state{transport = Transport, socket = Socket}) ->
+    Packet = term_to_binary({<<"message">>, <<"pong">>}),
+    case Transport:send(Socket, Packet) of
+        ok ->
+            {noreply, State};
+        {error, Reason} ->
+            logger:warning("[ssl_channel] send pong failed, reason: ~p", [Reason]),
+            {stop, {send_failed, Reason}, State}
+    end;
+handle_message_frame({<<"data">>, #{<<"route_key">> := RouteKey, <<"metric">> := Metric}}, State = #state{host_pid = HostPid}) when is_pid(HostPid) ->
+    iot_host:handle(HostPid, {data, RouteKey, Metric}),
+    {noreply, State};
+handle_message_frame({<<"task_event">>, Event}, State = #state{uuid = UUID, host_pid = HostPid}) when is_binary(UUID), is_pid(HostPid) ->
+    handle_event_stream_frame(UUID, Event),
+    {noreply, State};
+handle_message_frame(Body, State) ->
+    logger:warning("[ssl_channel] unsupported message body: ~p", [Body]),
+    {noreply, State}.
+
+-spec handle_event_stream_frame(binary(), map()) -> ok.
+handle_event_stream_frame(UUID, #{<<"task_id">> := TaskId, <<"type">> := <<"close">>, <<"stream">> := Reason}) ->
+    logger:debug("[ssl_channel] get uuid: ~p, task_id: ~p, close with reason: ~p", [UUID, TaskId, Reason]),
+    iot_container_task_sup:close(UUID, TaskId, Reason);
+handle_event_stream_frame(UUID, #{<<"task_id">> := TaskId, <<"type">> := Type, <<"stream">> := Stream}) ->
+    logger:debug("[ssl_channel] get uuid: ~p, task_id: ~p, type: ~ts, stream: ~ts", [UUID, TaskId, Type, Stream]),
+    iot_container_task_sup:stream(UUID, TaskId, Type, Stream);
+handle_event_stream_frame(UUID, Event) ->
+    logger:warning("[ssl_channel] invalid task_event, uuid: ~p, event: ~p", [UUID, Event]),
+    ok.
+
+-spec handle_command_response_frame(request_ref(), tuple(), #state{}) ->
+    {noreply, #state{}}.
+handle_command_response_frame(Ref, Reply, State = #state{inflight = Inflight}) when is_binary(Ref) ->
+    case maps:take(Ref, Inflight) of
+        error ->
+            {noreply, State};
+        {#inflight_command{receiver_pid = ReceiverPid, timer_ref = TimerRef}, NInflight} ->
+            erlang:cancel_timer(TimerRef),
+            deliver_command_response(ReceiverPid, Ref, Reply),
+            {noreply, State#state{inflight = NInflight}}
+    end;
+handle_command_response_frame(Ref, Reply, State) ->
+    logger:warning("[ws_channel] unexpected command_response frame, ref: ~p, reply: ~p", [Ref, Reply]),
+    {noreply, State}.
+
+-spec send_reply_frame(module(), any(), request_ref(), tuple()) -> ok | {error, term()}.
+send_reply_frame(Transport, Socket, Ref, Reply) ->
+    Packet = term_to_binary({<<"response">>, Ref, Reply}),
+    Transport:send(Socket, Packet).
+
+-spec decode_command_response({binary(), term()} | tuple()) ->
+    ok | {ok, term()} | {error, term()}.
+decode_command_response({<<"container">>, <<"ok">>}) ->
+    ok;
+decode_command_response({<<"container">>, {<<"ok">>, Result}}) ->
+    {ok, Result};
+decode_command_response({<<"container">>, {<<"error">>, Reason}}) ->
+    {error, Reason};
+decode_command_response(_Reply) ->
+    {error, invalid_response}.
+
+-spec deliver_command_response(undefined | pid(), request_ref(), tuple()) -> ok.
+deliver_command_response(undefined, _Ref, _Reply) ->
+    ok;
+deliver_command_response(ReceiverPid, Ref, Reply) when is_pid(ReceiverPid) ->
+    case is_process_alive(ReceiverPid) of
+        true ->
+            ReceiverPid ! {command_reply, Ref, decode_command_response(Reply)};
+        false ->
+            logger:warning("[ws_channel] get command_response: ~p, ref: ~p, but receiver_pid is deaded", [Reply, Ref])
+    end.
+
+-spec deliver_command_error(undefined | pid(), request_ref(), term()) -> ok.
+deliver_command_error(ReceiverPid, Ref, Reason) when is_pid(ReceiverPid) ->
+    ReceiverPid ! {command_reply, Ref, {error, Reason}},
+    ok;
+deliver_command_error(_ReceiverPid, _Ref, _Reason) ->
+    ok.
+
+-spec reply_command_timeout(undefined | pid(), request_ref()) -> ok.
+reply_command_timeout(ReceiverPid, Ref) when is_pid(ReceiverPid) ->
+    ReceiverPid ! {command_reply, Ref, {error, timeout}},
+    ok;
+reply_command_timeout(_ReceiverPid, _Ref) ->
+    ok.
+
+-spec reply_command_closed(request_ref(), #inflight_command{}, term()) -> ok.
+reply_command_closed(Ref, #inflight_command{receiver_pid = ReceiverPid}, Reason) when is_pid(ReceiverPid) ->
+    ReceiverPid ! {command_reply, Ref, {error, {channel_closed, Reason}}},
+    ok;
+reply_command_closed(_Ref, _CommandInfo, _Reason) ->
+    ok.
+
+%% 检测token是否是合法值
+-spec auth(Token :: binary(), UUID :: binary(), Timestamp :: integer()) ->
+    ok | {error, Reason :: binary()}.
+auth(Token, UUID, Timestamp) when is_binary(Token), is_binary(UUID), is_integer(Timestamp) ->
+    %% 1分钟内有效
+    Now = iot_util:current_time(),
+    case Timestamp =< Now andalso Now - Timestamp =< 60 of
+        true ->
+            case efka_client_store:auth(UUID, Token) of
+                true ->
+                    ok;
+                false ->
+                    {error, <<"invalid token">>}
+            end;
+        false ->
+            {error, <<"invalid timestamp">>}
+    end.
+
+-spec encode_command_body({container, map()}) -> {binary(), map()}.
+encode_command_body({container, Request}) ->
+    {<<"container">>, safe_term(Request)}.
+
+-spec safe_term(term()) -> term().
+safe_term(true) ->
+    true;
+safe_term(false) ->
+    false;
+safe_term(undefined) ->
+    undefined;
+safe_term(Value) when is_atom(Value) ->
+    atom_to_binary(Value, utf8);
+safe_term(Value) when is_map(Value) ->
+    maps:from_list([{safe_term(K), safe_term(V)} || {K, V} <- maps:to_list(Value)]);
+safe_term(Value) when is_list(Value) ->
+    [safe_term(Item) || Item <- Value];
+safe_term(Value) when is_tuple(Value) ->
+    list_to_tuple([safe_term(Item) || Item <- tuple_to_list(Value)]);
+safe_term(Value) ->
+    Value.
+
+-spec start_idle_timer() -> reference().
+start_idle_timer() ->
+    erlang:start_timer(?SSL_IDLE_TIMEOUT, self(), ssl_idle_timeout).
+
+-spec reset_idle_timer(#state{}) -> #state{}.
+reset_idle_timer(State = #state{idle_timer_ref = TimerRef}) ->
+    cancel_timer(TimerRef),
+    State#state{idle_timer_ref = start_idle_timer()}.
+
+-spec cancel_timer(undefined | reference()) -> ok.
+cancel_timer(undefined) ->
+    ok;
+cancel_timer(TimerRef) ->
+    _ = erlang:cancel_timer(TimerRef),
+    ok.

apps/iot/src/transport/udp_server.erl

@@ -1,37 +1,36 @@
 %%%-------------------------------------------------------------------
 %%% @author anlicheng
-%%% @copyright (C) 2023, <COMPANY>
+%%% @copyright (C) 2026, <COMPANY>
 %%% @doc
 %%%
 %%% @end
-%%% Created : 24. 12月 2023 15:42
+%%% Created : 09. 5月 2026 18:02
 %%%-------------------------------------------------------------------
--module(iot_api).
+-module(udp_server).
 -author("anlicheng").
 
 -behaviour(gen_server).
 
+-define(HEARTBEAT_VERSION, 1).
+-define(HEARTBEAT_NONCE_BYTES, 16).
+-define(HEARTBEAT_MAC_BYTES, 32).
+
 %% API
 -export([start_link/0]).
--export([ai_event/1]).
 
 %% gen_server callbacks
 -export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
 
 -define(SERVER, ?MODULE).
--define(API_TOKEN, <<"wv6fGyBhl*7@AsD9">>).
 
 -record(state, {
-
+    socket :: gen_udp:socket()
 }).
 
 %%%===================================================================
 %%% API
 %%%===================================================================
 
-ai_event(Id) when is_integer(Id) ->
-   gen_server:cast(?MODULE, {ai_event, Id}).
-
 %% @doc Spawns the server and registers the local name (unique)
 -spec(start_link() ->
     {ok, Pid :: pid()} | ignore | {error, Reason :: term()}).
@@ -48,7 +47,12 @@ start_link() ->
     {ok, State :: #state{}} | {ok, State :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term()} | ignore).
 init([]) ->
-    {ok, #state{}}.
+    ok = iot_log:set_metadata(),
+    {ok, UdpServerProps} = application:get_env(iot, udp_server),
+    Port = proplists:get_value(port, UdpServerProps),
+    logger:debug("[udp_server] start at port: ~p", [Port]),
+    {ok, Socket} = gen_udp:open(Port, [binary, {active, true}]),
+    {ok, #state{socket = Socket}}.
 
 %% @private
 %% @doc Handling call messages
@@ -69,33 +73,7 @@ handle_call(_Request, _From, State = #state{}) ->
     {noreply, NewState :: #state{}} |
     {noreply, NewState :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term(), NewState :: #state{}}).
-handle_cast({ai_event, Id}, State = #state{}) ->
-    spawn_monitor(fun() ->
-        Token = iot_util:md5(<<?API_TOKEN/binary, (integer_to_binary(Id))/binary, ?API_TOKEN/binary>>),
-        {ok, Url} = application:get_env(iot, api_url),
-
-        Headers = [
-            {<<"content-type">>, <<"application/json">>}
-        ],
-        ReqData = #{
-            <<"token">> => Token,
-            <<"id">> => Id
-        },
-        Body = iolist_to_binary(jiffy:encode(ReqData, [force_utf8])),
-        case hackney:request(post, Url, Headers, Body, [{pool, false}]) of
-            {ok, 200, _, ClientRef} ->
-                {ok, RespBody} = hackney:body(ClientRef),
-                lager:debug("[iot_api] send body: ~p, get error is: ~p", [Body, RespBody]),
-                hackney:close(ClientRef);
-            {ok, HttpCode, _, ClientRef} ->
-                {ok, RespBody} = hackney:body(ClientRef),
-                hackney:close(ClientRef),
-                lager:warning("[iot_api] send body: ~p, get error is: ~p", [Body, {HttpCode, RespBody}]);
-            {error, Reason} ->
-                lager:warning("[iot_api] send body: ~p, get error is: ~p", [Body, Reason])
-        end
-                  end),
-
+handle_cast(_Request, State = #state{}) ->
     {noreply, State}.
 
 %% @private
@@ -104,15 +82,11 @@ handle_cast({ai_event, Id}, State = #state{}) ->
     {noreply, NewState :: #state{}} |
     {noreply, NewState :: #state{}, timeout() | hibernate} |
     {stop, Reason :: term(), NewState :: #state{}}).
-%% Task进程挂掉
-handle_info({'DOWN', _MRef, process, _Pid, normal}, State) ->
+handle_info({udp, Socket, Ip, Port, Packet}, State = #state{socket = Socket}) ->
+    handle_heartbeat_packet(Packet, {Ip, Port}),
     {noreply, State};
-
-handle_info({'DOWN', _MRef, process, _Pid, Reason}, State) ->
-    lager:notice("[iot_api] task process down with reason: ~p", [Reason]),
-    {noreply, State};
-
-handle_info(_Info, State = #state{}) ->
+handle_info(Info, State = #state{}) ->
+    logger:warning("[udp_server] ignore unknown info: ~p", [Info]),
     {noreply, State}.
 
 %% @private
@@ -122,7 +96,8 @@ handle_info(_Info, State = #state{}) ->
 %% with Reason. The return value is ignored.
 -spec(terminate(Reason :: (normal | shutdown | {shutdown, term()} | term()),
     State :: #state{}) -> term()).
-terminate(_Reason, _State = #state{}) ->
+terminate(_Reason, _State = #state{socket = Socket}) ->
+    gen_udp:close(Socket),
     ok.
 
 %% @private
@@ -136,3 +111,31 @@ code_change(_OldVsn, State = #state{}, _Extra) ->
 %%%===================================================================
 %%% Internal functions
 %%%===================================================================
+
+-spec handle_heartbeat_packet(binary(), term()) -> ok.
+handle_heartbeat_packet(Packet, Peer) when is_binary(Packet) ->
+    case decode_heartbeat_packet(Packet) of
+        {ok, UUID, Timestamp, Payload, Mac} ->
+            case efka_client_store:verify_heartbeat(UUID, Timestamp, Payload, Mac) of
+                true ->
+                    Pid = iot_host:get_pid(UUID),
+                    iot_host:heartbeat(Pid);
+                false ->
+                    logger:warning("[udp_server] ignore invalid heartbeat auth from peer: ~p, uuid: ~p", [Peer, UUID]),
+                    ok
+            end;
+        error ->
+            logger:warning("[udp_server] ignore invalid heartbeat packet from peer: ~p, packet: ~p", [Peer, Packet]),
+            ok
+    end.
+
+-spec decode_heartbeat_packet(binary()) ->
+    {ok, UUID :: binary(), Timestamp :: integer(), Payload :: binary(), Mac :: binary()} | error.
+decode_heartbeat_packet(Packet = <<?HEARTBEAT_VERSION:8, UUIDLen:16, UUID:UUIDLen/binary,
+    Timestamp:64/unsigned-big, _Nonce:?HEARTBEAT_NONCE_BYTES/binary, _Mac:?HEARTBEAT_MAC_BYTES/binary>>)
+    when UUIDLen > 0 ->
+    PayloadSize = byte_size(Packet) - ?HEARTBEAT_MAC_BYTES,
+    <<Payload:PayloadSize/binary, Mac:?HEARTBEAT_MAC_BYTES/binary>> = Packet,
+    {ok, UUID, Timestamp, Payload, Mac};
+decode_heartbeat_packet(_Packet) ->
+    error.

apps/iot/src/util/http_client.erl

@@ -1,34 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author anlicheng
-%%% @copyright (C) 2025, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 18. 8月 2025 18:40
-%%%-------------------------------------------------------------------
--module(http_client).
--author("anlicheng").
-
-%% API
--export([post/3]).
-
-%% Headers = [
-%%   {<<"content-type">>, <<"application/json">>}
-%% ]
--spec post(Url :: string(), Headers :: list(), Body :: binary()) -> {ok, RespBody :: binary()} | {error, Reason :: any()}.
-post(Url, Headers, Body) when is_list(Url), is_list(Headers), is_binary(Body) ->
-    case hackney:request(post, Url, Headers, Body, [{pool, false}]) of
-        {ok, 200, _, ClientRef} ->
-            {ok, RespBody} = hackney:body(ClientRef),
-            lager:debug("[iot_api] send body: ~p, get error is: ~p", [Body, RespBody]),
-            hackney:close(ClientRef),
-            {ok, RespBody};
-        {ok, HttpCode, _, ClientRef} ->
-            {ok, RespBody} = hackney:body(ClientRef),
-            hackney:close(ClientRef),
-            lager:warning("[iot_api] send body: ~p, get error is: ~p", [Body, {HttpCode, RespBody}]),
-            {error, {HttpCode, RespBody}};
-        {error, Reason} ->
-            lager:warning("[iot_api] send body: ~p, get error is: ~p", [Body, Reason]),
-            {error, Reason}
-    end.

backup/endpoint_mnesia.erl

@@ -1,97 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author anlicheng
-%%% @copyright (C) 2025, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 12. 8月 2025 15:12
-%%%-------------------------------------------------------------------
--module(endpoint_mnesia).
--author("aresei").
--include("endpoint.hrl").
--include_lib("stdlib/include/qlc.hrl").
-
--define(TAB, endpoint).
-
-%% API
--export([create_table/0]).
--export([insert/1, delete/1, check_name/1]).
--export([get_endpoint/1]).
--export([as_map/1]).
-
-create_table() ->
-    %% id生成器
-    mnesia:create_table(endpoint, [
-        {attributes, record_info(fields, endpoint)},
-        {record_name, endpoint},
-        {disc_copies, [node()]},
-        {type, ordered_set}
-    ]).
-
--spec check_name(Name :: binary()) -> boolean() | {error, Reason :: any()}.
-check_name(Name) when is_binary(Name) ->
-    Fun = fun() ->
-        Q = qlc:q([E || E <- mnesia:table(?TAB), E#endpoint.name =:= Name]),
-        case qlc:e(Q) of
-            [] ->
-                false;
-            [_|_] ->
-                true
-        end
-          end,
-    case mnesia:transaction(Fun) of
-        {'atomic', Res} ->
-            Res;
-        {'aborted', Reason} ->
-            {error, Reason}
-    end.
-
--spec get_endpoint(Id :: integer()) -> error | {ok, Endpoint :: #endpoint{}}.
-get_endpoint(Id) when is_integer(Id) ->
-    case mnesia:dirty_read(?TAB, Id) of
-        [] ->
-            error;
-        [Endpoint | _] ->
-            {ok, Endpoint}
-    end.
-
--spec insert(Endpoint :: #endpoint{}) -> ok | {error, Reason :: term()}.
-insert(Endpoint = #endpoint{}) ->
-    case mnesia:transaction(fun() -> mnesia:write(?TAB, Endpoint, write) end) of
-        {'atomic', ok} ->
-            ok;
-        {'aborted', Reason} ->
-            {error, Reason}
-    end.
-
--spec delete(Id :: integer()) -> ok | {error, Reason :: any()}.
-delete(Id) when is_integer(Id) ->
-    case mnesia:transaction(fun() -> mnesia:delete(?TAB, Id, write) end) of
-        {'atomic', ok} ->
-            ok;
-        {'aborted', Reason} ->
-            {error, Reason}
-    end.
-
--spec as_map(Endpoint :: #endpoint{}) -> map().
-as_map(#endpoint{id = Id, name = Name, title = Title, config = Config, updated_at = UpdateTs, created_at = CreateTs}) ->
-    {ConfigKey, ConfigMap} =
-        case Config of
-            #http_endpoint{url = Url, pool_size = PoolSize} ->
-                {<<"http">>, #{<<"url">> => Url, <<"pool_size">> => PoolSize}};
-            #mqtt_endpoint{host = Host, port = Port, client_id = ClientId, username = Username, password = Password, topic = Topic, qos = Qos} ->
-                {<<"mqtt">>, #{<<"host">> => Host, <<"port">> => Port, <<"client_id">> => ClientId, <<"username">> => Username, <<"password">> => Password, <<"topic">> => Topic, <<"qos">> => Qos}};
-            #kafka_endpoint{username = Username, password = Password, bootstrap_servers = BootstrapServers, topic = Topic} ->
-                {<<"kafka">>, #{<<"username">> => Username, <<"password">> => Password, <<"bootstrap_servers">> => BootstrapServers, <<"topic">> => Topic}};
-            #mysql_endpoint{host = Host, port = Port, username = Username, password = Password, database = Database, table_name = TableName} ->
-                {<<"mysql">>, #{<<"host">> => Host, <<"port">> => Port, <<"username">> => Username, <<"password">> => Password, <<"database">> => Database, <<"table_name">> => TableName}}
-        end,
-
-    Map = #{
-        <<"id">> => Id,
-        <<"name">> => Name,
-        <<"title">> => Title,
-        <<"update_ts">> => UpdateTs,
-        <<"create_ts">> => CreateTs
-    },
-    Map#{ConfigKey => ConfigMap}.

backup/endpoint_mysql.erl

@@ -1,214 +0,0 @@
-
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 06. 7月 2023 12:02
-%%%-------------------------------------------------------------------
--module(endpoint_mysql).
-
--include("endpoint.hrl").
--behaviour(gen_server).
-
-%% API
--export([start_link/3]).
-
-%% gen_server callbacks
--export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
-
-%% 消息重发间隔
--define(RETRY_INTERVAL, 5000).
-
--define(DISCONNECTED, disconnected).
--define(CONNECTED, connected).
-
--record(state, {
-    endpoint :: #endpoint{},
-    buffer :: endpoint_buffer:buffer(),
-    pool_pid :: undefined | pid(),
-
-    status = ?DISCONNECTED
-}).
-
-%%%===================================================================
-%%% API
-%%%===================================================================
-
-%% @doc Creates a gen_statem process which calls Module:init/1 to
-%% initialize. To ensure a synchronized start-up procedure, this
-%% function does not return until Module:init/1 has returned.
-start_link(LocalName, AliasName, Endpoint = #endpoint{}) when is_atom(LocalName), is_atom(AliasName) ->
-    gen_statem:start_link({local, LocalName}, ?MODULE, [AliasName, Endpoint], []).
-
-%%%===================================================================
-%%% gen_statem callbacks
-%%%===================================================================
-
-%% @private
-%% @doc Whenever a gen_statem is started using gen_statem:start/[3,4] or
-%% gen_statem:start_link/[3,4], this function is called by the new
-%% process to initialize.
-init([AliasName, Endpoint]) ->
-    iot_name_server:register(AliasName, self()),
-    erlang:process_flag(trap_exit, true),
-    %% 创建转发器, 避免阻塞当前进程的创建，因此采用了延时初始化的机制
-    erlang:start_timer(0, self(), create_postman),
-    %% 初始化存储
-    Buffer = endpoint_buffer:new(Endpoint, 10),
-
-    {ok, #state{endpoint = Endpoint, buffer = Buffer, status = ?DISCONNECTED}}.
-
-%% @private
-%% @doc Handling call messages
--spec(handle_call(Request :: term(), From :: {pid(), Tag :: term()},
-    State :: #state{}) ->
-    {reply, Reply :: term(), NewState :: #state{}} |
-    {reply, Reply :: term(), NewState :: #state{}, timeout() | hibernate} |
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), Reply :: term(), NewState :: #state{}} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_call(get_stat, _From, State = #state{buffer = Buffer}) ->
-    Stat = endpoint_buffer:stat(Buffer),
-    {reply, {ok, Stat}, State}.
-
-%% @private
-%% @doc Handling cast messages
--spec(handle_cast(Request :: term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_cast({forward, ServiceId, Format, Metric}, State = #state{buffer = Buffer}) ->
-    NBuffer = endpoint_buffer:append({ServiceId, Format, Metric}, Buffer),
-    {noreply, State#state{buffer = NBuffer}};
-
-handle_cast(cleanup, State = #state{buffer = Buffer}) ->
-    endpoint_buffer:cleanup(Buffer),
-    {noreply, State}.
-
-%% @private
-%% @doc Handling all non call/cast messages
--spec(handle_info(Info :: timeout() | term(), State :: #state{}) ->
-    {noreply, NewState :: #state{}} |
-    {noreply, NewState :: #state{}, timeout() | hibernate} |
-    {stop, Reason :: term(), NewState :: #state{}}).
-handle_info({timeout, _, create_postman}, State = #state{status = ?DISCONNECTED, buffer = Buffer,
-    endpoint = #endpoint{title = Title, config = #mysql_endpoint{host = Host, port = Port, username = Username, password = Password, database = Database}}}) ->
-
-    lager:debug("[iot_endpoint] endpoint: ~p, create postman", [Title]),
-    WorkerArgs = [
-        {host, binary_to_list(Host)},
-        {port, Port},
-        {user, binary_to_list(Username)},
-        {password, binary_to_list(Password)},
-        {keep_alive, true},
-        {database, binary_to_list(Database)},
-        {queries, [<<"set names utf8">>]}
-    ],
-
-    %% 启动工作的线程池
-    PoolSize = 5,
-    case poolboy:start_link([{size, PoolSize}, {max_overflow, PoolSize}, {worker_module, mysql}], WorkerArgs) of
-        {ok, PoolPid} ->
-            NBuffer = endpoint_buffer:trigger_n(Buffer),
-            {noreply, State#state{pool_pid = PoolPid, buffer = NBuffer, status = ?CONNECTED}};
-        ignore ->
-            retry_connect(),
-            {noreply, State};
-        {error, Reason} ->
-            lager:warning("[mqtt_postman] start connect pool, get error: ~p", [Reason]),
-            retry_connect(),
-            {noreply, State}
-    end;
-
-%% 离线时，忽略数据发送逻辑
-handle_info({next_data, _Id, _Tuple}, State = #state{status = ?DISCONNECTED}) ->
-    {noreply, State};
-%% 发送数据到mqtt服务器
-handle_info({next_data, Id, {ServiceId, Metric}}, State = #state{status = ?CONNECTED, pool_pid = PoolPid, buffer = Buffer,
-    endpoint = #endpoint{title = Title, config = #mysql_endpoint{table_name = Table, fields_map = FieldsMap}}}) ->
-
-    case insert_sql(Table, ServiceId, FieldsMap, Metric) of
-        {ok, InsertSql, Values} ->
-            case poolboy:transaction(PoolPid, fun(ConnPid) -> mysql:query(ConnPid, InsertSql, Values) end) of
-                ok ->
-                    NBuffer = endpoint_buffer:ack(Id, Buffer),
-                    {noreply, State#state{buffer = NBuffer}};
-                Error ->
-                    lager:warning("[endpoint_mysql] endpoint: ~p, insert mysql get error: ~p", [Title, Error]),
-                    {noreply, State}
-            end;
-        error ->
-            lager:debug("[endpoint_mysql] endpoint: ~p, make sql error", [Title]),
-            {noreply, State}
-    end;
-
-%% postman进程挂掉时，重新建立新的
-handle_info({'EXIT', PoolPid, Reason}, State = #state{endpoint = #endpoint{title = Title}, pool_pid = PoolPid}) ->
-    lager:warning("[enpoint_mqtt] endpoint: ~p, conn pid exit with reason: ~p", [Title, Reason]),
-    retry_connect(),
-    {noreply, disconnected, State#state{pool_pid = undefined, status = ?DISCONNECTED}};
-
-handle_info(Info, State = #state{status = Status}) ->
-    lager:warning("[iot_endpoint] unknown message: ~p, status: ~p", [Info, Status]),
-    {noreply, State}.
-
-%% @private
-%% @doc This function is called by a gen_server when it is about to
-%% terminate. It should be the opposite of Module:init/1 and do any
-%% necessary cleaning up. When it returns, the gen_server terminates
-%% with Reason. The return value is ignored.
--spec(terminate(Reason :: (normal | shutdown | {shutdown, term()} | term()),
-    State :: #state{}) -> term()).
-terminate(Reason, #state{endpoint = #endpoint{title = Title}, buffer = Buffer}) ->
-    lager:debug("[iot_endpoint] endpoint: ~p, terminate with reason: ~p", [Title, Reason]),
-    endpoint_buffer:cleanup(Buffer),
-    ok.
-
-%% @private
-%% @doc Convert process state when code is changed
--spec(code_change(OldVsn :: term() | {down, term()}, State :: #state{},
-    Extra :: term()) ->
-    {ok, NewState :: #state{}} | {error, Reason :: term()}).
-code_change(_OldVsn, State = #state{}, _Extra) ->
-    {ok, State}.
-
-%%%===================================================================
-%%% Internal functions
-%%%===================================================================
-
-retry_connect() ->
-    erlang:start_timer(?RETRY_INTERVAL, self(), create_postman).
-
--spec insert_sql(Table :: binary(), ServiceId :: binary(), FieldsMap :: map(), Metric :: binary()) ->
-    error | {ok, Sql :: binary(), Values :: list()}.
-insert_sql(Table, ServiceId, FieldsMap, Metric) when is_binary(Table), is_binary(ServiceId), is_binary(Metric) ->
-    case line_format:parse(Metric) of
-        error ->
-            error;
-        {ok, #{<<"measurement">> := Measurement, <<"tags">> := Tags, <<"fields">> := Fields, <<"timestamp">> := Timestamp}} ->
-            Map = maps:merge(Tags, Fields),
-            NMap = Map#{<<"measurement">> => Measurement, <<"timestamp">> => Timestamp},
-            TableFields = lists:flatmap(fun({TableField, F}) ->
-                case maps:find(F, NMap) of
-                    error ->
-                        [];
-                    {ok, Val} ->
-                        [{TableField, Val}]
-                end
-                                        end, maps:to_list(FieldsMap)),
-
-            {Keys, Values} = kvs(TableFields),
-            FieldSql = iolist_to_binary(lists:join(<<", ">>, Keys)),
-            Placeholders = lists:duplicate(length(Keys), <<"?">>),
-            ValuesPlaceholder = iolist_to_binary(lists:join(<<", ">>, Placeholders)),
-
-            {ok, <<"INSERT INTO ", Table/binary, "(", FieldSql/binary, ") VALUES(", ValuesPlaceholder/binary, ")">>, Values}
-    end.
-
--spec kvs(Fields :: list()) -> {Keys :: list(), Values :: list()}.
-kvs(Fields) when is_list(Fields) ->
-    {Keys0, Values0} = lists:foldl(fun({K, V}, {Acc0, Acc1}) -> {[K|Acc0], [V|Acc1]} end, {[], []}, Fields),
-    {lists:reverse(Keys0), lists:reverse(Values0)}.

backup/service_config_model.erl

@@ -1,121 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author aresei
-%%% @copyright (C) 2023, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 04. 7月 2023 12:31
-%%%-------------------------------------------------------------------
--module(service_config_model).
--author("aresei").
--include("iot_tables.hrl").
--include_lib("stdlib/include/qlc.hrl").
-
--define(TAB, service_config).
-
-%% API
--export([create_table/0]).
--export([insert/4, update/4, get_config/1, delete/1]).
--export([as_map/1]).
-
-create_table() ->
-    %% id生成器
-    mnesia:create_table(service_config, [
-        {attributes, record_info(fields, service_config)},
-        {record_name, service_config},
-        {disc_copies, [node()]},
-        {type, ordered_set}
-    ]).
-
--spec insert(ServiceId :: binary(), HostUUID :: binary(), ConfigJson :: binary(), LastEditUser :: integer()) -> ok | {error, Reason :: term()}.
-insert(ServiceId, HostUUID, ConfigJson, LastEditUser) when is_binary(ServiceId), is_binary(HostUUID), is_binary(ConfigJson), is_integer(LastEditUser) ->
-    ServiceConfig = #service_config{
-        service_id = ServiceId,
-        host_uuid = HostUUID,
-        config_json = ConfigJson,
-        last_config_json = <<>>,
-        last_edit_user = LastEditUser,
-        create_ts = iot_util:current_time(),
-        update_ts = iot_util:current_time()
-    },
-    case mnesia:transaction(fun() -> mnesia:write(?TAB, ServiceConfig, write) end) of
-        {'atomic', ok} ->
-            ok;
-        {'aborted', Reason} ->
-            {error, Reason}
-    end.
-
--spec update(ServiceId :: binary(), HostUUID :: binary(), ConfigJson :: binary(), LastEditUser :: integer()) -> ok | {error, Reason :: term()}.
-update(ServiceId, HostUUID, ConfigJson, LastEditUser) when is_binary(ServiceId), is_binary(HostUUID), is_binary(ConfigJson), is_integer(LastEditUser) ->
-    Fun = fun() ->
-        case mnesia:read(?TAB, ServiceId, write) of
-            [] ->
-                ServiceConfig = #service_config{
-                    service_id = ServiceId,
-                    host_uuid = HostUUID,
-                    config_json = ConfigJson,
-                    last_config_json = <<>>,
-                    last_edit_user = LastEditUser,
-                    create_ts = iot_util:current_time(),
-                    update_ts = iot_util:current_time()
-                },
-                mnesia:write(?TAB, ServiceConfig, write);
-            [ServiceConfig0 = #service_config{config_json = OldConfigJson}] ->
-                NServiceConfig = ServiceConfig0#service_config{
-                    config_json = ConfigJson,
-                    last_config_json = OldConfigJson,
-                    last_edit_user = LastEditUser,
-                    update_ts = iot_util:current_time()
-                },
-                mnesia:write(?TAB, NServiceConfig, write)
-        end
-          end,
-
-    case mnesia:transaction(Fun) of
-        {'atomic', ok} ->
-            ok;
-        {'aborted', Reason} ->
-            {error, Reason}
-    end.
-
--spec get_config(ServiceId :: any()) -> error | {ok, Config :: #service_config{}}.
-get_config(ServiceId) when is_binary(ServiceId) ->
-    case mnesia:dirty_read(?TAB, ServiceId) of
-        [] ->
-            error;
-        [Config] ->
-            {ok, Config}
-    end.
-
--spec delete(ServiceId :: binary()) -> ok | {error, Reason :: any()}.
-delete(ServiceId) when is_binary(ServiceId) ->
-    Fun = fun() ->
-        case mnesia:read(?TAB, ServiceId, write) of
-            [] ->
-                ok;
-            [ServiceConfig0 = #service_config{config_json = OldConfigJson}] ->
-                NServiceConfig = ServiceConfig0#service_config{
-                    config_json = <<"">>,
-                    last_config_json = OldConfigJson,
-                    update_ts = iot_util:current_time()
-                },
-                mnesia:write(?TAB, NServiceConfig, write)
-        end
-          end,
-    case mnesia:transaction(Fun) of
-        {'atomic', ok} ->
-            ok;
-        {'aborted', Reason} ->
-            {error, Reason}
-    end.
-
--spec as_map(ServiceConfig :: #service_config{}) -> map().
-as_map(#service_config{service_id = ServiceId, config_json = ConfigJson, last_config_json = LastConfigJson, last_edit_user = LastEditUser, update_ts = UpdateTs, create_ts = CreateTs}) ->
-    #{
-        <<"service_id">> => ServiceId,
-        <<"config_json">> => ConfigJson,
-        <<"last_config_json">> => LastConfigJson,
-        <<"last_edit_user">> => LastEditUser,
-        <<"update_ts">> => UpdateTs,
-        <<"create_ts">> => CreateTs
-    }.

backup/service_handler.erl

@@ -1,176 +0,0 @@
-%%%-------------------------------------------------------------------
-%%% @author licheng5
-%%% @copyright (C) 2020, <COMPANY>
-%%% @doc
-%%%
-%%% @end
-%%% Created : 26. 4月 2020 3:36 下午
-%%%-------------------------------------------------------------------
--module(service_handler).
--author("licheng5").
--include("iot.hrl").
-
-%% API
--export([handle_request/4]).
-
-%% 下发config.json, 微服务接受后，保存服务配置
-handle_request("POST", "/service/push_config", _,
-    #{<<"uuid">> := UUID, <<"service_id">> := ServiceId, <<"last_edit_user">> := LastEditUser, <<"config_json">> := ConfigJson, <<"timeout">> := Timeout0})
-    when is_binary(UUID), is_binary(ServiceId), is_binary(ConfigJson), is_integer(Timeout0) ->
-
-    %% 检查ConfigJson是否是合法的json字符串
-    case iot_util:is_json(ConfigJson) of
-        true ->
-            case iot_host:get_pid(UUID) of
-                undefined ->
-                    {ok, 200, iot_util:json_error(-1, <<"host not found">>)};
-                Pid when is_pid(Pid) ->
-                    Timeout = Timeout0 * 1000,
-                    case iot_host:async_service_config(Pid, ServiceId, ConfigJson, Timeout) of
-                        {ok, Ref} ->
-                            case iot_host:await_reply(Ref, Timeout) of
-                                {ok, Result} ->
-                                    %% 更新配置信息到数据库
-                                    case service_config_model:update(ServiceId, UUID, ConfigJson, LastEditUser) of
-                                        ok ->
-                                            {ok, 200, iot_util:json_data(Result)};
-                                        {error, Reason} ->
-                                            lager:debug("[service_handler] set_config service_id: ~p, get error: ~p", [ServiceId, Reason]),
-                                            {ok, 200, iot_util:json_error(-1, <<"set service config failed">>)}
-                                    end;
-                                {error, Reason} ->
-                                    {ok, 200, iot_util:json_error(-1, Reason)}
-                            end;
-                        {error, Reason} when is_binary(Reason) ->
-                            {ok, 200, iot_util:json_error(-1, Reason)}
-                    end
-            end;
-        false ->
-            {ok, 200, iot_util:json_error(-1, <<"config is invalid json">>)}
-    end;
-
-%% 获取服务配置信息
-handle_request("GET", "/service/get_config", #{<<"service_id">> := ServiceId}, _) when is_binary(ServiceId) ->
-    case service_config_model:get_config(ServiceId) of
-        error ->
-            {ok, 200, iot_util:json_error(-1, <<"service config not found">>)};
-        {ok, Config} ->
-            {ok, 200, iot_util:json_data(service_config_model:as_map(Config))}
-    end;
-
-%% 删除对应的主机信息
-handle_request("POST", "/service/delete_config", _, #{<<"service_id">> := ServiceId}) when is_binary(ServiceId) ->
-    case service_config_model:delete(ServiceId) of
-        ok ->
-            {ok, 200, iot_util:json_data(<<"success">>)};
-        {error, Reason} ->
-            lager:debug("[service_handler] delete config of service_id: ~p, error: ~p", [ServiceId, Reason]),
-            {ok, 200, iot_util:json_error(-1, <<"delete service config errror">>)}
-    end;
-
-%% 部署微服务
-handle_request("POST", "/service/deploy", _, #{<<"uuid">> := UUID, <<"task_id">> := TaskId, <<"service_id">> := ServiceId, <<"tar_url">> := TarUrl})
-    when is_binary(UUID), is_integer(TaskId), is_binary(ServiceId), is_binary(TarUrl) ->
-
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            case iot_host:deploy_service(Pid, TaskId, ServiceId, TarUrl) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, 5000) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-%% 启动服务
-handle_request("POST", "/service/start", _, #{<<"uuid">> := UUID, <<"service_id">> := ServiceId}) when is_binary(UUID), is_binary(ServiceId) ->
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            case iot_host:start_service(Pid, ServiceId) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, 5000) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-%% 停止服务
-handle_request("POST", "/service/stop", _, #{<<"uuid">> := UUID, <<"service_id">> := ServiceId}) when is_binary(UUID), is_binary(ServiceId) ->
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            case iot_host:stop_service(Pid, ServiceId) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, 5000) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-%% 远程调用微服务, 返回值的格式为json
-handle_request("POST", "/service/invoke", _, #{<<"uuid">> := UUID, <<"service_id">> := ServiceId, <<"payload">> := Payload, <<"timeout">> := Timeout0})
-    when is_binary(UUID), is_binary(ServiceId), is_binary(Payload), is_integer(Timeout0) ->
-
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            Timeout = Timeout0 * 1000,
-            case iot_host:invoke_service(Pid, ServiceId, Payload, Timeout) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, Timeout) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-handle_request("POST", "/service/task_log", _, #{<<"uuid">> := UUID, <<"task_id">> := TaskId}) when is_binary(UUID), is_integer(TaskId) ->
-    case iot_host:get_pid(UUID) of
-        undefined ->
-            {ok, 200, iot_util:json_error(404, <<"host not found">>)};
-        Pid when is_pid(Pid) ->
-            case iot_host:task_log(Pid, TaskId) of
-                {ok, Ref} ->
-                    case iot_host:await_reply(Ref, 5000) of
-                        {ok, Result} ->
-                            {ok, 200, iot_util:json_data(Result)};
-                        {error, Reason} ->
-                            {ok, 200, iot_util:json_error(400, Reason)}
-                    end;
-                {error, Reason} when is_binary(Reason) ->
-                    {ok, 200, iot_util:json_error(400, Reason)}
-            end
-    end;
-
-handle_request(_, Path, _, _) ->
-    Path1 = list_to_binary(Path),
-    {ok, 200, iot_util:json_error(-1, <<"url: ", Path1/binary, " not found">>)}.
-
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-%% helper methods
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

config/sys-dev.config

@@ -1,110 +0,0 @@
-[
-  {iot, [
-    {http_server, [
-      {port, 18090},
-      {acceptors, 500},
-      {max_connections, 10240},
-      {backlog, 10240}
-    ]},
-
-    {tcp_server, [
-      {port, 18092},
-      {acceptors, 500},
-      {max_connections, 10240},
-      {backlog, 10240}
-    ]},
-
-    {redis_server, [
-      {port, 16379},
-      {acceptors, 500},
-      {max_connections, 10240},
-      {backlog, 10240}
-    ]},
-
-    {udp_server, [
-      {port, 18080}
-    ]},
-
-    {api_url, "http://39.98.184.67:8800/api/v1/taskLog"},
-
-    %% 目标服务器地址
-    {emqx_server, [
-      {host, {39, 98, 184, 67}},
-      {port, 1883},
-      {tcp_opts, []},
-      {username, "test"},
-      {password, "test1234"},
-      {keepalive, 86400},
-      {retry_interval, 5}
-    ]},
-
-    %% 权限检验时的预埋token
-    {pre_tokens, [
-      {<<"test">>, <<"iot2023">>}
-    ]},
-
-    {pools, [
-      %% mysql连接池配置
-      {mysql_iot,
-        [{size, 10}, {max_overflow, 20}, {worker_module, mysql}],
-        [
-          {host, "47.111.101.3"},
-          {port, 3306},
-          {user, "root"},
-          {connect_mode, synchronous},
-          {keep_alive, true},
-          {password, "r3a-7Qrh#3Q"},
-          {database, "nannong_demo"}
-        ]
-      },
-
-      %% redis连接池
-      {redis_pool,
-        [{size, 10}, {max_overflow, 20}, {worker_module, eredis}],
-        [
-          {host, "127.0.0.1"},
-          {port, 6379},
-          {database, 1}
-        ]
-      }
-    ]}
-
-  ]},
-
-
-  %% 系统日志配置，系统日志为lager, 支持日志按日期自动分割
-  {lager, [
-    {colored, true},
-    %% Whether to write a crash log, and where. Undefined means no crash logger.
-    {crash_log, "trade_hub.crash.log"},
-    %% Maximum size in bytes of events in the crash log - defaults to 65536
-    {crash_log_msg_size, 65536},
-    %% Maximum size of the crash log in bytes, before its rotated, set
-    %% to 0 to disable rotation - default is 0
-    {crash_log_size, 10485760},
-    %% What time to rotate the crash log - default is no time
-    %% rotation. See the README for a description of this format.
-    {crash_log_date, "$D0"},
-    %% Number of rotated crash logs to keep, 0 means keep only the
-    %% current one - default is 0
-    {crash_log_count, 5},
-    %% Whether to redirect error_logger messages into lager - defaults to true
-    {error_logger_redirect, true},
-
-    %% How big the gen_event mailbox can get before it is switched into sync mode
-    {async_threshold, 20},
-    %% Switch back to async mode, when gen_event mailbox size decrease from `async_threshold'
-    %% to async_threshold - async_threshold_window
-    {async_threshold_window, 5},
-
-    {handlers, [
-      %% debug | info | warning | error, 日志级别
-      {lager_console_backend, debug},
-      {lager_file_backend, [{file, "debug.log"}, {level, debug}, {size, 314572800}]},
-      {lager_file_backend, [{file, "notice.log"}, {level, notice}, {size, 314572800}]},
-      {lager_file_backend, [{file, "error.log"}, {level, error}, {size, 314572800}]},
-      {lager_file_backend, [{file, "info.log"}, {level, info}, {size, 314572800}]}
-    ]}
-  ]}
-
-].

config/sys-prod.config

@@ -1,85 +0,0 @@
-[
-  {iot, [
-    {http_server, [
-      {port, 18080},
-      {acceptors, 500},
-      {max_connections, 10240},
-      {backlog, 10240}
-    ]},
-
-    {redis_server, [
-      {port, 16379},
-      {acceptors, 500},
-      {max_connections, 10240},
-      {backlog, 10240}
-    ]},
-
-    {udp_server, [
-      {port, 18080}
-    ]},
-
-    %% 权限检验时的预埋token
-    {pre_tokens, [
-      {<<"test">>, <<"iot2023">>}
-    ]},
-
-    {api_url, "https://lgsiot.njau.edu.cn/api/v1/taskLog"},
-
-    {influxdb, [
-      {host, "172.19.0.4"},
-      {port, 8086},
-      {token, <<"A-ZRjqMK_7NR45lXXEiR7AEtYCd1ETzq9Z61FTMQLb5O4-1hSf8sCrjdPB84e__xsrItKHL3qjJALgbYN-H_VQ==">>}
-    ]},
-
-    {pools, [
-      %% redis连接池
-      {redis_pool,
-        [{size, 10}, {max_overflow, 20}, {worker_module, eredis}],
-        [
-          {host, "172.30.6.175"},
-          {port, 26379},
-          {database, 1}
-        ]
-      }
-
-    ]}
-
-  ]},
-
-
-  %% 系统日志配置，系统日志为lager, 支持日志按日期自动分割
-  {lager, [
-    {colored, true},
-    %% Whether to write a crash log, and where. Undefined means no crash logger.
-    {crash_log, "trade_hub.crash.log"},
-    %% Maximum size in bytes of events in the crash log - defaults to 65536
-    {crash_log_msg_size, 65536},
-    %% Maximum size of the crash log in bytes, before its rotated, set
-    %% to 0 to disable rotation - default is 0
-    {crash_log_size, 10485760},
-    %% What time to rotate the crash log - default is no time
-    %% rotation. See the README for a description of this format.
-    {crash_log_date, "$D0"},
-    %% Number of rotated crash logs to keep, 0 means keep only the
-    %% current one - default is 0
-    {crash_log_count, 5},
-    %% Whether to redirect error_logger messages into lager - defaults to true
-    {error_logger_redirect, true},
-
-    %% How big the gen_event mailbox can get before it is switched into sync mode
-    {async_threshold, 20},
-    %% Switch back to async mode, when gen_event mailbox size decrease from `async_threshold'
-    %% to async_threshold - async_threshold_window
-    {async_threshold_window, 5},
-
-    {handlers, [
-      %% debug | info | warning | error, 日志级别
-      {lager_console_backend, debug},
-      {lager_file_backend, [{file, "debug.log"}, {level, debug}, {size, 314572800}]},
-      {lager_file_backend, [{file, "notice.log"}, {level, notice}, {size, 314572800}]},
-      {lager_file_backend, [{file, "error.log"}, {level, error}, {size, 314572800}]},
-      {lager_file_backend, [{file, "info.log"}, {level, info}, {size, 314572800}]}
-    ]}
-  ]}
-
-].

config/sys.config.src

@@ -0,0 +1,82 @@
+[
+  {iot, [
+    {http_server, [
+      {port, 18090},
+      {acceptors, 5},
+      {max_connections, 1024},
+      {backlog, 1024}
+    ]},
+
+    {ssl_server, [
+      {port, 1443},
+      {acceptors, 5},
+      {max_connections, 1024},
+      {backlog, 1024}
+    ]},
+
+    {udp_server, [
+      {port, 24000}
+    ]},
+
+    {ctrl_server, [
+      {socket_path, "${IOT_CTRL_SOCKET_PATH:-/var/lib/iot/ctl.sock}"},
+      {backlog, ${IOT_CTRL_SOCKET_BACKLOG:-128}}
+    ]},
+
+    {api_url, "${IOT_API_URL}"}
+  ]},
+
+  {endpoint, [
+    %% 支持的协议
+    {endpoints, [
+      {root_dir, "${ENDPOINT_ROOT_DIR}"},
+      {support_protocols, [
+        http
+      ]}
+    ]},
+
+    {endpoint_log, [
+      {path, "${ENDPOINT_LOG_PATH}"},
+      {max_bytes, ${ENDPOINT_LOG_MAX_BYTES:-10485760}},
+      {max_files, ${ENDPOINT_LOG_MAX_FILES:-10}}
+    ]}
+  ]},
+
+  %% 系统日志配置，使用 OTP logger
+  {kernel, [
+    %% 设置 Logger 的 primary log level
+    {logger_level, ${IOT_LOG_LEVEL:-debug}},
+    {logger, [
+      {handler, default, logger_std_h,
+        #{
+          level => ${IOT_CONSOLE_LOG_LEVEL:-debug},
+          filter_default => stop,
+          filters => [
+            {iot_domain, {fun logger_filters:domain/2, {log, sub, [iot]}}},
+            {endpoint_domain, {fun logger_filters:domain/2, {log, sub, [endpoint]}}}
+          ],
+          formatter => {logger_formatter, #{template => [time, " [", level, "] ", msg, "\n"]}}
+        }
+      },
+
+      {handler, disk, logger_disk_log_h,
+        #{
+          level => ${IOT_DISK_LOG_LEVEL:-debug},
+          filter_default => stop,
+          filters => [
+            {iot_domain, {fun logger_filters:domain/2, {log, sub, [iot]}}},
+            {endpoint_domain, {fun logger_filters:domain/2, {log, sub, [endpoint]}}}
+          ],
+          config => #{
+            file => "${IOT_DISK_LOG_FILE:-log/debug.log}",
+            max_no_files => ${IOT_DISK_LOG_MAX_FILES:-10},
+            max_no_bytes => ${IOT_DISK_LOG_MAX_BYTES:-524288000}
+          },
+          formatter => {logger_formatter, #{template => [time, " [", level, "] ", msg, "\n"]}}
+        }
+      }
+
+    ]}
+  ]}
+
+].

config/vm.args.src

@@ -3,15 +3,13 @@
 -setcookie iot_cookie
 
 +K true
-+A30
++A 128
 
--mnesia dir '"/usr/local/var/mnesia/iot"'
+-mnesia dir '"${IOT_MNESIA_DIR}"'
 -mnesia dump_log_write_threshold 50000
 -mnesia dc_dump_limit 40
 
 -sbt db
 
-+K true
-+A 128
 +P 1048576
 +t 10485760