Добавлена оперативная и историческая статистика нод и узлов mnesia #22

2026-05-28 18:59:21 +03:00
parent 29fe0386be
commit f507d08e2c
16 changed files with 753 additions and 201 deletions
@@ -0,0 +1,162 @@
+%%%-------------------------------------------------------------------
+%%% @doc Административный обработчик исторических метрик узлов.
+%%% GET /v1/admin/nodes/metrics – возвращает массив метрик за период.
+%%% @end
+%%%-------------------------------------------------------------------
+-module(admin_handler_node_metrics).
+-behaviour(cowboy_handler).
+-export([init/2]).
+-export([trails/0]).
+
+-include("records.hrl").
+
+%%% cowboy_handler callback
+-spec init(cowboy_req:req(), any()) -> {ok, binary(), cowboy_req:req()}.
+init(Req, _Opts) ->
+  case cowboy_req:method(Req) of
+    <<"GET">> -> get_metrics(Req);
+    _ -> handler_utils:send_error(Req, 405, <<"Method not allowed">>)
+  end.
+
+%%% Swagger metadata
+-spec trails() -> [map()].
+trails() ->
+  [#{path => <<"/v1/admin/nodes/metrics">>,
+    method => <<"GET">>,
+    description => <<"Get historical node metrics">>,
+    tags => [<<"Node Metrics">>],
+    parameters => [
+      #{name => <<"node">>, in => <<"query">>, schema => #{type => string}, description => <<"Node name, default current">>},
+      #{name => <<"from">>, in => <<"query">>, required => true, schema => #{type => string, format => <<"date-time">>}},
+      #{name => <<"to">>, in => <<"query">>, required => true, schema => #{type => string, format => <<"date-time">>}}
+    ],
+    responses => #{
+      200 => #{
+        description => <<"Array of node metrics">>,
+        content => #{<<"application/json">> => #{schema => #{
+          type => array,
+          items => metric_schema()
+        }}}
+      },
+      400 => #{description => <<"Missing required parameters">>},
+      403 => #{description => <<"Admin access required">>}
+    }}
+  ].
+
+metric_schema() ->
+  #{
+    type => object,
+    properties => #{
+      <<"timestamp">> => #{
+        type => string,
+        format => <<"date-time">>,
+        description => <<"Metric collection timestamp (ISO8601)">>
+      },
+      <<"node">> => #{
+        type => string,
+        description => <<"Erlang node name">>
+      },
+      <<"memory_total">> => #{
+        type => integer,
+        description => <<"Total memory allocated by the Erlang VM (bytes)">>
+      },
+      <<"memory_processes">> => #{
+        type => integer,
+        description => <<"Memory used by processes (bytes)">>
+      },
+      <<"memory_atom">> => #{
+        type => integer,
+        description => <<"Memory used by atoms (bytes)">>
+      },
+      <<"memory_binary">> => #{
+        type => integer,
+        description => <<"Memory used by binaries (bytes)">>
+      },
+      <<"memory_ets">> => #{
+        type => integer,
+        description => <<"Memory used by ETS tables (bytes)">>
+      },
+      <<"process_count">> => #{
+        type => integer,
+        description => <<"Number of live Erlang processes">>
+      },
+      <<"run_queue">> => #{
+        type => integer,
+        description => <<"Length of the scheduler run queue">>
+      },
+      <<"mnesia_commits">> => #{
+        type => integer,
+        description => <<"Number of successful Mnesia transactions since node start">>
+      },
+      <<"mnesia_failures">> => #{
+        type => integer,
+        description => <<"Number of failed Mnesia transactions since node start">>
+      },
+      <<"table_sizes">> => #{
+        type => object,
+        description => <<"Map of Mnesia table names to their sizes (number of records)">>,
+        additionalProperties => #{type => integer}
+      },
+      <<"active_sessions">> => #{
+        type => integer,
+        description => <<"Total number of active user and admin sessions">>
+      },
+      <<"ws_connections">> => #{
+        type => integer,
+        description => <<"Number of open WebSocket connections">>
+      },
+      <<"io_in">> => #{
+        type => integer,
+        description => <<"Bytes received on sockets since last collection (delta)">>
+      },
+      <<"io_out">> => #{
+        type => integer,
+        description => <<"Bytes sent over sockets since last collection (delta)">>
+      }
+    }
+  }.
+
+%%% Internal functions
+
+%% @doc Возвращает исторические метрики по заданному периоду и узлу.
+-spec get_metrics(cowboy_req:req()) -> {ok, binary(), cowboy_req:req()}.
+get_metrics(Req) ->
+  case handler_utils:auth_admin(Req) of
+    {ok, _AdminId, Req1} ->
+      Qs = cowboy_req:parse_qs(Req1),
+      From = handler_utils:parse_datetime_qs(proplists:get_value(<<"from">>, Qs)),
+      To   = handler_utils:parse_datetime_qs(proplists:get_value(<<"to">>, Qs)),
+      Node = proplists:get_value(<<"node">>, Qs, node()),
+      case {From, To} of
+        {undefined, _} -> handler_utils:send_error(Req1, 400, <<"Missing from">>);
+        {_, undefined} -> handler_utils:send_error(Req1, 400, <<"Missing to">>);
+        {F, T} ->
+          Metrics = core_node_metric:get_by_timerange(F, T, Node),
+          Json = [metric_to_map(M) || M <- Metrics],
+          handler_utils:send_json(Req1, 200, Json)
+      end;
+    {error, Code, Msg, Req1} ->
+      handler_utils:send_error(Req1, Code, Msg)
+  end.
+
+%% @private Преобразует запись метрики в JSON-совместимую карту.
+-spec metric_to_map(#node_metric{}) -> map().
+metric_to_map(M) ->
+  #{
+    <<"timestamp">>        => handler_utils:datetime_to_iso8601(M#node_metric.timestamp),
+    <<"node">>             => atom_to_binary(M#node_metric.node, utf8),
+    <<"memory_total">>     => M#node_metric.memory_total,
+    <<"memory_processes">> => M#node_metric.memory_processes,
+    <<"memory_atom">>      => M#node_metric.memory_atom,
+    <<"memory_binary">>    => M#node_metric.memory_binary,
+    <<"memory_ets">>       => M#node_metric.memory_ets,
+    <<"process_count">>    => M#node_metric.process_count,
+    <<"run_queue">>        => M#node_metric.run_queue,
+    <<"mnesia_commits">>   => M#node_metric.mnesia_commits,
+    <<"mnesia_failures">>  => M#node_metric.mnesia_failures,
+    <<"table_sizes">>      => M#node_metric.table_sizes,
+    <<"active_sessions">>  => M#node_metric.active_sessions,
+    <<"ws_connections">>   => M#node_metric.ws_connections,
+    <<"io_in">>            => M#node_metric.io_in,
+    <<"io_out">>           => M#node_metric.io_out
+  }.
@@ -1,11 +1,13 @@
 %%%-------------------------------------------------------------------
 %%% @doc Административный WebSocket-обработчик.
 %%% Устанавливает WebSocket-соединение после проверки JWT-токена
-%%% и подписывает администратора на каналы уведомлений.
+%%% и подписывает администратора на каналы уведомлений, включая
+%%% канал метрик узлов `node_metrics`.
 %%% @end
 %%%-------------------------------------------------------------------
 -module(admin_ws_handler).
 -behaviour(cowboy_websocket).
+-include("records.hrl").

 -export([init/2]).
 -export([websocket_init/1]).
@@ -30,7 +32,7 @@ init(Req, _Opts) ->
      Resp = cowboy_req:reply(401, #{}, <<"Missing token">>, Req),
      {ok, Resp, undefined};
    Token ->
-      io:format("[ADMIN_WS] Token received: ~s...~n", [binary_part(Token, 0, 30)]),
+      io:format("[ADMIN_WS] Token received: ~s...~n", [safe_slice(Token, 30)]),
      case logic_auth:verify_jwt(Token) of
        {ok, UserId, Role} ->
          io:format("[ADMIN_WS] UserId: ~s, Role: ~s~n", [UserId, Role]),
@@ -57,8 +59,11 @@ init(Req, _Opts) ->
 %% @doc Вызывается после установки WebSocket-соединения.
 -spec websocket_init(#state{}) -> {ok, #state{}}.
 websocket_init(State) ->
+  core_counters:inc(admin_ws_connections),
  io:format("[ADMIN_WS] WebSocket initialized for admin ~s~n", [State#state.admin_id]),
  pg:join(eventhub_admin_ws, self()),
+  % Подписываемся на метрики узлов
+  pg:join(node_metrics, self()),
  {ok, State}.

 %% @doc Обрабатывает входящие текстовые сообщения (subscribe/unsubscribe/ping).
@@ -84,7 +89,7 @@ websocket_handle({text, Msg}, State) ->
 websocket_handle(_Frame, State) ->
  {ok, State}.

-%% @doc Отправляет административное уведомление через WebSocket.
+%% @doc Отправляет административное уведомление или метрику узла через WebSocket.
 -spec websocket_info(term(), #state{}) -> {reply, {text, binary()}, #state{}} | {ok, #state{}}.
 websocket_info({admin_notification, Type, Data}, State) ->
  Msg = jsx:encode(#{
@@ -93,6 +98,12 @@ websocket_info({admin_notification, Type, Data}, State) ->
    timestamp => os:system_time(seconds)
  }),
  {reply, {text, Msg}, State};
+websocket_info({node_metric, Metric}, State) ->
+  JSON = jsx:encode(#{
+    <<"type">> => <<"node_metric">>,
+    <<"data">> => metric_to_map(Metric)
+  }),
+  {reply, {text, JSON}, State};
 websocket_info(_Info, State) ->
  {ok, State}.

@@ -100,4 +111,37 @@ websocket_info(_Info, State) ->
 -spec terminate(term(), cowboy_req:req(), #state{}) -> ok.
 terminate(_Reason, _Req, _State) ->
  pg:leave(eventhub_admin_ws, self()),
-  ok.
+  core_counters:dec(admin_ws_connections),
+  ok.
+
+%%%-------------------------------------------------------------------
+%%% Вспомогательные функции
+%%%-------------------------------------------------------------------
+
+%% Безопасное получение первых Max байт бинарной строки
+safe_slice(Bin, Max) ->
+  case byte_size(Bin) > Max of
+    true  -> binary_part(Bin, 0, Max);
+    false -> Bin
+  end.
+
+%% Преобразование #node_metric{} в карту для JSON
+metric_to_map(M) ->
+  #{
+    <<"timestamp">>        => handler_utils:datetime_to_iso8601(M#node_metric.timestamp),
+    <<"node">>             => atom_to_binary(M#node_metric.node, utf8),
+    <<"memory_total">>     => M#node_metric.memory_total,
+    <<"memory_processes">> => M#node_metric.memory_processes,
+    <<"memory_atom">>      => M#node_metric.memory_atom,
+    <<"memory_binary">>    => M#node_metric.memory_binary,
+    <<"memory_ets">>       => M#node_metric.memory_ets,
+    <<"process_count">>    => M#node_metric.process_count,
+    <<"run_queue">>        => M#node_metric.run_queue,
+    <<"mnesia_commits">>   => M#node_metric.mnesia_commits,
+    <<"mnesia_failures">>  => M#node_metric.mnesia_failures,
+    <<"table_sizes">>      => M#node_metric.table_sizes,
+    <<"active_sessions">>  => M#node_metric.active_sessions,
+    <<"ws_connections">>   => M#node_metric.ws_connections,
+    <<"io_in">>            => M#node_metric.io_in,
+    <<"io_out">>           => M#node_metric.io_out
+  }.
@@ -55,6 +55,7 @@ init(Req, _Opts) ->
 websocket_init(#state{user_id = UserId} = State) ->
  pg:join(eventhub_ws, self()),
  io:format("[WS] User ~s connected~n", [UserId]),
+  core_counters:inc(ws_connections),
  {ok, State#state{subscriptions = []}}.

 -spec websocket_handle(term(), #state{}) ->
@@ -99,6 +100,7 @@ websocket_info(_Info, State) ->
 -spec terminate(term(), cowboy_req:req(), #state{}) -> ok.
 terminate(_Reason, _Req, #state{user_id = UserId}) ->
  pg:leave(eventhub_ws, self()),
+  core_counters:dec(ws_connections),
  io:format("[WS] User ~s disconnected~n", [UserId]),
  ok.