update healthcheck api with additional checks

heroku · Dec 14, 2010 · b488f03 · b488f03
1 parent d6f1cad
commit b488f03
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 6 deletions.
diff --git a/src/logplex_api.erl b/src/logplex_api.erl
@@ -72,8 +72,15 @@ loop(Req) ->
 handlers() ->
     [{['GET', "/healthcheck"], fun(Req, _Match) ->
         authorize(Req),
+
+        [throw({500, io_lib:format("Zero ~p child processes running", [Worker])}) || {Worker, 0} <- logplex_stats:workers()],
+
+        RegisteredMods = [logplex_grid, logplex_rate_limit, logplex_realtime, logplex_stats, logplex_channel, logplex_token, logplex_drain, logplex_session, logplex_tail, logplex_shard, syslog_acceptor],
+        [(whereis(Name) == undefined orelse not is_process_alive(whereis(Name))) andalso throw({500, io_lib:format("Process dead: ~p", [Name])}) || Name <- RegisteredMods],
+
         Count = logplex_stats:healthcheck(),
-        not is_integer(Count) andalso exit({expected_integer, Count}),
+        not is_integer(Count) andalso throw({500, io_lib:format("Increment healthcheck counter failed: ~p", [Count])}),
+
         {200, integer_to_list(Count)}
     end},
 

diff --git a/src/logplex_app.erl b/src/logplex_app.erl
@@ -44,7 +44,7 @@ init([]) ->
     {ok, {{one_for_one, 5, 10}, [
         {logplex_grid, {logplex_grid, start_link, []}, permanent, 2000, worker, [logplex_grid]},
         {logplex_rate_limit, {logplex_rate_limit, start_link, []}, permanent, 2000, worker, [logplex_rate_limit]},
-        %{logplex_realtime, {logplex_realtime, start_link, []}, permanent, 2000, worker, [logplex_realtime]},
+        {logplex_realtime, {logplex_realtime, start_link, []}, permanent, 2000, worker, [logplex_realtime]},
         {logplex_stats, {logplex_stats, start_link, []}, permanent, 2000, worker, [logplex_stats]},
 
         {logplex_channel, {logplex_channel, start_link, []}, permanent, 2000, worker, [logplex_channel]},

diff --git a/src/logplex_realtime.erl b/src/logplex_realtime.erl
@@ -35,7 +35,7 @@
     message_received=0,
     message_processed=0,
     message_routed=0,
-    queue_dropped=0,
+    work_queue_dropped=0,
     drain_buffer_dropped=0,
     redis_buffer_dropped=0
 }).
@@ -96,8 +96,8 @@ handle_cast({incr, message_processed, Incr}, #state{message_processed=Count}=Sta
 handle_cast({incr, message_routed, Incr}, #state{message_routed=Count}=State) ->
     {noreply, State#state{message_routed=Count+Incr}};
 
-handle_cast({incr, queue_dropped, Incr}, #state{queue_dropped=Count}=State) ->
-    {noreply, State#state{queue_dropped=Count+Incr}};
+handle_cast({incr, work_queue_dropped, Incr}, #state{work_queue_dropped=Count}=State) ->
+    {noreply, State#state{work_queue_dropped=Count+Incr}};
 
 handle_cast({incr, drain_buffer_dropped, Incr}, #state{drain_buffer_dropped=Count}=State) ->
     {noreply, State#state{drain_buffer_dropped=Count+Incr}};
@@ -121,7 +121,7 @@ handle_info(flush, State) ->
             {message_received, State#state.message_received},
             {message_processed, State#state.message_processed},
             {message_routed, State#state.message_routed},
-            {queue_dropped, State#state.queue_dropped},
+            {work_queue_dropped, State#state.work_queue_dropped},
             {drain_buffer_dropped, State#state.drain_buffer_dropped},
             {redis_buffer_dropped, State#state.redis_buffer_dropped}
         ]})),