Merge pull request #6948 from zhongwencool/fix-watermark-master

fix(watermark): watermark should use > instead of >=

apps/emqx/src/emqx_os_mon.erl

@@ -76,10 +76,16 @@ set_procmem_high_watermark(Float) ->
 %%--------------------------------------------------------------------
 
 init([]) ->
-    Opts = emqx:get_config([sysmon, os]),
-    set_mem_check_interval(maps:get(mem_check_interval, Opts)),
-    set_sysmem_high_watermark(maps:get(sysmem_high_watermark, Opts)),
-    set_procmem_high_watermark(maps:get(procmem_high_watermark, Opts)),
+    #{
+        sysmem_high_watermark := SysHW,
+        procmem_high_watermark := PHW,
+        mem_check_interval := MCI
+    } = emqx:get_config([sysmon, os]),
+
+    set_sysmem_high_watermark(SysHW),
+    set_procmem_high_watermark(PHW),
+    set_mem_check_interval(MCI),
+    ensure_system_memory_alarm(SysHW),
     _ = start_check_timer(),
     {ok, #{}}.
 
@@ -95,7 +101,7 @@ handle_info({timeout, _Timer, check}, State) ->
     CPULowWatermark = emqx:get_config([sysmon, os, cpu_low_watermark]) * 100,
     _ = case emqx_vm:cpu_util() of %% TODO: should be improved?
         0 -> ok;
-        Busy when Busy >= CPUHighWatermark ->
+        Busy when Busy > CPUHighWatermark ->
             Usage = list_to_binary(io_lib:format("~.2f%", [Busy])),
             Message = <<Usage/binary, " cpu usage">>,
             emqx_alarm:activate(high_cpu_usage,
@@ -106,7 +112,7 @@ handle_info({timeout, _Timer, check}, State) ->
                 },
                 Message),
             start_check_timer();
-        Busy when Busy =< CPULowWatermark ->
+        Busy when Busy < CPULowWatermark ->
             Usage = list_to_binary(io_lib:format("~.2f%", [Busy])),
             Message = <<Usage/binary, " cpu usage">>,
             emqx_alarm:deactivate(high_cpu_usage,
@@ -142,3 +148,20 @@ start_check_timer() ->
         "x86_64-pc-linux-musl" -> ok;
         _ -> emqx_misc:start_timer(Interval, check)
     end.
+
+%% At startup, memsup starts first and checks for memory alarms,
+%% but emqx_alarm_handler is not yet used instead of alarm_handler,
+%% so alarm_handler is used directly for notification (normally emqx_alarm_handler should be used).
+%%The internal memsup will no longer trigger events that have been alerted,
+%% and there is no exported function to remove the alerted flag,
+%% so it can only be checked again at startup.
+ensure_system_memory_alarm(HW) ->
+    case erlang:whereis(memsup) of
+        undefined -> ok;
+        _Pid ->
+            {Allocated, Total, _Worst} = memsup:get_memory_data(),
+            case Total =/= 0 andalso Allocated/Total * 100 > HW of
+                true -> emqx_alarm:activate(high_system_memory_usage, #{high_watermark => HW});
+                false -> ok
+            end
+    end.

apps/emqx/src/emqx_vm_mon.erl

@@ -61,7 +61,7 @@ handle_info({timeout, _Timer, check}, State) ->
     ProcLowWatermark = emqx:get_config([sysmon, vm, process_low_watermark]),
     ProcessCount = erlang:system_info(process_count),
     case ProcessCount / erlang:system_info(process_limit) of
-        Percent when Percent >= ProcHighWatermark ->
+        Percent when Percent > ProcHighWatermark ->
             Usage = io_lib:format("~p%", [Percent*100]),
             Message = [Usage, " process usage"],
             emqx_alarm:activate(too_many_processes,

apps/emqx/test/emqx_os_mon_SUITE.erl

@@ -24,20 +24,24 @@
 all() -> emqx_common_test_helpers:all(?MODULE).
 
 init_per_suite(Config) ->
-    emqx_config:put([sysmon, os], #{
-        cpu_check_interval => 60000,cpu_high_watermark => 0.8,
-        cpu_low_watermark => 0.6,mem_check_interval => 60000,
-        procmem_high_watermark => 0.05,sysmem_high_watermark => 0.7}),
-    application:ensure_all_started(os_mon),
+    emqx_common_test_helpers:boot_modules(all),
+    emqx_common_test_helpers:start_apps([],
+        fun(emqx) ->
+            application:set_env(emqx, os_mon, [
+                {cpu_check_interval, 1},
+                {cpu_high_watermark, 5},
+                {cpu_low_watermark, 80},
+                {mem_check_interval, 60},
+                {sysmem_high_watermark, 70},
+                {procmem_high_watermark, 5}]);
+            (_) -> ok
+        end),
     Config.
 
 end_per_suite(_Config) ->
-    application:stop(os_mon).
+    emqx_common_test_helpers:stop_apps([]).
 
 t_api(_) ->
-    gen_event:swap_handler(alarm_handler, {emqx_alarm_handler, swap}, {alarm_handler, []}),
-    {ok, _} = emqx_os_mon:start_link(),
-
     ?assertEqual(60000, emqx_os_mon:get_mem_check_interval()),
     ?assertEqual(ok, emqx_os_mon:set_mem_check_interval(30000)),
     ?assertEqual(60000, emqx_os_mon:get_mem_check_interval()),
@@ -58,4 +62,3 @@ t_api(_) ->
     emqx_os_mon ! ignored,
     gen_server:stop(emqx_os_mon),
     ok.
-