WIP

big_tests/default.spec

@@ -51,7 +51,7 @@
 {suites, "tests", graphql_http_upload_SUITE}.
 {suites, "tests", graphql_server_SUITE}.
 {suites, "tests", graphql_metric_SUITE}.
-{suites, "tests", instrument_cets_SUITE}.
+{suites, "tests", instrument_SUITE}.
 {suites, "tests", inbox_SUITE}.
 {suites, "tests", inbox_extensions_SUITE}.
 {suites, "tests", jingle_SUITE}.

big_tests/dynamic_domains.spec

@@ -70,7 +70,7 @@
 {suites, "tests", graphql_server_SUITE}.
 {suites, "tests", graphql_metric_SUITE}.
 
-{suites, "tests", instrument_cets_SUITE}.
+{suites, "tests", instrument_SUITE}.
 
 {suites, "tests", inbox_SUITE}.
 {suites, "tests", inbox_extensions_SUITE}.

big_tests/tests/cets_disco_SUITE.erl

@@ -363,9 +363,7 @@ start_cets_discovery_with_file_backnend(Config) ->
 
 stop_cets_discovery() ->
     ok = rpc(mim(), supervisor, terminate_child, [ejabberd_sup, cets_discovery]),
-    ok = rpc(mim2(), supervisor, terminate_child, [ejabberd_sup, cets_discovery]),
-    rpc(mim(), mongoose_instrument_probe_cets, stop, []),
-    rpc(mim2(), mongoose_instrument_probe_cets, stop, []).
+    ok = rpc(mim2(), supervisor, terminate_child, [ejabberd_sup, cets_discovery]).
 
 stop_and_delete_cets_discovery() ->
     stop_cets_discovery(),

big_tests/tests/cluster_commands_SUITE.erl

@@ -83,13 +83,21 @@ init_per_suite(Config) ->
     NodeCtlPath = distributed_helper:ctl_path(Node1, Config3),
     Node2CtlPath = distributed_helper:ctl_path(Node2, Config3),
     Node3CtlPath = distributed_helper:ctl_path(Node3, Config3),
-    escalus:init_per_suite([{ctl_path_atom(Node1), NodeCtlPath},
-                            {ctl_path_atom(Node2), Node2CtlPath},
-                            {ctl_path_atom(Node3), Node3CtlPath}]
-                           ++ Config3).
+    Config4 = escalus:init_per_suite([{ctl_path_atom(Node1), NodeCtlPath},
+                                      {ctl_path_atom(Node2), Node2CtlPath},
+                                      {ctl_path_atom(Node3), Node3CtlPath}]
+                                     ++ Config3),
+    Config5 = mongoose_helper:backup_and_set_config_option(
+                Config4, [instrumentation, probe_interval], 1),
+    restart(mongoose_system_probes),
+    instrument_helper:start([{system_dist_data, #{}}]),
+    Config5.
 
 end_per_suite(Config) ->
-    escalus:end_per_suite(Config).
+    mongoose_helper:restore_config_option(Config, [instrumentation, probe_interval]),
+    restart(mongoose_system_probes),
+    escalus:end_per_suite(Config),
+    instrument_helper:stop().
 
 init_per_group(Group, Config) when Group == clustered orelse Group == mnesia ->
     Node2 = mim2(),
@@ -164,10 +172,10 @@ end_per_testcase(CaseName, Config) ->
 %% Message tests
 %%--------------------------------------------------------------------
 
-one_to_one_message(ConfigIn) ->
+one_to_one_message(Config) ->
     %% Given Alice connected to node one and ClusterGuy connected to node two
-    Metrics = [{[global, data, dist], [{recv_oct, '>'}, {send_oct, '>'}]}],
-    Config = [{mongoose_metrics, Metrics} | ConfigIn],
+    #{send_oct := InitialSendOct, recv_oct := InitialRecvOct}
+        = rpc(mim(), mongoose_system_probes, probe, [system_dist_data, #{}]),
     escalus:story(Config, [{alice, 1}, {clusterguy, 1}], fun(Alice, ClusterGuy) ->
         %% When Alice sends a message to ClusterGuy
         Msg1 = escalus_stanza:chat_to(ClusterGuy, <<"Hi!">>),
@@ -182,7 +190,12 @@ one_to_one_message(ConfigIn) ->
         %% Then Alice also receives it
         Stanza2 = escalus:wait_for_stanza(Alice, 5000),
         escalus:assert(is_chat_message, [<<"Oh hi!">>], Stanza2)
-    end).
+    end),
+    instrument_helper:wait_and_assert(
+      system_dist_data, #{},
+      fun(#{send_oct := SendOct, recv_oct := RecvOct}) ->
+              SendOct > InitialSendOct andalso RecvOct > InitialRecvOct
+      end).
 
 %%--------------------------------------------------------------------
 %% Manage cluster commands tests
@@ -469,3 +482,7 @@ wait_for_process_to_stop(Pid, Timeout) ->
     after Timeout ->
             ct:fail(wait_for_process_to_stop_timeout)
     end.
+
+restart(Module) ->
+    rpc(mim(), Module, stop, []),
+    rpc(mim(), Module, start, []).

big_tests/tests/graphql_metric_SUITE.erl

@@ -119,38 +119,23 @@ type_to_keys(<<"counter">>) ->
 type_to_keys(<<"spiral">>) ->
     [<<"one">>, <<"count">>];
 type_to_keys(<<"gauge">>) ->
-    [<<"value">>];
-type_to_keys(<<"merged_inet_stats">>) ->
-    [<<"connections">>, <<"recv_cnt">>, <<"recv_max">>, <<"recv_oct">>,
-     <<"send_cnt">>, <<"send_max">>, <<"send_oct">>, <<"send_pend">>];
-type_to_keys(<<"rdbms_stats">>) ->
-    [<<"workers">>, <<"recv_cnt">>, <<"recv_max">>, <<"recv_oct">>,
-     <<"send_cnt">>, <<"send_max">>, <<"send_oct">>, <<"send_pend">>];
-type_to_keys(<<"vm_stats_memory">>) ->
-    [<<"atom_used">>, <<"binary">>, <<"ets">>,
-     <<"processes_used">>, <<"system">>, <<"total">>];
-type_to_keys(<<"vm_system_info">>) ->
-    [<<"ets_limit">>, <<"port_count">>, <<"port_limit">>,
-     <<"process_count">>, <<"process_limit">>];
-type_to_keys(<<"probe_queues">>) ->
-    [<<"fsm">>, <<"regular">>, <<"total">>].
+    [<<"value">>].
 
 cets_info_keys() ->
     [<<"available_nodes">>, <<"unavailable_nodes">>,
-    <<"remote_nodes_without_disco">>, <<"joined_nodes">>,
-    <<"remote_nodes_with_unknown_tables">>, <<"remote_unknown_tables">>,
-    <<"remote_nodes_with_missing_tables">>, <<"remote_missing_tables">>,
-    <<"conflict_nodes">>, <<"conflict_tables">>,
-    <<"discovered_nodes">>, <<"discovery_works">>].
+     <<"remote_nodes_without_disco">>, <<"joined_nodes">>,
+     <<"remote_nodes_with_unknown_tables">>, <<"remote_unknown_tables">>,
+     <<"remote_nodes_with_missing_tables">>, <<"remote_missing_tables">>,
+     <<"conflict_nodes">>, <<"conflict_tables">>,
+     <<"discovered_nodes">>, <<"discovery_works">>].
 
 get_by_name_global_erlang_metrics(Config) ->
     %% Filter by name works
-    Result = get_metrics([<<"global">>, <<"erlang">>], Config),
+    Result = get_metrics([<<"global">>, <<"system_info">>], Config),
     ParsedResult = get_ok_value([data, metric, getMetrics], Result),
     Map = maps:from_list([{Name, X} || X = #{<<"name">> := Name} <- ParsedResult]),
-    Info = maps:get([<<"global">>, <<"erlang">>, <<"system_info">>], Map),
-    %% VMSystemInfoMetric type
-    #{<<"type">> := <<"vm_system_info">>} = Info,
+    Info = maps:get([<<"global">>, <<"system_info">>, <<"port_count">>], Map),
+    #{<<"type">> := <<"counter">>} = Info,
     check_metric_by_type(Info),
     %% Other metrics are filtered out
     undef = maps:get(roster_reads_key(), Map, undef).
@@ -178,30 +163,27 @@ get_metrics_for_specific_host_type(Config) ->
     [_|_] = ParsedResult.
 
 get_process_queue_length(Config) ->
-    Result = get_metrics([<<"global">>, <<"processQueueLengths">>], Config),
+    Result = get_metrics([<<"global">>, <<"system_process_queue_lengths">>], Config),
     ParsedResult = get_ok_value([data, metric, getMetrics], Result),
     Map = maps:from_list([{Name, X} || X = #{<<"name">> := Name} <- ParsedResult]),
-    Lens = maps:get([<<"global">>, <<"processQueueLengths">>], Map),
-    %% ProbeQueuesMetric type
-    #{<<"type">> := <<"probe_queues">>} = Lens,
+    Lens = maps:get([<<"global">>, <<"system_process_queue_lengths">>, <<"total">>], Map),
+    #{<<"type">> := <<"counter">>} = Lens,
     check_metric_by_type(Lens).
 
 get_inet_stats(Config) ->
-    Result = get_metrics([<<"global">>, <<"data">>, <<"dist">>], Config),
+    Result = get_metrics([<<"global">>, <<"system_dist_data">>], Config),
     ParsedResult = get_ok_value([data, metric, getMetrics], Result),
     Map = maps:from_list([{Name, X} || X = #{<<"name">> := Name} <- ParsedResult]),
-    Stats = maps:get([<<"global">>, <<"data">>, <<"dist">>], Map),
-    %% MergedInetStatsMetric type
-    #{<<"type">> := <<"merged_inet_stats">>} = Stats,
+    Stats = maps:get([<<"global">>, <<"system_dist_data">>, <<"connections">>], Map),
+    #{<<"type">> := <<"counter">>} = Stats,
     check_metric_by_type(Stats).
 
 get_vm_stats_memory(Config) ->
     Result = get_metrics([<<"global">>], Config),
     ParsedResult = get_ok_value([data, metric, getMetrics], Result),
     Map = maps:from_list([{Name, X} || X = #{<<"name">> := Name} <- ParsedResult]),
-    Mem = maps:get([<<"global">>, <<"erlang">>, <<"memory">>], Map),
-    %% VMStatsMemoryMetric type
-    #{<<"type">> := <<"vm_stats_memory">>} = Mem,
+    Mem = maps:get([<<"global">>, <<"system_memory">>, <<"total">>], Map),
+    #{<<"type">> := <<"counter">>} = Mem,
     check_metric_by_type(Mem).
 
 get_cets_system(Config) ->
@@ -265,17 +247,17 @@ get_metrics_as_dicts_empty_args(Config) ->
     [#{<<"key">> := <<"median">>, <<"value">> := Median}] = maps:get(RecvName, Map),
     ?assert(is_integer(Median)),
     %% Empty keys
-    Result2 = get_metrics_as_dicts([<<"global">>, <<"erlang">>], [], Config),
+    Result2 = get_metrics_as_dicts([<<"global">>, <<"system_info">>], [], Config),
     ParsedResult2 = get_ok_value([data, metric, getMetricsAsDicts], Result2),
-    ?assertEqual(length(ParsedResult2), 2).
+    ?assertEqual(6, length(ParsedResult2)).
 
 get_metrics_as_dicts_empty_strings(Config) ->
     %% Name is an empty string
     Result = get_metrics_as_dicts([<<>>], [<<"median">>], Config),
     ParsedResult = get_ok_value([data, metric, getMetricsAsDicts], Result),
     [] = ParsedResult,
     %% Key is an empty string
-    Result2 = get_metrics_as_dicts([<<"global">>, <<"erlang">>], [<<>>], Config),
+    Result2 = get_metrics_as_dicts([<<"global">>, <<"system_info">>], [<<>>], Config),
     ParsedResult2 = get_ok_value([data, metric, getMetricsAsDicts], Result2),
     [_|_] = ParsedResult2.