Merge pull request #7838 from terry-xiaoyu/fix_ci_emqx_modules

fix(test): race condition in emqx_telemetry_SUITE

apps/emqx_conf/.gitignore

@@ -0,0 +1,3 @@
+data/
+etc/certs/
+etc/emqx.conf.all.rendered

apps/emqx_exhook/test/emqx_exhook_demo_svr.erl

@@ -66,9 +66,14 @@ start() ->
     start(?NAME, ?PORT).
 
 start(Name, Port) ->
-    Pid = spawn(fun() -> mgr_main(Name, Port) end),
+    Parent = self(),
+    Pid = spawn(fun() -> mgr_main(Parent, Name, Port) end),
     register(to_atom_name(Name), Pid),
-    {ok, Pid}.
+    receive
+        grpc_server_started -> {ok, Pid}
+    after 2000 ->
+        error({failed_to_start_grpc_server, Port})
+    end.
 
 stop() ->
     stop(?NAME).
@@ -87,7 +92,7 @@ take() ->
 in({FunName, Req}) ->
     to_atom_name(?NAME) ! {in, FunName, Req}.
 
-mgr_main(Name, Port) ->
+mgr_main(Parent, Name, Port) ->
     application:ensure_all_started(grpc),
     Services = #{
         protos => [emqx_exhook_pb],
@@ -95,6 +100,7 @@ mgr_main(Name, Port) ->
     },
     Options = [],
     Svr = grpc:start_server(Name, Port, Services, Options),
+    Parent ! grpc_server_started,
     mgr_loop([Svr, queue:new(), queue:new()]).
 
 mgr_loop([Svr, Q, Takes]) ->

apps/emqx_modules/test/emqx_telemetry_SUITE.erl

@@ -140,9 +140,15 @@ init_per_testcase(t_exhook_info, Config) ->
                 }
         },
     {ok, _} = emqx_exhook_demo_svr:start(),
+    {ok, Sock} = gen_tcp:connect("localhost", 9000, [], 3000),
+    _ = gen_tcp:close(Sock),
     ok = emqx_common_test_helpers:load_config(emqx_exhook_schema, ExhookConf),
     {ok, _} = application:ensure_all_started(emqx_exhook),
     Config;
+init_per_testcase(t_cluster_uuid, Config) ->
+    Node = start_slave(n1),
+    ok = setup_slave(Node),
+    [{n1, Node} | Config];
 init_per_testcase(_Testcase, Config) ->
     mock_httpc(),
     Config.
@@ -188,6 +194,9 @@ end_per_testcase(t_exhook_info, _Config) ->
     emqx_exhook_demo_svr:stop(),
     application:stop(emqx_exhook),
     ok;
+end_per_testcase(t_cluster_uuid, Config) ->
+    Node = proplists:get_value(n1, Config),
+    ok = stop_slave(Node);
 end_per_testcase(_Testcase, _Config) ->
     meck:unload([httpc]),
     ok.
@@ -211,23 +220,16 @@ t_node_uuid(_) ->
     ?assertEqual(NodeUUID3, NodeUUID4),
     ?assertMatch({badrpc, nodedown}, emqx_telemetry_proto_v1:get_node_uuid('fake@node')).
 
-t_cluster_uuid(_Config) ->
+t_cluster_uuid(Config) ->
+    Node = proplists:get_value(n1, Config),
     {ok, ClusterUUID0} = emqx_telemetry:get_cluster_uuid(),
     {ok, ClusterUUID1} = emqx_telemetry_proto_v1:get_cluster_uuid(node()),
     ?assertEqual(ClusterUUID0, ClusterUUID1),
     {ok, NodeUUID0} = emqx_telemetry:get_node_uuid(),
-
-    Node = start_slave(n1),
-    try
-        ok = setup_slave(Node),
     {ok, ClusterUUID2} = emqx_telemetry_proto_v1:get_cluster_uuid(Node),
     ?assertEqual(ClusterUUID0, ClusterUUID2),
     {ok, NodeUUID1} = emqx_telemetry_proto_v1:get_node_uuid(Node),
     ?assertNotEqual(NodeUUID0, NodeUUID1),
-        ok
-    after
-        ok = stop_slave(Node)
-    end,
     ok.
 
 t_official_version(_) ->
@@ -674,7 +676,10 @@ setup_slave(Node) ->
     ok.
 
 stop_slave(Node) ->
-    slave:stop(Node).
+    ok = ekka:force_leave(Node),
+    emqx_cluster_rpc:skip_failed_commit(Node),
+    ok = slave:stop(Node),
+    ?assertEqual([node()], mria_mnesia:running_nodes()).
 
 host() ->
     [_, Host] = string:tokens(atom_to_list(node()), "@"),

apps/emqx_modules/test/emqx_topic_metrics_api_SUITE.erl

@@ -27,6 +27,8 @@
     <<"topic_metrics">> => []
 }).
 
+suite() -> [{timetrap, {seconds, 30}}].
+
 all() ->
     emqx_common_test_helpers:all(?MODULE).