From 8b5a717a1f7617a25e2d1e6a132063441f7850a5 Mon Sep 17 00:00:00 2001
From: Thales Macedo Garitezi <thalesmg@gmail.com>
Date: Fri, 31 Mar 2023 11:02:15 -0300
Subject: [PATCH] test(peer): increase init and startup timeout for peer nodes

Attempt to stabilize tests that use cluster nodes.
---
 apps/emqx/test/emqx_common_test_helpers.erl   |  6 +++---
 .../emqx_bridge_impl_kafka_consumer_SUITE.erl | 20 ++++++++++++-------
 2 files changed, 16 insertions(+), 10 deletions(-)

diff --git a/apps/emqx/test/emqx_common_test_helpers.erl b/apps/emqx/test/emqx_common_test_helpers.erl
index 9a4461fac..077ebe138 100644
--- a/apps/emqx/test/emqx_common_test_helpers.erl
+++ b/apps/emqx/test/emqx_common_test_helpers.erl
@@ -660,6 +660,7 @@ start_slave(Name, Opts) when is_list(Opts) ->
 start_slave(Name, Opts) when is_map(Opts) ->
     SlaveMod = maps:get(peer_mod, Opts, ct_slave),
     Node = node_name(Name),
+    put_peer_mod(Node, SlaveMod),
     DoStart =
         fun() ->
             case SlaveMod of
@@ -669,8 +670,8 @@ start_slave(Name, Opts) when is_map(Opts) ->
                         [
                             {kill_if_fail, true},
                             {monitor_master, true},
-                            {init_timeout, 10000},
-                            {startup_timeout, 10000},
+                            {init_timeout, 20_000},
+                            {startup_timeout, 20_000},
                             {erl_flags, erl_flags()}
                         ]
                     );
@@ -687,7 +688,6 @@ start_slave(Name, Opts) when is_map(Opts) ->
             throw(Other)
     end,
     pong = net_adm:ping(Node),
-    put_peer_mod(Node, SlaveMod),
     setup_node(Node, Opts),
     ok = snabbkaffe:forward_trace(Node),
     Node.
diff --git a/lib-ee/emqx_ee_bridge/test/emqx_bridge_impl_kafka_consumer_SUITE.erl b/lib-ee/emqx_ee_bridge/test/emqx_bridge_impl_kafka_consumer_SUITE.erl
index 8664f9745..4019a9c42 100644
--- a/lib-ee/emqx_ee_bridge/test/emqx_bridge_impl_kafka_consumer_SUITE.erl
+++ b/lib-ee/emqx_ee_bridge/test/emqx_bridge_impl_kafka_consumer_SUITE.erl
@@ -388,7 +388,9 @@ end_per_testcase(_Testcase, Config) ->
                 maps:values(ProducersMapping)
             ),
             ok = wolff:stop_and_delete_supervised_client(KafkaProducerClientId),
-            emqx_common_test_helpers:call_janitor(30_000),
+            %% in CI, apparently this needs more time since the
+            %% machines struggle with all the containers running...
+            emqx_common_test_helpers:call_janitor(60_000),
             ok = snabbkaffe:stop(),
             ok
     end.
@@ -1664,7 +1666,7 @@ t_cluster_group(Config) ->
                  || {Name, Opts} <- Cluster
                 ],
             on_exit(fun() ->
-                lists:foreach(
+                emqx_misc:pmap(
                     fun(N) ->
                         ct:pal("stopping ~p", [N]),
                         ok = emqx_common_test_helpers:stop_slave(N)
@@ -1875,7 +1877,7 @@ t_cluster_node_down(Config) ->
                     Cluster
                 ),
             on_exit(fun() ->
-                lists:foreach(
+                emqx_misc:pmap(
                     fun(N) ->
                         ct:pal("stopping ~p", [N]),
                         ok = emqx_common_test_helpers:stop_slave(N)
@@ -1894,10 +1896,14 @@ t_cluster_node_down(Config) ->
             {ok, _} = snabbkaffe:receive_events(SRef0),
             lists:foreach(
                 fun(N) ->
-                    ?assertMatch(
-                        {ok, _},
-                        erpc:call(N, emqx_bridge, lookup, [BridgeId]),
-                        #{node => N}
+                    ?retry(
+                        _Sleep1 = 100,
+                        _Attempts1 = 50,
+                        ?assertMatch(
+                            {ok, _},
+                            erpc:call(N, emqx_bridge, lookup, [BridgeId]),
+                            #{node => N}
+                        )
                     )
                 end,
                 Nodes