From c73e755a000a6e47d53913e7c2d2988cd45b9d6f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Jean-S=C3=A9bastien=20P=C3=A9dron?=
 <jean-sebastien.pedron@dumbbell.fr>
Date: Thu, 19 Dec 2024 19:39:15 +0100
Subject: [PATCH 1/2] rabbitmq_peer_discovery_etcd: Wait for etcd start in
 system_SUITE

[Why]
It was possible that testcases were executed before the etcd daemon was
ready, leading to test failures.

[How]
There was already a santy check to verify that the etcd daemon was
working correctly, but it was itself a testcase.

This patch moves this code to the etcd start code to wait for it to be
ready.

This replaces the previous workaround of waiting for 2 seconds.

While here, log anything printed to stdout/stderr by etcd after it
exited.

Fixes #12981.

(cherry picked from commit d54b2bff3cd0700fc7f63f1225e3928ef660d09b)

# Conflicts:
#	deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl
---
 .../test/system_SUITE.erl                     | 78 +++++++++++++------
 1 file changed, 53 insertions(+), 25 deletions(-)

diff --git a/deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl b/deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl
index 5224ff4ebd96..d9f81a375f52 100644
--- a/deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl
+++ b/deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl
@@ -26,7 +26,6 @@
          init_per_testcase/2,
          end_per_testcase/2,
 
-         etcd_connection_sanity_check_test/1,
          init_opens_a_connection_test/1,
          registration_with_locking_test/1,
          start_one_member_at_a_time/1,
@@ -41,7 +40,6 @@ all() ->
 groups() ->
     [
      {v3_client, [], [
-                    etcd_connection_sanity_check_test,
                     init_opens_a_connection_test,
                     registration_with_locking_test
                 ]},
@@ -202,13 +200,61 @@ start_etcd(Config) ->
            "--initial-cluster-state", "new",
            "--initial-cluster-token", "test-token",
            "--log-level", "debug", "--log-outputs", "stdout"],
+<<<<<<< HEAD
     EtcdPid = spawn(fun() -> rabbit_ct_helpers:exec(Cmd) end),
+=======
+    EtcdPid = spawn(fun() -> do_start_etcd(Cmd) end),
+>>>>>>> d54b2bff3 (rabbitmq_peer_discovery_etcd: Wait for etcd start in system_SUITE)
 
-    EtcdEndpoint = rabbit_misc:format("~s:~b", [EtcdHost, EtcdClientPort]),
-    rabbit_ct_helpers:set_config(
-      Config,
-      [{etcd_pid, EtcdPid},
-       {etcd_endpoints, [EtcdEndpoint]}]).
+    EtcdEndpoints = [rabbit_misc:format("~s:~b", [EtcdHost, EtcdClientPort])],
+    Config1 = rabbit_ct_helpers:set_config(
+                Config,
+                [{etcd_pid, EtcdPid},
+                 {etcd_endpoints, EtcdEndpoints}]),
+
+    #{level := Level} = logger:get_primary_config(),
+    logger:set_primary_config(level, critical),
+    try
+        wait_for_etcd(EtcdEndpoints),
+        Config1
+    catch
+        exit:{test_case_failed, _} ->
+            stop_etcd(Config1),
+            {skip, "Failed to start etcd"}
+    after
+        logger:set_primary_config(level, Level)
+    end.
+
+do_start_etcd(Cmd) ->
+    case rabbit_ct_helpers:exec(Cmd) of
+        {ok, Stdout} ->
+            ct:pal("etcd daemon exited:~n~s", [Stdout]);
+        {error, Reason, Stdout} ->
+            ct:pal(
+              "etcd daemon exited with error ~0p:~n~s",
+              [Reason, Stdout])
+    end.
+
+wait_for_etcd(EtcdEndpoints) ->
+    application:ensure_all_started(eetcd),
+
+    Timeout = 60000,
+    rabbit_ct_helpers:await_condition(
+      fun() ->
+              case eetcd:open(test, EtcdEndpoints) of
+                  {ok, _Pid} -> true;
+                  _          -> false
+              end
+      end, Timeout),
+
+    Condition1 = fun() -> 1 =:= length(eetcd_conn_sup:info()) end,
+    try
+        rabbit_ct_helpers:await_condition(Condition1, Timeout)
+    after
+        eetcd:close(test)
+    end,
+    Condition2 = fun() -> 0 =:= length(eetcd_conn_sup:info()) end,
+    rabbit_ct_helpers:await_condition(Condition2, Timeout).
 
 stop_etcd(Config) ->
     case rabbit_ct_helpers:get_config(Config, etcd_pid) of
@@ -226,24 +272,6 @@ stop_etcd(Config) ->
 %% Test cases
 %%
 
-etcd_connection_sanity_check_test(Config) ->
-    application:ensure_all_started(eetcd),
-    Endpoints = ?config(etcd_endpoints, Config),
-    ?assertMatch({ok, _Pid}, eetcd:open(test, Endpoints)),
-
-    Condition1 = fun() ->
-                    1 =:= length(eetcd_conn_sup:info())
-                end,
-    try
-        rabbit_ct_helpers:await_condition(Condition1, 60000)
-    after
-        eetcd:close(test)
-    end,
-    Condition2 = fun() ->
-                    0 =:= length(eetcd_conn_sup:info())
-                end,
-    rabbit_ct_helpers:await_condition(Condition2, 60000).
-
 init_opens_a_connection_test(Config) ->
     Endpoints = ?config(etcd_endpoints, Config),
     {ok, Pid} = start_client(Endpoints),

From 1c9918c564c4876646d6762924e7b1571d75c346 Mon Sep 17 00:00:00 2001
From: Michael Klishin <michaelklishin@icloud.com>
Date: Thu, 19 Dec 2024 18:35:57 -0500
Subject: [PATCH 2/2] rabbitmq_peer_discovery_etcd system_SUITE: resolve a
 conflict #12985 #12988

---
 deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl b/deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl
index d9f81a375f52..2f7c0bcda85e 100644
--- a/deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl
+++ b/deps/rabbitmq_peer_discovery_etcd/test/system_SUITE.erl
@@ -200,11 +200,7 @@ start_etcd(Config) ->
            "--initial-cluster-state", "new",
            "--initial-cluster-token", "test-token",
            "--log-level", "debug", "--log-outputs", "stdout"],
-<<<<<<< HEAD
-    EtcdPid = spawn(fun() -> rabbit_ct_helpers:exec(Cmd) end),
-=======
     EtcdPid = spawn(fun() -> do_start_etcd(Cmd) end),
->>>>>>> d54b2bff3 (rabbitmq_peer_discovery_etcd: Wait for etcd start in system_SUITE)
 
     EtcdEndpoints = [rabbit_misc:format("~s:~b", [EtcdHost, EtcdClientPort])],
     Config1 = rabbit_ct_helpers:set_config(