From 982fa22ffad01ebe640e5c23f3c6d0c67bde6742 Mon Sep 17 00:00:00 2001
From: Ryan Leung <rleungx@gmail.com>
Date: Fri, 11 Oct 2024 12:16:14 +0800
Subject: [PATCH] *: fix the race of `TestRaftClusterMultipleRestart` (#8686)

close tikv/pd#8543

Signed-off-by: Ryan Leung <rleungx@gmail.com>

Co-authored-by: ti-chi-bot[bot] <108142056+ti-chi-bot[bot]@users.noreply.github.com>
---
 pkg/replication/replication_mode.go  | 2 ++
 server/cluster/cluster.go            | 5 +++++
 tests/server/cluster/cluster_test.go | 2 ++
 3 files changed, 9 insertions(+)

diff --git a/pkg/replication/replication_mode.go b/pkg/replication/replication_mode.go
index fd55874ce30..cba83cf1ebb 100644
--- a/pkg/replication/replication_mode.go
+++ b/pkg/replication/replication_mode.go
@@ -345,6 +345,7 @@ func (m *ModeManager) Run(ctx context.Context) {
 	select {
 	case <-timer.C:
 	case <-ctx.Done():
+		log.Info("replication mode manager is stopped")
 		return
 	}
 
@@ -383,6 +384,7 @@ func (m *ModeManager) Run(ctx context.Context) {
 	}()
 
 	wg.Wait()
+	log.Info("replication mode manager is stopped")
 }
 
 func minimalUpVoters(rule *placement.Rule, upStores, downStores []*core.StoreInfo) int {
diff --git a/server/cluster/cluster.go b/server/cluster/cluster.go
index f1630d433d1..4cce39fa093 100644
--- a/server/cluster/cluster.go
+++ b/server/cluster/cluster.go
@@ -766,6 +766,11 @@ func (c *RaftCluster) Stop() {
 	log.Info("raft cluster is stopped")
 }
 
+// Wait blocks until the cluster is stopped. Only for test purpose.
+func (c *RaftCluster) Wait() {
+	c.wg.Wait()
+}
+
 // IsRunning return if the cluster is running.
 func (c *RaftCluster) IsRunning() bool {
 	c.RLock()
diff --git a/tests/server/cluster/cluster_test.go b/tests/server/cluster/cluster_test.go
index a1a7f68646e..a9be92d19e9 100644
--- a/tests/server/cluster/cluster_test.go
+++ b/tests/server/cluster/cluster_test.go
@@ -615,6 +615,8 @@ func TestRaftClusterMultipleRestart(t *testing.T) {
 	// let the job run at small interval
 	re.NoError(failpoint.Enable("github.com/tikv/pd/server/cluster/highFrequencyClusterJobs", `return(true)`))
 	for i := 0; i < 100; i++ {
+		// See https://github.com/tikv/pd/issues/8543
+		rc.Wait()
 		err = rc.Start(leaderServer.GetServer())
 		re.NoError(err)
 		time.Sleep(time.Millisecond)