Merge branch 'master' into unify-metrics

nolouch · Dec 20, 2023 · 3f1d841 · 3f1d841
2 parents 5789434 + a8aaae9
commit 3f1d841
Show file tree

Hide file tree

Showing 5 changed files with 56 additions and 11 deletions.
diff --git a/errors.toml b/errors.toml
@@ -766,6 +766,11 @@ error = '''
 cannot set invalid configuration
 '''
 
+["PD:server:ErrLeaderFrequentlyChange"]
+error = '''
+leader %s frequently changed, leader-key is [%s]
+'''
+
 ["PD:server:ErrLeaderNil"]
 error = '''
 leader is nil

diff --git a/pkg/election/leadership.go b/pkg/election/leadership.go
@@ -19,6 +19,7 @@ import (
 	"sync/atomic"
 	"time"
 
+	"github.com/pingcap/errors"
 	"github.com/pingcap/failpoint"
 	"github.com/pingcap/kvproto/pkg/pdpb"
 	"github.com/pingcap/log"
@@ -156,6 +157,16 @@ func (ls *Leadership) Campaign(leaseTimeout int64, leaderData string, cmps ...cl
 		lease:   clientv3.NewLease(ls.client),
 	}
 	ls.setLease(newLease)
+
+	failpoint.Inject("skipGrantLeader", func(val failpoint.Value) {
+		var member pdpb.Member
+		member.Unmarshal([]byte(leaderData))
+		name, ok := val.(string)
+		if ok && member.Name == name {
+			failpoint.Return(errors.Errorf("failed to grant lease"))
+		}
+	})
+
 	if err := newLease.Grant(leaseTimeout); err != nil {
 		return err
 	}

diff --git a/pkg/errs/errno.go b/pkg/errs/errno.go
@@ -208,14 +208,15 @@ var (
 
 // server errors
 var (
-	ErrServiceRegistered     = errors.Normalize("service with path [%s] already registered", errors.RFCCodeText("PD:server:ErrServiceRegistered"))
-	ErrAPIInformationInvalid = errors.Normalize("invalid api information, group %s version %s", errors.RFCCodeText("PD:server:ErrAPIInformationInvalid"))
-	ErrClientURLEmpty        = errors.Normalize("client url empty", errors.RFCCodeText("PD:server:ErrClientEmpty"))
-	ErrLeaderNil             = errors.Normalize("leader is nil", errors.RFCCodeText("PD:server:ErrLeaderNil"))
-	ErrCancelStartEtcd       = errors.Normalize("etcd start canceled", errors.RFCCodeText("PD:server:ErrCancelStartEtcd"))
-	ErrConfigItem            = errors.Normalize("cannot set invalid configuration", errors.RFCCodeText("PD:server:ErrConfiguration"))
-	ErrServerNotStarted      = errors.Normalize("server not started", errors.RFCCodeText("PD:server:ErrServerNotStarted"))
-	ErrRateLimitExceeded     = errors.Normalize("rate limit exceeded", errors.RFCCodeText("PD:server:ErrRateLimitExceeded"))
+	ErrServiceRegistered      = errors.Normalize("service with path [%s] already registered", errors.RFCCodeText("PD:server:ErrServiceRegistered"))
+	ErrAPIInformationInvalid  = errors.Normalize("invalid api information, group %s version %s", errors.RFCCodeText("PD:server:ErrAPIInformationInvalid"))
+	ErrClientURLEmpty         = errors.Normalize("client url empty", errors.RFCCodeText("PD:server:ErrClientEmpty"))
+	ErrLeaderNil              = errors.Normalize("leader is nil", errors.RFCCodeText("PD:server:ErrLeaderNil"))
+	ErrCancelStartEtcd        = errors.Normalize("etcd start canceled", errors.RFCCodeText("PD:server:ErrCancelStartEtcd"))
+	ErrConfigItem             = errors.Normalize("cannot set invalid configuration", errors.RFCCodeText("PD:server:ErrConfiguration"))
+	ErrServerNotStarted       = errors.Normalize("server not started", errors.RFCCodeText("PD:server:ErrServerNotStarted"))
+	ErrRateLimitExceeded      = errors.Normalize("rate limit exceeded", errors.RFCCodeText("PD:server:ErrRateLimitExceeded"))
+	ErrLeaderFrequentlyChange = errors.Normalize("leader %s frequently changed, leader-key is [%s]", errors.RFCCodeText("PD:server:ErrLeaderFrequentlyChange"))
 )
 
 // logutil errors

diff --git a/pkg/member/member.go b/pkg/member/member.go
@@ -185,12 +185,15 @@ func (m *EmbeddedEtcdMember) CampaignLeader(ctx context.Context, leaseTimeout in
 	failpoint.Inject("skipCampaignLeaderCheck", func() {
 		failpoint.Return(m.leadership.Campaign(leaseTimeout, m.MemberValue()))
 	})
+
 	if m.leadership.GetCampaignTimesNum() >= campaignLeaderFrequencyTimes {
-		log.Warn("campaign times is too frequent, resign and campaign again",
-			zap.String("leader-name", m.Name()), zap.String("leader-key", m.GetLeaderPath()))
 		m.leadership.ResetCampaignTimes()
-		return m.ResignEtcdLeader(ctx, m.Name(), "")
+		if err := m.ResignEtcdLeader(ctx, m.Name(), ""); err != nil {
+			return err
+		}
+		return errs.ErrLeaderFrequentlyChange.FastGenByArgs(m.Name(), m.GetLeaderPath())
 	}
+
 	return m.leadership.Campaign(leaseTimeout, m.MemberValue())
 }
 

diff --git a/tests/server/member/member_test.go b/tests/server/member/member_test.go
@@ -346,6 +346,31 @@ func TestCampaignLeaderFrequently(t *testing.T) {
 	re.NotEqual(leader, cluster.GetLeader())
 }
 
+func TestGrantLeaseFailed(t *testing.T) {
+	re := require.New(t)
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	cluster, err := tests.NewTestCluster(ctx, 5)
+	defer cluster.Destroy()
+	re.NoError(err)
+
+	err = cluster.RunInitialServers()
+	re.NoError(err)
+	cluster.WaitLeader()
+	leader := cluster.GetLeader()
+	re.NotEmpty(cluster.GetLeader())
+	re.NoError(failpoint.Enable("github.com/tikv/pd/pkg/election/skipGrantLeader", fmt.Sprintf("return(\"%s\")", leader)))
+
+	for i := 0; i < 3; i++ {
+		cluster.GetLeaderServer().ResetPDLeader()
+		cluster.WaitLeader()
+	}
+	// PD leader should be different from before because etcd leader changed.
+	re.NotEmpty(cluster.GetLeader())
+	re.NotEqual(leader, cluster.GetLeader())
+	re.NoError(failpoint.Disable("github.com/tikv/pd/pkg/election/skipGrantLeader"))
+}
+
 func TestGetLeader(t *testing.T) {
 	re := require.New(t)
 	ctx, cancel := context.WithCancel(context.Background())