[release-3.5] [Solution 1] Auto sync members in v3store if Islearner is the only field that differs between v2store and v3store

server/embed/etcd.go

@@ -299,6 +299,9 @@ func StartEtcd(inCfg *Config) (e *Etcd, err error) {
 		zap.Strings("listen-metrics-urls", e.cfg.getMetricsURLs()),
 	)
 	serving = true
+
+	e.Server.SyncLearnerPromotionIfNeeded()
+
 	return e, nil
 }
 

server/etcdserver/api/membership/cluster.go

@@ -23,6 +23,7 @@ import (
 	"errors"
 	"fmt"
 	"path"
+	"reflect"
 	"sort"
 	"strings"
 	"sync"
@@ -542,6 +543,53 @@ func (c *RaftCluster) PromoteMember(id types.ID, shouldApplyV3 ShouldApplyV3) {
 	)
 }
 
+func (c *RaftCluster) SyncLearnerPromotionIfNeeded() {
+	c.Lock()
+	defer c.Unlock()
+
+	v2Members, _ := membersFromStore(c.lg, c.v2store)
+	v3Members, _ := membersFromBackend(c.lg, c.be)
+
+	for id, v2Member := range v2Members {
+		v3Member, ok := v3Members[id]
+
+		if !ok {
+			c.lg.Error("Detected member only in v2store but missing in v3store", zap.String("member", fmt.Sprintf("%+v", *v2Member)))
+			continue
+		}
+
+		// A peerURL list is considered the same regardless of order.
+		clonedV2Member := v2Member.Clone()
+		sort.Strings(clonedV2Member.PeerURLs)
+
+		clonedV3Member := v3Member.Clone()
+		sort.Strings(clonedV3Member.PeerURLs)
+
+		if reflect.DeepEqual(clonedV2Member.RaftAttributes, clonedV3Member.RaftAttributes) {
+			c.lg.Info("Member's RaftAttributes is consistent between v2store and v3store", zap.String("member", fmt.Sprintf("%+v", *v2Member)))
+			continue
+		}
+
+		// Sync member iff both conditions below are true,
+		//  1. IsLearner is the only field in RaftAttributes that differs between v2store and v3store.
+		//  2. v2store.IsLearner == false && v3store.IsLearner == true.
+		clonedV3Member.IsLearner = false
+		if reflect.DeepEqual(clonedV2Member.RaftAttributes, clonedV3Member.RaftAttributes) {
+			syncedV3Member := v3Member.Clone()
+			syncedV3Member.IsLearner = false
+			c.lg.Warn("Syncing member in v3store", zap.String("member", fmt.Sprintf("%+v", *syncedV3Member)))
+			unsafeHackySaveMemberToBackend(c.lg, c.be, syncedV3Member)
+			c.be.ForceCommit()
+			continue
+		}
+
+		c.lg.Error("Cannot sync member in v3store due to IsLearner not being the only field that differs between v2store amd v3store",
+			zap.String("v2member", fmt.Sprintf("%+v", *v2Member)),
+			zap.String("v3member", fmt.Sprintf("%+v", *v3Member)),
+		)
+	}
+}
+
 func (c *RaftCluster) UpdateRaftAttributes(id types.ID, raftAttr RaftAttributes, shouldApplyV3 ShouldApplyV3) {
 	c.Lock()
 	defer c.Unlock()

server/etcdserver/api/membership/cluster_test.go

@@ -24,7 +24,7 @@ import (
 
 	"github.com/coreos/go-semver/semver"
 	"github.com/stretchr/testify/assert"
-	betesting "go.etcd.io/etcd/server/v3/mvcc/backend/testing"
+	"github.com/stretchr/testify/require"
 	"go.uber.org/zap"
 	"go.uber.org/zap/zaptest"
 
@@ -33,6 +33,7 @@ import (
 	"go.etcd.io/etcd/raft/v3/raftpb"
 	"go.etcd.io/etcd/server/v3/etcdserver/api/v2store"
 	"go.etcd.io/etcd/server/v3/mock/mockstore"
+	betesting "go.etcd.io/etcd/server/v3/mvcc/backend/testing"
 )
 
 func TestClusterMember(t *testing.T) {
@@ -1210,3 +1211,160 @@ func TestRemoveMemberSyncsBackendAndStoreV2(t *testing.T) {
 		})
 	}
 }
+
+func TestSyncLearnerPromotion(t *testing.T) {
+	tcs := []struct {
+		name string
+
+		storeV2Members []*Member
+		backendMembers []*Member
+
+		expectV3Members map[types.ID]*Member
+	}{
+		{
+			name: "v3store should keep unchanged if IsLearner isn't the only field that differs",
+			storeV2Members: []*Member{
+				{
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.100:2380"},
+						IsLearner: false,
+					},
+				},
+			},
+			backendMembers: []*Member{
+				{
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.9:2380"},
+						IsLearner: true,
+					},
+				},
+			},
+			expectV3Members: map[types.ID]*Member{
+				100: {
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.9:2380"},
+						IsLearner: true,
+					},
+				},
+			},
+		},
+		{
+			name: "v3store should keep unchanged if IsLearner is the only field that differs but v3store.IsLearner is false",
+			storeV2Members: []*Member{
+				{
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.9:2380"},
+						IsLearner: true,
+					},
+				},
+			},
+			backendMembers: []*Member{
+				{
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.9:2380"},
+						IsLearner: false,
+					},
+				},
+			},
+			expectV3Members: map[types.ID]*Member{
+				100: {
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.9:2380"},
+						IsLearner: false,
+					},
+				},
+			},
+		},
+		{
+			name: "v3store should be updated if IsLearner is the only field that differs and v3store.IsLearner is true",
+			storeV2Members: []*Member{
+				{
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.9:2380"},
+						IsLearner: false,
+					},
+				},
+			},
+			backendMembers: []*Member{
+				{
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.9:2380"},
+						IsLearner: true,
+					},
+				},
+			},
+			expectV3Members: map[types.ID]*Member{
+				100: {
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.9:2380"},
+						IsLearner: false,
+					},
+				},
+			},
+		},
+		{
+			name: "v3store should be updated if IsLearner is the only field that differs and peerURLs are in different order in v2store and v3store",
+			storeV2Members: []*Member{
+				{
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://10.0.0.9:2380", "http://127.0.0.1:2380"},
+						IsLearner: false,
+					},
+				},
+			},
+			backendMembers: []*Member{
+				{
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://127.0.0.1:2380", "http://10.0.0.9:2380"},
+						IsLearner: true,
+					},
+				},
+			},
+			expectV3Members: map[types.ID]*Member{
+				100: {
+					ID: 100,
+					RaftAttributes: RaftAttributes{
+						PeerURLs:  []string{"http://127.0.0.1:2380", "http://10.0.0.9:2380"},
+						IsLearner: false,
+					},
+				},
+			},
+		},
+	}
+	for _, tc := range tcs {
+		t.Run(tc.name, func(t *testing.T) {
+			lg := zaptest.NewLogger(t)
+			be, _ := betesting.NewDefaultTmpBackend(t)
+			defer be.Close()
+			mustCreateBackendBuckets(be)
+			for _, m := range tc.backendMembers {
+				unsafeSaveMemberToBackend(lg, be, m)
+			}
+			be.ForceCommit()
+
+			st := v2store.New()
+			for _, m := range tc.storeV2Members {
+				mustSaveMemberToStore(lg, st, m)
+			}
+
+			cluster := NewCluster(lg)
+			cluster.SetBackend(be)
+			cluster.SetStore(st)
+
+			cluster.SyncLearnerPromotionIfNeeded()
+			v3Members, _ := cluster.MembersFromBackend()
+			require.Equal(t, tc.expectV3Members, v3Members)
+		})
+	}
+}

server/etcdserver/api/membership/store.go

@@ -58,6 +58,22 @@ func unsafeSaveMemberToBackend(lg *zap.Logger, be backend.Backend, m *Member) er
 	return nil
 }
 
+// unsafeHackySaveMemberToBackend updates the member in a hacky way.
+// It's only used to workaround https://github.com/etcd-io/etcd/issues/19557.
+func unsafeHackySaveMemberToBackend(lg *zap.Logger, be backend.Backend, m *Member) error {
+	mkey := backendMemberKey(m.ID)
+	mvalue, err := json.Marshal(m)
+	if err != nil {
+		lg.Panic("failed to marshal member", zap.Error(err))
+	}
+
+	tx := be.BatchTx()
+	tx.LockOutsideApply()
+	defer tx.Unlock()
+	tx.UnsafePut(buckets.Members, mkey, mvalue)
+	return nil
+}
+
 // TrimClusterFromBackend removes all information about cluster (versions)
 // from the v3 backend.
 func TrimClusterFromBackend(be backend.Backend) error {

server/etcdserver/server.go

@@ -2931,3 +2931,25 @@ func maybeDefragBackend(cfg config.ServerConfig, be backend.Backend) error {
 func (s *EtcdServer) CorruptionChecker() CorruptionChecker {
 	return s.corruptionChecker
 }
+
+// SyncLearnerPromotionIfNeeded provides a workaround for the users who have
+// already been affected by https://github.com/etcd-io/etcd/issues/19557.
+// It automatically syncs the v3store (bbolt) from v2store iff all the
+// conditions below are true for each member,
+//  1. IsLearner is the only field that differs between v2store and v3store.
+//  2. v2store.IsLearner == false && v3store.IsLearner == true.
+//  3. etcd is ready to serve client requests, which means it has finished
+//     replaying the WAL records.
+func (s *EtcdServer) SyncLearnerPromotionIfNeeded() {
+	lg := s.Logger()
+	select {
+	case <-s.StoppingNotify():
+		lg.Warn("stop sync learner promotion operations as the server is stopping")
+		return
+	case <-s.ReadyNotify():
+	}
+
+	lg.Info("Trying to sync the learner promotion operations for v3store if needed")
+	s.cluster.SyncLearnerPromotionIfNeeded()
+	lg.Info("Finished syncing the learner promotion operations for v3store")
+}

tests/e2e/ctl_v3_member_test.go

@@ -24,7 +24,12 @@ import (
 
 	"github.com/stretchr/testify/require"
 
+	"go.etcd.io/bbolt"
 	"go.etcd.io/etcd/api/v3/etcdserverpb"
+	"go.etcd.io/etcd/client/pkg/v3/types"
+	"go.etcd.io/etcd/server/v3/datadir"
+	"go.etcd.io/etcd/server/v3/etcdserver/api/membership"
+	"go.etcd.io/etcd/server/v3/mvcc/buckets"
 	"go.etcd.io/etcd/tests/v3/framework/e2e"
 )
 
@@ -255,6 +260,75 @@ func TestCtlV3PromotingLearner(t *testing.T) {
 	require.NoError(t, err)
 
 	t.Logf("Promoting the learner %x", learnerID)
-	_, err = etcdctl.MemberPromote(learnerID)
+	resp, err := etcdctl.MemberPromote(learnerID)
 	require.NoError(t, err)
+
+	var promotedMember *etcdserverpb.Member
+	for _, m := range resp.Members {
+		if m.ID == learnerID {
+			promotedMember = m
+			break
+		}
+	}
+	require.NotNil(t, promotedMember)
+	t.Logf("The promoted member: %+v", promotedMember)
+
+	t.Log("Ensure all members are voting members")
+	ensureAllMembersAreVotingMembers(t, etcdctl)
+
+	t.Logf("Stopping the first member")
+	require.NoError(t, epc.Procs[0].Stop())
+
+	t.Log("Manually changing the already promoted learner to a learner again")
+	promotedMember.IsLearner = true
+	mustSaveMemberIntoBbolt(t, epc.Procs[0].Config().DataDirPath, promotedMember)
+
+	t.Log("Starting the first member again")
+	require.NoError(t, epc.Procs[0].Start())
+
+	t.Log("Checking the auto-sync learner log message")
+	e2e.AssertProcessLogs(t, epc.Procs[0], "Syncing member in v3store")
+
+	t.Log("Ensure all members are voting members again")
+	ensureAllMembersAreVotingMembers(t, etcdctl)
+}
+
+func mustSaveMemberIntoBbolt(t *testing.T, dataDir string, protoMember *etcdserverpb.Member) {
+	dbPath := datadir.ToBackendFileName(dataDir)
+	db, err := bbolt.Open(dbPath, 0666, nil)
+	require.NoError(t, err)
+	defer func() {
+		require.NoError(t, db.Close())
+	}()
+
+	m := &membership.Member{
+		ID: types.ID(protoMember.ID),
+		RaftAttributes: membership.RaftAttributes{
+			PeerURLs:  protoMember.PeerURLs,
+			IsLearner: protoMember.IsLearner,
+		},
+		Attributes: membership.Attributes{
+			Name:       protoMember.Name,
+			ClientURLs: protoMember.ClientURLs,
+		},
+	}
+
+	err = db.Update(func(tx *bbolt.Tx) error {
+		b := tx.Bucket(buckets.Members.Name())
+
+		mkey := []byte(m.ID.String())
+		mvalue, err := json.Marshal(m)
+		require.NoError(t, err)
+
+		return b.Put(mkey, mvalue)
+	})
+	require.NoError(t, err)
+}
+
+func ensureAllMembersAreVotingMembers(t *testing.T, etcdctl *e2e.Etcdctl) {
+	memberListResp, err := etcdctl.MemberList()
+	require.NoError(t, err)
+	for _, m := range memberListResp.Members {
+		require.False(t, m.IsLearner)
+	}
 }