Merge pull request #11725 from YoyinZyc/downgrade-api

[Etcd Downgrade] Store downgrade info to backend

etcdserver/api/membership/cluster.go

@@ -59,6 +59,16 @@ type RaftCluster struct {
 	// removed contains the ids of removed members in the cluster.
 	// removed id cannot be reused.
 	removed map[types.ID]bool
+
+	downgradeInfo *DowngradeInfo
+}
+
+type DowngradeInfo struct {
+	// TargetVersion is the target downgrade version, if the cluster is not under downgrading,
+	// the targetVersion will be an empty string
+	TargetVersion string `json:"target-version"`
+	// Enabled indicates whether the cluster is enabled to downgrade
+	Enabled bool `json:"enabled"`
 }
 
 // ConfigChangeContext represents a context for confChange.
@@ -106,6 +116,7 @@ func NewCluster(lg *zap.Logger, token string) *RaftCluster {
 		token:         token,
 		members:       make(map[types.ID]*Member),
 		removed:       make(map[types.ID]bool),
+		downgradeInfo: &DowngradeInfo{Enabled: false},
 	}
 }
 
@@ -691,6 +702,39 @@ func clusterVersionFromBackend(lg *zap.Logger, be backend.Backend) *semver.Versi
 	return semver.Must(semver.NewVersion(string(vals[0])))
 }
 
+func downgradeInfoFromBackend(lg *zap.Logger, be backend.Backend) *DowngradeInfo {
+	dkey := backendDowngradeKey()
+	tx := be.ReadTx()
+	tx.Lock()
+	defer tx.Unlock()
+	keys, vals := tx.UnsafeRange(clusterBucketName, dkey, nil, 0)
+	if len(keys) == 0 {
+		return nil
+	}
+
+	if len(keys) != 1 {
+		lg.Panic(
+			"unexpected number of keys when getting cluster version from backend",
+			zap.Int("number-of-key", len(keys)),
+		)
+	}
+	var d DowngradeInfo
+	if err := json.Unmarshal(vals[0], &d); err != nil {
+		lg.Panic("failed to unmarshal downgrade information", zap.Error(err))
+	}
+
+	// verify the downgrade info from backend
+	if d.Enabled {
+		if _, err := semver.NewVersion(d.TargetVersion); err != nil {
+			lg.Panic(
+				"unexpected version format of the downgrade target version from backend",
+				zap.String("target-version", d.TargetVersion),
+			)
+		}
+	}
+	return &d
+}
+
 // ValidateClusterAndAssignIDs validates the local cluster by matching the PeerURLs
 // with the existing cluster. If the validation succeeds, it assigns the IDs
 // from the existing cluster to the local cluster.
@@ -752,6 +796,36 @@ func (c *RaftCluster) IsLocalMemberLearner() bool {
 	return localMember.IsLearner
 }
 
+// DowngradeInfo returns the downgrade status of the cluster
+func (c *RaftCluster) DowngradeInfo() *DowngradeInfo {
+	c.Lock()
+	defer c.Unlock()
+	if c.downgradeInfo == nil {
+		return &DowngradeInfo{Enabled: false}
+	}
+	d := &DowngradeInfo{Enabled: c.downgradeInfo.Enabled, TargetVersion: c.downgradeInfo.TargetVersion}
+	return d
+}
+
+func (c *RaftCluster) SetDowngradeInfo(d *DowngradeInfo) {
+	c.Lock()
+	defer c.Unlock()
+
+	if c.be != nil {
+		mustSaveDowngradeToBackend(c.lg, c.be, d)
+	}
+
+	c.downgradeInfo = d
+
+	if d.Enabled {
+		c.lg.Info(
+			"The server is ready to downgrade",
+			zap.String("target-version", d.TargetVersion),
+			zap.String("server-version", version.Version),
+		)
+	}
+}
+
 // IsMemberExist returns if the member with the given id exists in cluster.
 func (c *RaftCluster) IsMemberExist(id types.ID) bool {
 	c.Lock()

etcdserver/api/membership/store.go

@@ -53,8 +53,8 @@ func mustSaveMemberToBackend(lg *zap.Logger, be backend.Backend, m *Member) {
 
 	tx := be.BatchTx()
 	tx.Lock()
+	defer tx.Unlock()
 	tx.UnsafePut(membersBucketName, mkey, mvalue)
-	tx.Unlock()
 }
 
 func mustDeleteMemberFromBackend(be backend.Backend, id types.ID) {
@@ -62,9 +62,9 @@ func mustDeleteMemberFromBackend(be backend.Backend, id types.ID) {
 
 	tx := be.BatchTx()
 	tx.Lock()
+	defer tx.Unlock()
 	tx.UnsafeDelete(membersBucketName, mkey)
 	tx.UnsafePut(membersRemovedBucketName, mkey, []byte("removed"))
-	tx.Unlock()
 }
 
 func mustSaveClusterVersionToBackend(be backend.Backend, ver *semver.Version) {
@@ -76,6 +76,18 @@ func mustSaveClusterVersionToBackend(be backend.Backend, ver *semver.Version) {
 	tx.UnsafePut(clusterBucketName, ckey, []byte(ver.String()))
 }
 
+func mustSaveDowngradeToBackend(lg *zap.Logger, be backend.Backend, downgrade *DowngradeInfo) {
+	dkey := backendDowngradeKey()
+	dvalue, err := json.Marshal(downgrade)
+	if err != nil {
+		lg.Panic("failed to marshal downgrade information", zap.Error(err))
+	}
+	tx := be.BatchTx()
+	tx.Lock()
+	defer tx.Unlock()
+	tx.UnsafePut(clusterBucketName, dkey, dvalue)
+}
+
 func mustSaveMemberToStore(lg *zap.Logger, s v2store.Store, m *Member) {
 	b, err := json.Marshal(m.RaftAttributes)
 	if err != nil {
@@ -184,6 +196,10 @@ func backendClusterVersionKey() []byte {
 	return []byte("clusterVersion")
 }
 
+func backendDowngradeKey() []byte {
+	return []byte("downgrade")
+}
+
 func mustCreateBackendBuckets(be backend.Backend) {
 	tx := be.BatchTx()
 	tx.Lock()