br: leader selection with tikv store balance score during the ebs data restore

br/pkg/errors/errors.go

@@ -83,8 +83,9 @@ var (
 	ErrStorageInvalidPermission = errors.Normalize("external storage permission", errors.RFCCodeText("BR:ExternalStorage:ErrStorageInvalidPermission"))
 
 	// Snapshot restore
-	ErrRestoreTotalKVMismatch = errors.Normalize("restore total tikvs mismatch", errors.RFCCodeText("BR:EBS:ErrRestoreTotalKVMismatch"))
-	ErrRestoreInvalidPeer     = errors.Normalize("restore met a invalid peer", errors.RFCCodeText("BR:EBS:ErrRestoreInvalidPeer"))
+	ErrRestoreTotalKVMismatch   = errors.Normalize("restore total tikvs mismatch", errors.RFCCodeText("BR:EBS:ErrRestoreTotalKVMismatch"))
+	ErrRestoreInvalidPeer       = errors.Normalize("restore met a invalid peer", errors.RFCCodeText("BR:EBS:ErrRestoreInvalidPeer"))
+	ErrRestoreRegionWithoutPeer = errors.Normalize("restore met a region without any peer", errors.RFCCodeText("BR:EBS:ErrRestoreRegionWithoutPeer"))
 
 	// Errors reported from TiKV.
 	ErrKVStorage           = errors.Normalize("tikv storage occur I/O error", errors.RFCCodeText("BR:KV:ErrKVStorage"))

br/pkg/restore/data.go

@@ -372,6 +372,7 @@ type RecoverRegion struct {
 // 2. build a leader list for all region during the tikv startup
 // 3. get max allocate id
 func (recovery *Recovery) MakeRecoveryPlan() error {
+	storeBalanceScore := make(map[uint64]int, len(recovery.allStores))
 	// Group region peer info by region id. find the max allocateId
 	// region [id] [peer[0-n]]
 	var regions = make(map[uint64][]*RecoverRegion, 0)
@@ -410,16 +411,20 @@ func (recovery *Recovery) MakeRecoveryPlan() error {
 			}
 		} else {
 			// Generate normal commands.
-			log.Debug("detected valid peer", zap.Uint64("region id", regionId))
-			for i, peer := range peers {
-				log.Debug("make plan", zap.Uint64("store id", peer.StoreId), zap.Uint64("region id", peer.RegionId))
-				plan := &recovpb.RecoverRegionRequest{RegionId: peer.RegionId, AsLeader: i == 0}
-				// sorted by log term -> last index -> commit index in a region
-				if plan.AsLeader {
-					log.Debug("as leader peer", zap.Uint64("store id", peer.StoreId), zap.Uint64("region id", peer.RegionId))
-					recovery.RecoveryPlan[peer.StoreId] = append(recovery.RecoveryPlan[peer.StoreId], plan)
-				}
+			log.Debug("detected valid region", zap.Uint64("region id", regionId))
+			// calc the leader candidates
+			leaderCandidates, err := LeaderCandidates(peers)
+			if err != nil {
+				return errors.Trace(err)
 			}
+
+			// select the leader base on tikv storeBalanceScore
+			leader := SelectRegionLeader(storeBalanceScore, leaderCandidates)
+			log.Debug("as leader peer", zap.Uint64("store id", leader.StoreId), zap.Uint64("region id", leader.RegionId))
+			plan := &recovpb.RecoverRegionRequest{RegionId: leader.RegionId, AsLeader: true}
+			recovery.RecoveryPlan[leader.StoreId] = append(recovery.RecoveryPlan[leader.StoreId], plan)
+			storeBalanceScore[leader.StoreId] += 1
+
 		}
 	}
 	return nil

br/pkg/restore/util.go

@@ -750,3 +750,49 @@ func CheckConsistencyAndValidPeer(regionInfos []*RecoverRegionInfo) (map[uint64]
 	}
 	return validPeers, nil
 }
+
+// in cloud, since iops and bandwidth limitation, write operator in raft is slow, so raft state (logterm, lastlog, commitlog...) are the same among the peers
+// LeaderCandidates select all peers can be select as a leader during the restore
+func LeaderCandidates(peers []*RecoverRegion) ([]*RecoverRegion, error) {
+	if peers == nil {
+		log.Warn("region without peer")
+		return nil, errors.Annotatef(berrors.ErrRestoreRegionWithoutPeer,
+			"invalid region range")
+	}
+	candidates := make([]*RecoverRegion, 0, len(peers))
+	// by default, the peers[0] to be assign as a leader, since peers already sorted by leader selection rule
+	leader := peers[0]
+	candidates = append(candidates, leader)
+	for i, peer := range peers {
+		if i == 0 {
+			continue
+		}
+		// qualificated candidate is leader.logterm = candidate.logterm && leader.lastindex = candidate.lastindex && && leader.commitindex = candidate.commitindex
+		if peer.LastLogTerm == leader.LastLogTerm && peer.LastIndex == leader.LastIndex && peer.CommitIndex == leader.CommitIndex {
+			log.Debug("leader candidate", zap.Uint64("store id", peer.StoreId), zap.Uint64("region id", peer.RegionId), zap.Uint64("peer id", peer.PeerId))
+			candidates = append(candidates, peer)
+		}
+
+	}
+	return candidates, nil
+}
+
+// for region A, has candidate leader x, y, z
+// peer x on store 1 with storeBalanceScore 3
+// peer y on store 3 with storeBalanceScore 2
+// peer z on store 4 with storeBalanceScore 1
+// result: peer z will be select as leader on store 4
+func SelectRegionLeader(storeBalanceScore map[uint64]int, peers []*RecoverRegion) *RecoverRegion {
+	// by default, the peers[0] to be assign as a leader
+	leader := peers[0]
+	minLeaderStore := storeBalanceScore[leader.StoreId]
+	for _, peer := range peers {
+		log.Debug("leader candidate", zap.Int("score", storeBalanceScore[peer.StoreId]), zap.Int("min-score", minLeaderStore), zap.Uint64("store id", peer.StoreId), zap.Uint64("region id", peer.RegionId), zap.Uint64("peer id", peer.PeerId))
+		if storeBalanceScore[peer.StoreId] < minLeaderStore {
+			minLeaderStore = storeBalanceScore[peer.StoreId]
+			leader = peer
+		}
+	}
+
+	return leader
+}

br/pkg/restore/util_test.go

@@ -460,3 +460,53 @@ func TestCheckConsistencyAndValidPeer(t *testing.T) {
 	require.Error(t, err)
 	require.Regexp(t, ".*invalid restore range.*", err.Error())
 }
+
+func TestLeaderCandidates(t *testing.T) {
+	//key space is continuous
+	validPeer1 := newPeerMeta(9, 11, 2, []byte(""), []byte("bb"), 2, 1, 0, 0, false)
+	validPeer2 := newPeerMeta(19, 22, 3, []byte("bb"), []byte("cc"), 2, 1, 0, 1, false)
+	validPeer3 := newPeerMeta(29, 30, 1, []byte("cc"), []byte(""), 2, 1, 0, 2, false)
+
+	peers := []*restore.RecoverRegion{
+		validPeer1,
+		validPeer2,
+		validPeer3,
+	}
+
+	candidates, err := restore.LeaderCandidates(peers)
+	require.NoError(t, err)
+	require.Equal(t, 3, len(candidates))
+}
+
+func TestSelectRegionLeader(t *testing.T) {
+
+	validPeer1 := newPeerMeta(9, 11, 2, []byte(""), []byte("bb"), 2, 1, 0, 0, false)
+	validPeer2 := newPeerMeta(19, 22, 3, []byte("bb"), []byte("cc"), 2, 1, 0, 1, false)
+	validPeer3 := newPeerMeta(29, 30, 1, []byte("cc"), []byte(""), 2, 1, 0, 2, false)
+
+	peers := []*restore.RecoverRegion{
+		validPeer1,
+		validPeer2,
+		validPeer3,
+	}
+	// init store banlance score all is 0
+	storeBalanceScore := make(map[uint64]int, len(peers))
+	leader := restore.SelectRegionLeader(storeBalanceScore, peers)
+	require.Equal(t, validPeer1, leader)
+
+	// change store banlance store
+	storeBalanceScore[2] = 3
+	storeBalanceScore[3] = 2
+	storeBalanceScore[1] = 1
+	leader = restore.SelectRegionLeader(storeBalanceScore, peers)
+	require.Equal(t, validPeer3, leader)
+
+	// one peer
+	peer := []*restore.RecoverRegion{
+		validPeer3,
+	}
+	// init store banlance score all is 0
+	storeScore := make(map[uint64]int, len(peer))
+	leader = restore.SelectRegionLeader(storeScore, peer)
+	require.Equal(t, validPeer3, leader)
+}