Golang topo.TabletAliasEqual函数代码示例

// updateShardCellsAndMaster will update the 'Cells' and possibly
// MasterAlias records for the shard, if needed.
func (wr *Wrangler) updateShardCellsAndMaster(ctx context.Context, si *topo.ShardInfo, tabletAlias *pb.TabletAlias, tabletType pb.TabletType, force bool) error {
	// See if we need to update the Shard:
	// - add the tablet's cell to the shard's Cells if needed
	// - change the master if needed
	shardUpdateRequired := false
	if !si.HasCell(tabletAlias.Cell) {
		shardUpdateRequired = true
	}
	if tabletType == pb.TabletType_MASTER && !topo.TabletAliasEqual(si.MasterAlias, tabletAlias) {
		shardUpdateRequired = true
	}
	if !shardUpdateRequired {
		return nil
	}

	actionNode := actionnode.UpdateShard()
	keyspace := si.Keyspace()
	shard := si.ShardName()
	lockPath, err := wr.lockShard(ctx, keyspace, shard, actionNode)
	if err != nil {
		return err
	}

	// re-read the shard with the lock
	si, err = wr.ts.GetShard(ctx, keyspace, shard)
	if err != nil {
		return wr.unlockShard(ctx, keyspace, shard, actionNode, lockPath, err)
	}

	// update it
	wasUpdated := false
	if !si.HasCell(tabletAlias.Cell) {
		si.Cells = append(si.Cells, tabletAlias.Cell)
		wasUpdated = true
	}
	if tabletType == pb.TabletType_MASTER && !topo.TabletAliasEqual(si.MasterAlias, tabletAlias) {
		if !topo.TabletAliasIsZero(si.MasterAlias) && !force {
			return wr.unlockShard(ctx, keyspace, shard, actionNode, lockPath, fmt.Errorf("creating this tablet would override old master %v in shard %v/%v", topo.TabletAliasString(si.MasterAlias), keyspace, shard))
		}
		si.MasterAlias = tabletAlias
		wasUpdated = true
	}

	if wasUpdated {
		// write it back
		if err := topo.UpdateShard(ctx, wr.ts, si); err != nil {
			return wr.unlockShard(ctx, keyspace, shard, actionNode, lockPath, err)
		}
	}

	// and unlock
	return wr.unlockShard(ctx, keyspace, shard, actionNode, lockPath, err)
}

// ValidateVersionKeyspace validates all versions are the same in all
// tablets in a keyspace
func (wr *Wrangler) ValidateVersionKeyspace(ctx context.Context, keyspace string) error {
	// find all the shards
	shards, err := wr.ts.GetShardNames(ctx, keyspace)
	if err != nil {
		return err
	}

	// corner cases
	if len(shards) == 0 {
		return fmt.Errorf("No shards in keyspace %v", keyspace)
	}
	sort.Strings(shards)
	if len(shards) == 1 {
		return wr.ValidateVersionShard(ctx, keyspace, shards[0])
	}

	// find the reference version using the first shard's master
	si, err := wr.ts.GetShard(ctx, keyspace, shards[0])
	if err != nil {
		return err
	}
	if topo.TabletAliasIsZero(si.MasterAlias) {
		return fmt.Errorf("No master in shard %v/%v", keyspace, shards[0])
	}
	referenceAlias := si.MasterAlias
	log.Infof("Gathering version for reference master %v", topo.TabletAliasString(referenceAlias))
	referenceVersion, err := wr.GetVersion(ctx, referenceAlias)
	if err != nil {
		return err
	}

	// then diff with all tablets but master 0
	er := concurrency.AllErrorRecorder{}
	wg := sync.WaitGroup{}
	for _, shard := range shards {
		aliases, err := topo.FindAllTabletAliasesInShard(ctx, wr.ts, keyspace, shard)
		if err != nil {
			er.RecordError(err)
			continue
		}

		for _, alias := range aliases {
			if topo.TabletAliasEqual(alias, si.MasterAlias) {
				continue
			}

			wg.Add(1)
			go wr.diffVersion(ctx, referenceVersion, referenceAlias, alias, &wg, &er)
		}
	}
	wg.Wait()
	if er.HasErrors() {
		return fmt.Errorf("Version diffs:\n%v", er.Error().Error())
	}
	return nil
}

// FIXME(msolomon) This validate presumes the master is up and running.
// Even when that isn't true, there are validation processes that might be valuable.
func (wr *Wrangler) validateShard(ctx context.Context, keyspace, shard string, pingTablets bool, wg *sync.WaitGroup, results chan<- error) {
	shardInfo, err := wr.ts.GetShard(ctx, keyspace, shard)
	if err != nil {
		results <- fmt.Errorf("TopologyServer.GetShard(%v, %v) failed: %v", keyspace, shard, err)
		return
	}

	aliases, err := topo.FindAllTabletAliasesInShard(ctx, wr.ts, keyspace, shard)
	if err != nil {
		results <- fmt.Errorf("TopologyServer.FindAllTabletAliasesInShard(%v, %v) failed: %v", keyspace, shard, err)
		return
	}

	tabletMap, _ := topo.GetTabletMap(ctx, wr.ts, aliases)

	var masterAlias *pb.TabletAlias
	for _, alias := range aliases {
		tabletInfo, ok := tabletMap[*alias]
		if !ok {
			results <- fmt.Errorf("tablet %v not found in map", alias)
			continue
		}
		if tabletInfo.Type == pb.TabletType_MASTER {
			if masterAlias != nil {
				results <- fmt.Errorf("shard %v/%v already has master %v but found other master %v", keyspace, shard, masterAlias, alias)
			} else {
				masterAlias = alias
			}
		}
	}

	if masterAlias == nil {
		results <- fmt.Errorf("no master for shard %v/%v", keyspace, shard)
	} else if !topo.TabletAliasEqual(shardInfo.MasterAlias, masterAlias) {
		results <- fmt.Errorf("master mismatch for shard %v/%v: found %v, expected %v", keyspace, shard, masterAlias, shardInfo.MasterAlias)
	}

	for _, alias := range aliases {
		wg.Add(1)
		go func(alias *pb.TabletAlias) {
			defer wg.Done()
			if err := topo.Validate(ctx, wr.ts, alias); err != nil {
				results <- fmt.Errorf("Validate(%v) failed: %v", alias, err)
			} else {
				wr.Logger().Infof("tablet %v is valid", alias)
			}
		}(alias)
	}

	if pingTablets {
		wr.validateReplication(ctx, shardInfo, tabletMap, results)
		wr.pingTablets(ctx, tabletMap, wg, results)
	}

	return
}

// ValidateVersionShard validates all versions are the same in all
// tablets in a shard
func (wr *Wrangler) ValidateVersionShard(ctx context.Context, keyspace, shard string) error {
	si, err := wr.ts.GetShard(ctx, keyspace, shard)
	if err != nil {
		return err
	}

	// get version from the master, or error
	if topo.TabletAliasIsZero(si.MasterAlias) {
		return fmt.Errorf("No master in shard %v/%v", keyspace, shard)
	}
	log.Infof("Gathering version for master %v", topo.TabletAliasString(si.MasterAlias))
	masterVersion, err := wr.GetVersion(ctx, si.MasterAlias)
	if err != nil {
		return err
	}

	// read all the aliases in the shard, that is all tablets that are
	// replicating from the master
	aliases, err := topo.FindAllTabletAliasesInShard(ctx, wr.ts, keyspace, shard)
	if err != nil {
		return err
	}

	// then diff with all slaves
	er := concurrency.AllErrorRecorder{}
	wg := sync.WaitGroup{}
	for _, alias := range aliases {
		if topo.TabletAliasEqual(alias, si.MasterAlias) {
			continue
		}

		wg.Add(1)
		go wr.diffVersion(ctx, masterVersion, si.MasterAlias, alias, &wg, &er)
	}
	wg.Wait()
	if er.HasErrors() {
		return fmt.Errorf("Version diffs:\n%v", er.Error().Error())
	}
	return nil
}

// Scrap a tablet. If force is used, we write to topo.Server
// directly and don't remote-execute the command.
//
// If we scrap the master for a shard, we will clear its record
// from the Shard object (only if that was the right master)
func (wr *Wrangler) Scrap(ctx context.Context, tabletAlias topo.TabletAlias, force, skipRebuild bool) error {
	// load the tablet, see if we'll need to rebuild
	ti, err := wr.ts.GetTablet(ctx, tabletAlias)
	if err != nil {
		return err
	}
	rebuildRequired := ti.IsInServingGraph()
	wasMaster := ti.Type == topo.TYPE_MASTER

	if force {
		err = topotools.Scrap(ctx, wr.ts, ti.Alias, force)
	} else {
		err = wr.tmc.Scrap(ctx, ti)
	}
	if err != nil {
		return err
	}

	if !rebuildRequired {
		wr.Logger().Infof("Rebuild not required")
		return nil
	}
	if skipRebuild {
		wr.Logger().Warningf("Rebuild required, but skipping it")
		return nil
	}

	// update the Shard object if the master was scrapped
	if wasMaster {
		actionNode := actionnode.UpdateShard()
		lockPath, err := wr.lockShard(ctx, ti.Keyspace, ti.Shard, actionNode)
		if err != nil {
			return err
		}

		// read the shard with the lock
		si, err := wr.ts.GetShard(ctx, ti.Keyspace, ti.Shard)
		if err != nil {
			return wr.unlockShard(ctx, ti.Keyspace, ti.Shard, actionNode, lockPath, err)
		}

		// update it if the right alias is there
		if topo.TabletAliasEqual(si.MasterAlias, topo.TabletAliasToProto(tabletAlias)) {
			si.MasterAlias = nil

			// write it back
			if err := topo.UpdateShard(ctx, wr.ts, si); err != nil {
				return wr.unlockShard(ctx, ti.Keyspace, ti.Shard, actionNode, lockPath, err)
			}
		} else {
			wr.Logger().Warningf("Scrapping master %v from shard %v/%v but master in Shard object was %v", tabletAlias, ti.Keyspace, ti.Shard, si.MasterAlias)
		}

		// and unlock
		if err := wr.unlockShard(ctx, ti.Keyspace, ti.Shard, actionNode, lockPath, err); err != nil {
			return err
		}
	}

	// and rebuild the original shard
	_, err = wr.RebuildShardGraph(ctx, ti.Keyspace, ti.Shard, []string{ti.Alias.Cell})
	return err
}

// ValidateSchemaKeyspace will diff the schema from all the tablets in
// the keyspace.
func (wr *Wrangler) ValidateSchemaKeyspace(ctx context.Context, keyspace string, excludeTables []string, includeViews bool) error {
	// find all the shards
	shards, err := wr.ts.GetShardNames(ctx, keyspace)
	if err != nil {
		return err
	}

	// corner cases
	if len(shards) == 0 {
		return fmt.Errorf("No shards in keyspace %v", keyspace)
	}
	sort.Strings(shards)
	if len(shards) == 1 {
		return wr.ValidateSchemaShard(ctx, keyspace, shards[0], excludeTables, includeViews)
	}

	// find the reference schema using the first shard's master
	si, err := wr.ts.GetShard(ctx, keyspace, shards[0])
	if err != nil {
		return err
	}
	if topo.TabletAliasIsZero(si.MasterAlias) {
		return fmt.Errorf("No master in shard %v/%v", keyspace, shards[0])
	}
	referenceAlias := si.MasterAlias
	log.Infof("Gathering schema for reference master %v", referenceAlias)
	referenceSchema, err := wr.GetSchema(ctx, referenceAlias, nil, excludeTables, includeViews)
	if err != nil {
		return err
	}

	// then diff with all other tablets everywhere
	er := concurrency.AllErrorRecorder{}
	wg := sync.WaitGroup{}

	// first diff the slaves in the reference shard 0
	aliases, err := topo.FindAllTabletAliasesInShard(ctx, wr.ts, keyspace, shards[0])
	if err != nil {
		return err
	}

	for _, alias := range aliases {
		if topo.TabletAliasEqual(alias, si.MasterAlias) {
			continue
		}

		wg.Add(1)
		go wr.diffSchema(ctx, referenceSchema, referenceAlias, alias, excludeTables, includeViews, &wg, &er)
	}

	// then diffs all tablets in the other shards
	for _, shard := range shards[1:] {
		si, err := wr.ts.GetShard(ctx, keyspace, shard)
		if err != nil {
			er.RecordError(err)
			continue
		}

		if topo.TabletAliasIsZero(si.MasterAlias) {
			er.RecordError(fmt.Errorf("No master in shard %v/%v", keyspace, shard))
			continue
		}

		aliases, err := topo.FindAllTabletAliasesInShard(ctx, wr.ts, keyspace, shard)
		if err != nil {
			er.RecordError(err)
			continue
		}

		for _, alias := range aliases {
			wg.Add(1)
			go wr.diffSchema(ctx, referenceSchema, referenceAlias, alias, excludeTables, includeViews, &wg, &er)
		}
	}
	wg.Wait()
	if er.HasErrors() {
		return fmt.Errorf("Schema diffs:\n%v", er.Error().Error())
	}
	return nil
}

// TestInitMasterShard is the good scenario test, where everything
// works as planned
func TestInitMasterShard(t *testing.T) {
	ctx := context.Background()
	ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
	wr := wrangler.New(logutil.NewConsoleLogger(), ts, tmclient.NewTabletManagerClient(), time.Second)
	vp := NewVtctlPipe(t, ts)
	defer vp.Close()

	// Create a master, a couple good slaves
	master := NewFakeTablet(t, wr, "cell1", 0, pb.TabletType_MASTER)
	goodSlave1 := NewFakeTablet(t, wr, "cell1", 1, pb.TabletType_REPLICA)
	goodSlave2 := NewFakeTablet(t, wr, "cell2", 2, pb.TabletType_REPLICA)

	// Master: set a plausible ReplicationPosition to return,
	// and expect to add entry in _vt.reparent_journal
	master.FakeMysqlDaemon.CurrentMasterPosition = myproto.ReplicationPosition{
		GTIDSet: myproto.MariadbGTID{
			Domain:   5,
			Server:   456,
			Sequence: 890,
		},
	}
	master.FakeMysqlDaemon.ReadOnly = true
	master.FakeMysqlDaemon.ResetReplicationResult = []string{"reset rep 1"}
	master.FakeMysqlDaemon.StartReplicationCommandsResult = []string{"new master shouldn't use this"}
	master.FakeMysqlDaemon.ExpectedExecuteSuperQueryList = []string{
		"reset rep 1",
		"CREATE DATABASE IF NOT EXISTS _vt",
		"SUBCREATE TABLE IF NOT EXISTS _vt.reparent_journal",
		"CREATE DATABASE IF NOT EXISTS _vt",
		"SUBCREATE TABLE IF NOT EXISTS _vt.reparent_journal",
		"SUBINSERT INTO _vt.reparent_journal (time_created_ns, action_name, master_alias, replication_position) VALUES",
	}
	master.StartActionLoop(t, wr)
	defer master.StopActionLoop(t)

	// Slave1: expect to be reset and re-parented
	goodSlave1.FakeMysqlDaemon.ReadOnly = true
	goodSlave1.FakeMysqlDaemon.ResetReplicationResult = []string{"reset rep 1"}
	goodSlave1.FakeMysqlDaemon.StartReplicationCommandsStatus = &myproto.ReplicationStatus{
		Position:           master.FakeMysqlDaemon.CurrentMasterPosition,
		MasterHost:         master.Tablet.Hostname,
		MasterPort:         int(master.Tablet.PortMap["mysql"]),
		MasterConnectRetry: 10,
	}
	goodSlave1.FakeMysqlDaemon.StartReplicationCommandsResult = []string{"cmd1"}
	goodSlave1.FakeMysqlDaemon.ExpectedExecuteSuperQueryList = []string{
		"reset rep 1",
		"cmd1",
	}
	goodSlave1.StartActionLoop(t, wr)
	defer goodSlave1.StopActionLoop(t)

	// Slave2: expect to be re-parented
	goodSlave2.FakeMysqlDaemon.ReadOnly = true
	goodSlave2.FakeMysqlDaemon.ResetReplicationResult = []string{"reset rep 2"}
	goodSlave2.FakeMysqlDaemon.StartReplicationCommandsStatus = &myproto.ReplicationStatus{
		Position:           master.FakeMysqlDaemon.CurrentMasterPosition,
		MasterHost:         master.Tablet.Hostname,
		MasterPort:         int(master.Tablet.PortMap["mysql"]),
		MasterConnectRetry: 10,
	}
	goodSlave2.FakeMysqlDaemon.StartReplicationCommandsResult = []string{"cmd1", "cmd2"}
	goodSlave2.FakeMysqlDaemon.ExpectedExecuteSuperQueryList = []string{
		"reset rep 2",
		"cmd1",
		"cmd2",
	}
	goodSlave2.StartActionLoop(t, wr)
	defer goodSlave2.StopActionLoop(t)

	// run InitShardMaster
	if err := vp.Run([]string{"InitShardMaster", "-wait_slave_timeout", "10s", master.Tablet.Keyspace + "/" + master.Tablet.Shard, topo.TabletAliasString(master.Tablet.Alias)}); err != nil {
		t.Fatalf("InitShardMaster failed: %v", err)
	}

	// check what was run
	if master.FakeMysqlDaemon.ReadOnly {
		t.Errorf("master was not turned read-write")
	}
	si, err := ts.GetShard(ctx, master.Tablet.Keyspace, master.Tablet.Shard)
	if err != nil {
		t.Fatalf("GetShard failed: %v", err)
	}
	if !topo.TabletAliasEqual(si.MasterAlias, master.Tablet.Alias) {
		t.Errorf("unexpected shard master alias, got %v expected %v", si.MasterAlias, master.Tablet.Alias)
	}
	if err := master.FakeMysqlDaemon.CheckSuperQueryList(); err != nil {
		t.Fatalf("master.FakeMysqlDaemon.CheckSuperQueryList failed: %v", err)
	}
	if err := goodSlave1.FakeMysqlDaemon.CheckSuperQueryList(); err != nil {
		t.Fatalf("goodSlave1.FakeMysqlDaemon.CheckSuperQueryList failed: %v", err)
	}
	if err := goodSlave2.FakeMysqlDaemon.CheckSuperQueryList(); err != nil {
		t.Fatalf("goodSlave2.FakeMysqlDaemon.CheckSuperQueryList failed: %v", err)
	}
}

// InitTablet creates or updates a tablet. If no parent is specified
// in the tablet, and the tablet has a slave type, we will find the
// appropriate parent. If createShardAndKeyspace is true and the
// parent keyspace or shard don't exist, they will be created.  If
// update is true, and a tablet with the same ID exists, update it.
// If Force is true, and a tablet with the same ID already exists, it
// will be scrapped and deleted, and then recreated.
func (wr *Wrangler) InitTablet(ctx context.Context, tablet *pb.Tablet, force, createShardAndKeyspace, update bool) error {
	if err := topo.TabletComplete(tablet); err != nil {
		return err
	}

	if topo.IsInReplicationGraph(tablet.Type) {
		// get the shard, possibly creating it
		var err error
		var si *topo.ShardInfo

		if createShardAndKeyspace {
			// create the parent keyspace and shard if needed
			si, err = topotools.GetOrCreateShard(ctx, wr.ts, tablet.Keyspace, tablet.Shard)
		} else {
			si, err = wr.ts.GetShard(ctx, tablet.Keyspace, tablet.Shard)
			if err == topo.ErrNoNode {
				return fmt.Errorf("missing parent shard, use -parent option to create it, or CreateKeyspace / CreateShard")
			}
		}

		// get the shard, checks a couple things
		if err != nil {
			return fmt.Errorf("cannot get (or create) shard %v/%v: %v", tablet.Keyspace, tablet.Shard, err)
		}
		if !key.KeyRangeEqual(si.KeyRange, tablet.KeyRange) {
			return fmt.Errorf("shard %v/%v has a different KeyRange: %v != %v", tablet.Keyspace, tablet.Shard, si.KeyRange, tablet.KeyRange)
		}
		if tablet.Type == pb.TabletType_MASTER && !topo.TabletAliasIsZero(si.MasterAlias) && !topo.TabletAliasEqual(si.MasterAlias, tablet.Alias) && !force {
			return fmt.Errorf("creating this tablet would override old master %v in shard %v/%v", topo.TabletAliasString(si.MasterAlias), tablet.Keyspace, tablet.Shard)
		}

		// update the shard record if needed
		if err := wr.updateShardCellsAndMaster(ctx, si, tablet.Alias, tablet.Type, force); err != nil {
			return err
		}
	}

	err := topo.CreateTablet(ctx, wr.ts, tablet)
	if err != nil && err == topo.ErrNodeExists {
		// Try to update nicely, but if it fails fall back to force behavior.
		if update || force {
			oldTablet, err := wr.ts.GetTablet(ctx, tablet.Alias)
			if err != nil {
				wr.Logger().Warningf("failed reading tablet %v: %v", topo.TabletAliasString(tablet.Alias), err)
			} else {
				if oldTablet.Keyspace == tablet.Keyspace && oldTablet.Shard == tablet.Shard {
					*(oldTablet.Tablet) = *tablet
					if err := topo.UpdateTablet(ctx, wr.ts, oldTablet); err != nil {
						wr.Logger().Warningf("failed updating tablet %v: %v", topo.TabletAliasString(tablet.Alias), err)
						// now fall through the Scrap case
					} else {
						if !topo.IsInReplicationGraph(tablet.Type) {
							return nil
						}

						if err := topo.UpdateTabletReplicationData(ctx, wr.ts, tablet); err != nil {
							wr.Logger().Warningf("failed updating tablet replication data for %v: %v", topo.TabletAliasString(tablet.Alias), err)
							// now fall through the Scrap case
						} else {
							return nil
						}
					}
				}
			}
		}
		if force {
			if err = wr.Scrap(ctx, tablet.Alias, force, false); err != nil {
				wr.Logger().Errorf("failed scrapping tablet %v: %v", topo.TabletAliasString(tablet.Alias), err)
				return err
			}
			if err := wr.ts.DeleteTablet(ctx, tablet.Alias); err != nil {
				// we ignore this
				wr.Logger().Errorf("failed deleting tablet %v: %v", topo.TabletAliasString(tablet.Alias), err)
			}
			return topo.CreateTablet(ctx, wr.ts, tablet)
		}
	}
	return err
}

// TestInitMasterShardOneSlaveFails makes sure that if one slave fails to
// proceed, the action completes anyway
func TestInitMasterShardOneSlaveFails(t *testing.T) {
	ctx := context.Background()
	ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
	wr := wrangler.New(logutil.NewConsoleLogger(), ts, tmclient.NewTabletManagerClient(), time.Second)

	// Create a master, a couple slaves
	master := NewFakeTablet(t, wr, "cell1", 0, pb.TabletType_MASTER)
	goodSlave := NewFakeTablet(t, wr, "cell1", 1, pb.TabletType_REPLICA)
	badSlave := NewFakeTablet(t, wr, "cell2", 2, pb.TabletType_REPLICA)

	// Master: set a plausible ReplicationPosition to return,
	// and expect to add entry in _vt.reparent_journal
	master.FakeMysqlDaemon.CurrentMasterPosition = myproto.ReplicationPosition{
		GTIDSet: myproto.MariadbGTID{
			Domain:   5,
			Server:   456,
			Sequence: 890,
		},
	}
	master.FakeMysqlDaemon.ReadOnly = true
	master.FakeMysqlDaemon.ExpectedExecuteSuperQueryList = []string{
		"CREATE DATABASE IF NOT EXISTS _vt",
		"SUBCREATE TABLE IF NOT EXISTS _vt.reparent_journal",
		"CREATE DATABASE IF NOT EXISTS _vt",
		"SUBCREATE TABLE IF NOT EXISTS _vt.reparent_journal",
		"SUBINSERT INTO _vt.reparent_journal (time_created_ns, action_name, master_alias, replication_position) VALUES",
	}
	master.StartActionLoop(t, wr)
	defer master.StopActionLoop(t)

	// goodSlave: expect to be re-parented
	goodSlave.FakeMysqlDaemon.ReadOnly = true
	goodSlave.FakeMysqlDaemon.StartReplicationCommandsStatus = &myproto.ReplicationStatus{
		Position:           master.FakeMysqlDaemon.CurrentMasterPosition,
		MasterHost:         master.Tablet.Hostname,
		MasterPort:         int(master.Tablet.PortMap["mysql"]),
		MasterConnectRetry: 10,
	}
	goodSlave.FakeMysqlDaemon.StartReplicationCommandsResult = []string{"cmd1"}
	goodSlave.FakeMysqlDaemon.ExpectedExecuteSuperQueryList = goodSlave.FakeMysqlDaemon.StartReplicationCommandsResult
	goodSlave.StartActionLoop(t, wr)
	defer goodSlave.StopActionLoop(t)

	// badSlave: insert an error by failing the ReplicationStatus input
	// on purpose
	badSlave.FakeMysqlDaemon.ReadOnly = true
	badSlave.FakeMysqlDaemon.StartReplicationCommandsStatus = &myproto.ReplicationStatus{
		Position:           master.FakeMysqlDaemon.CurrentMasterPosition,
		MasterHost:         "",
		MasterPort:         0,
		MasterConnectRetry: 10,
	}
	badSlave.StartActionLoop(t, wr)
	defer badSlave.StopActionLoop(t)

	// also change the master alias in the Shard object, to make sure it
	// is set back.
	si, err := ts.GetShard(ctx, master.Tablet.Keyspace, master.Tablet.Shard)
	if err != nil {
		t.Fatalf("GetShard failed: %v", err)
	}
	si.MasterAlias.Uid++
	if err := topo.UpdateShard(ctx, ts, si); err != nil {
		t.Fatalf("UpdateShard failed: %v", err)
	}

	// run InitShardMaster without force, it fails because master is
	// changing.
	if err := wr.InitShardMaster(ctx, master.Tablet.Keyspace, master.Tablet.Shard, master.Tablet.Alias, false /*force*/, 10*time.Second); err == nil || !strings.Contains(err.Error(), "is not the shard master") {
		t.Errorf("InitShardMaster with mismatched new master returned wrong error: %v", err)
	}

	// run InitShardMaster
	if err := wr.InitShardMaster(ctx, master.Tablet.Keyspace, master.Tablet.Shard, master.Tablet.Alias, true /*force*/, 10*time.Second); err == nil || !strings.Contains(err.Error(), "wrong status for StartReplicationCommands") {
		t.Errorf("InitShardMaster with one failed slave returned wrong error: %v", err)
	}

	// check what was run: master should still be good
	if master.FakeMysqlDaemon.ReadOnly {
		t.Errorf("master was not turned read-write")
	}
	si, err = ts.GetShard(ctx, master.Tablet.Keyspace, master.Tablet.Shard)
	if err != nil {
		t.Fatalf("GetShard failed: %v", err)
	}
	if !topo.TabletAliasEqual(si.MasterAlias, master.Tablet.Alias) {
		t.Errorf("unexpected shard master alias, got %v expected %v", si.MasterAlias, master.Tablet.Alias)
	}
}

func TestTabletExternallyReparented(t *testing.T) {
	tabletmanager.SetReparentFlags(time.Minute /* finalizeTimeout */)

	ctx := context.Background()
	ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
	wr := wrangler.New(logutil.NewConsoleLogger(), ts, tmclient.NewTabletManagerClient(), time.Second)
	vp := NewVtctlPipe(t, ts)
	defer vp.Close()

	// Create an old master, a new master, two good slaves, one bad slave
	oldMaster := NewFakeTablet(t, wr, "cell1", 0, pb.TabletType_MASTER)
	newMaster := NewFakeTablet(t, wr, "cell1", 1, pb.TabletType_REPLICA)
	goodSlave1 := NewFakeTablet(t, wr, "cell1", 2, pb.TabletType_REPLICA)
	goodSlave2 := NewFakeTablet(t, wr, "cell2", 3, pb.TabletType_REPLICA)
	badSlave := NewFakeTablet(t, wr, "cell1", 4, pb.TabletType_REPLICA)

	// Add a new Cell to the Shard, that doesn't map to any read topo cell,
	// to simulate a data center being unreachable.
	si, err := ts.GetShard(ctx, "test_keyspace", "0")
	if err != nil {
		t.Fatalf("GetShard failed: %v", err)
	}
	si.Cells = append(si.Cells, "cell666")
	if err := topo.UpdateShard(ctx, ts, si); err != nil {
		t.Fatalf("UpdateShard failed: %v", err)
	}

	// Slightly unrelated test: make sure we can find the tablets
	// even with a datacenter being down.
	tabletMap, err := topo.GetTabletMapForShardByCell(ctx, ts, "test_keyspace", "0", []string{"cell1"})
	if err != nil {
		t.Fatalf("GetTabletMapForShardByCell should have worked but got: %v", err)
	}
	master, err := topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.Ip, "vt", oldMaster.Tablet.PortMap["vt"])
	if err != nil || !topo.TabletAliasEqual(&master, oldMaster.Tablet.Alias) {
		t.Fatalf("FindTabletByIPAddrAndPort(master) failed: %v %v", err, master)
	}
	slave1, err := topotools.FindTabletByIPAddrAndPort(tabletMap, goodSlave1.Tablet.Ip, "vt", goodSlave1.Tablet.PortMap["vt"])
	if err != nil || !topo.TabletAliasEqual(&slave1, goodSlave1.Tablet.Alias) {
		t.Fatalf("FindTabletByIPAddrAndPort(slave1) failed: %v %v", err, master)
	}
	slave2, err := topotools.FindTabletByIPAddrAndPort(tabletMap, goodSlave2.Tablet.Ip, "vt", goodSlave2.Tablet.PortMap["vt"])
	if err != topo.ErrNoNode {
		t.Fatalf("FindTabletByIPAddrAndPort(slave2) worked: %v %v", err, slave2)
	}

	// Make sure the master is not exported in other cells
	tabletMap, err = topo.GetTabletMapForShardByCell(ctx, ts, "test_keyspace", "0", []string{"cell2"})
	master, err = topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.Ip, "vt", oldMaster.Tablet.PortMap["vt"])
	if err != topo.ErrNoNode {
		t.Fatalf("FindTabletByIPAddrAndPort(master) worked in cell2: %v %v", err, master)
	}

	tabletMap, err = topo.GetTabletMapForShard(ctx, ts, "test_keyspace", "0")
	if err != topo.ErrPartialResult {
		t.Fatalf("GetTabletMapForShard should have returned ErrPartialResult but got: %v", err)
	}
	master, err = topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.Ip, "vt", oldMaster.Tablet.PortMap["vt"])
	if err != nil || !topo.TabletAliasEqual(&master, oldMaster.Tablet.Alias) {
		t.Fatalf("FindTabletByIPAddrAndPort(master) failed: %v %v", err, master)
	}

	// On the elected master, we will respond to
	// TabletActionSlaveWasPromoted
	newMaster.StartActionLoop(t, wr)
	defer newMaster.StopActionLoop(t)

	// On the old master, we will only respond to
	// TabletActionSlaveWasRestarted.
	oldMaster.StartActionLoop(t, wr)
	defer oldMaster.StopActionLoop(t)

	// On the good slaves, we will respond to
	// TabletActionSlaveWasRestarted.
	goodSlave1.StartActionLoop(t, wr)
	defer goodSlave1.StopActionLoop(t)

	goodSlave2.StartActionLoop(t, wr)
	defer goodSlave2.StopActionLoop(t)

	// On the bad slave, we will respond to
	// TabletActionSlaveWasRestarted with bad data.
	badSlave.StartActionLoop(t, wr)
	defer badSlave.StopActionLoop(t)

	// First test: reparent to the same master, make sure it works
	// as expected.
	tmc := tmclient.NewTabletManagerClient()
	ti, err := ts.GetTablet(ctx, oldMaster.Tablet.Alias)
	if err != nil {
		t.Fatalf("GetTablet failed: %v", err)
	}
	if err := vp.Run([]string{"TabletExternallyReparented", topo.TabletAliasString(oldMaster.Tablet.Alias)}); err != nil {
		t.Fatalf("TabletExternallyReparented(same master) should have worked: %v", err)
	}

	// Second test: reparent to a replica, and pretend the old
	// master is still good to go.

	// This tests a bad case; the new designated master is a slave,
//.........这里部分代码省略.........

func TestShardReplicationStatuses(t *testing.T) {
	ctx := context.Background()
	ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
	wr := wrangler.New(logutil.NewConsoleLogger(), ts, tmclient.NewTabletManagerClient(), time.Second)

	// create shard and tablets
	if err := topo.CreateShard(ctx, ts, "test_keyspace", "0"); err != nil {
		t.Fatalf("CreateShard failed: %v", err)
	}
	master := NewFakeTablet(t, wr, "cell1", 1, pb.TabletType_MASTER)
	slave := NewFakeTablet(t, wr, "cell1", 2, pb.TabletType_REPLICA)

	// mark the master inside the shard
	si, err := ts.GetShard(ctx, "test_keyspace", "0")
	if err != nil {
		t.Fatalf("GetShard failed: %v", err)
	}
	si.MasterAlias = master.Tablet.Alias
	if err := topo.UpdateShard(ctx, ts, si); err != nil {
		t.Fatalf("UpdateShard failed: %v", err)
	}

	// master action loop (to initialize host and port)
	master.FakeMysqlDaemon.CurrentMasterPosition = myproto.ReplicationPosition{
		GTIDSet: myproto.MariadbGTID{
			Domain:   5,
			Server:   456,
			Sequence: 892,
		},
	}
	master.StartActionLoop(t, wr)
	defer master.StopActionLoop(t)

	// slave loop
	slave.FakeMysqlDaemon.CurrentMasterPosition = myproto.ReplicationPosition{
		GTIDSet: myproto.MariadbGTID{
			Domain:   5,
			Server:   456,
			Sequence: 890,
		},
	}
	slave.FakeMysqlDaemon.CurrentMasterHost = master.Tablet.Hostname
	slave.FakeMysqlDaemon.CurrentMasterPort = int(master.Tablet.PortMap["mysql"])
	slave.StartActionLoop(t, wr)
	defer slave.StopActionLoop(t)

	// run ShardReplicationStatuses
	ti, rs, err := wr.ShardReplicationStatuses(ctx, "test_keyspace", "0")
	if err != nil {
		t.Fatalf("ShardReplicationStatuses failed: %v", err)
	}

	// check result (make master first in the array)
	if len(ti) != 2 || len(rs) != 2 {
		t.Fatalf("ShardReplicationStatuses returned wrong results: %v %v", ti, rs)
	}
	if topo.TabletAliasEqual(ti[0].Alias, slave.Tablet.Alias) {
		ti[0], ti[1] = ti[1], ti[0]
		rs[0], rs[1] = rs[1], rs[0]
	}
	if !topo.TabletAliasEqual(ti[0].Alias, master.Tablet.Alias) ||
		!topo.TabletAliasEqual(ti[1].Alias, slave.Tablet.Alias) ||
		rs[0].MasterHost != "" ||
		rs[1].MasterHost != master.Tablet.Hostname {
		t.Fatalf("ShardReplicationStatuses returend wrong results: %v %v", ti, rs)
	}
}

func (wr *Wrangler) emergencyReparentShardLocked(ctx context.Context, ev *events.Reparent, keyspace, shard string, masterElectTabletAlias *pb.TabletAlias, waitSlaveTimeout time.Duration) error {
	shardInfo, err := wr.ts.GetShard(ctx, keyspace, shard)
	if err != nil {
		return err
	}
	ev.ShardInfo = *shardInfo

	event.DispatchUpdate(ev, "reading all tablets")
	tabletMap, err := topo.GetTabletMapForShard(ctx, wr.ts, keyspace, shard)
	if err != nil {
		return err
	}

	// Check corner cases we're going to depend on
	masterElectTabletInfo, ok := tabletMap[*masterElectTabletAlias]
	if !ok {
		return fmt.Errorf("master-elect tablet %v is not in the shard", masterElectTabletAlias)
	}
	ev.NewMaster = *masterElectTabletInfo.Tablet
	if topo.TabletAliasEqual(shardInfo.MasterAlias, masterElectTabletAlias) {
		return fmt.Errorf("master-elect tablet %v is already the master", masterElectTabletAlias)
	}

	// Deal with the old master: try to remote-scrap it, if it's
	// truely dead we force-scrap it. Remove it from our map in any case.
	if !topo.TabletAliasIsZero(shardInfo.MasterAlias) {
		scrapOldMaster := true
		oldMasterTabletInfo, ok := tabletMap[*shardInfo.MasterAlias]
		if ok {
			delete(tabletMap, *shardInfo.MasterAlias)
		} else {
			oldMasterTabletInfo, err = wr.ts.GetTablet(ctx, shardInfo.MasterAlias)
			if err != nil {
				wr.logger.Warningf("cannot read old master tablet %v, won't touch it: %v", shardInfo.MasterAlias, err)
				scrapOldMaster = false
			}
		}

		if scrapOldMaster {
			ev.OldMaster = *oldMasterTabletInfo.Tablet
			wr.logger.Infof("scrapping old master %v", shardInfo.MasterAlias)

			ctx, cancel := context.WithTimeout(ctx, waitSlaveTimeout)
			defer cancel()

			if err := wr.tmc.Scrap(ctx, oldMasterTabletInfo); err != nil {
				wr.logger.Warningf("remote scrapping failed master failed, will force the scrap: %v", err)

				if err := topotools.Scrap(ctx, wr.ts, shardInfo.MasterAlias, true); err != nil {
					wr.logger.Warningf("old master topo scrapping failed, continuing anyway: %v", err)
				}
			}
		}
	}

	// Stop replication on all slaves, get their current
	// replication position
	event.DispatchUpdate(ev, "stop replication on all slaves")
	wg := sync.WaitGroup{}
	mu := sync.Mutex{}
	statusMap := make(map[pb.TabletAlias]myproto.ReplicationStatus)
	for alias, tabletInfo := range tabletMap {
		wg.Add(1)
		go func(alias pb.TabletAlias, tabletInfo *topo.TabletInfo) {
			defer wg.Done()
			wr.logger.Infof("getting replication position from %v", alias)
			ctx, cancel := context.WithTimeout(ctx, waitSlaveTimeout)
			defer cancel()
			rp, err := wr.TabletManagerClient().StopReplicationAndGetStatus(ctx, tabletInfo)
			if err != nil {
				wr.logger.Warningf("failed to get replication status from %v, ignoring tablet: %v", alias, err)
				return
			}
			mu.Lock()
			statusMap[alias] = rp
			mu.Unlock()
		}(alias, tabletInfo)
	}
	wg.Wait()

	// Verify masterElect is alive and has the most advanced position
	masterElectStatus, ok := statusMap[*masterElectTabletAlias]
	if !ok {
		return fmt.Errorf("couldn't get master elect %v replication position", masterElectTabletAlias)
	}
	for alias, status := range statusMap {
		if topo.TabletAliasEqual(&alias, masterElectTabletAlias) {
			continue
		}
		if !masterElectStatus.Position.AtLeast(status.Position) {
			return fmt.Errorf("tablet %v is more advanced than master elect tablet %v: %v > %v", alias, masterElectTabletAlias, status.Position, masterElectStatus)
		}
	}

	// Promote the masterElect
	wr.logger.Infof("promote slave %v", masterElectTabletAlias)
	event.DispatchUpdate(ev, "promoting slave")
	rp, err := wr.tmc.PromoteSlave(ctx, masterElectTabletInfo)
	if err != nil {
		return fmt.Errorf("master-elect tablet %v failed to be upgraded to master: %v", masterElectTabletAlias, err)
//.........这里部分代码省略.........

func (wr *Wrangler) plannedReparentShardLocked(ctx context.Context, ev *events.Reparent, keyspace, shard string, masterElectTabletAlias *pb.TabletAlias, waitSlaveTimeout time.Duration) error {
	shardInfo, err := wr.ts.GetShard(ctx, keyspace, shard)
	if err != nil {
		return err
	}
	ev.ShardInfo = *shardInfo

	event.DispatchUpdate(ev, "reading tablet map")
	tabletMap, err := topo.GetTabletMapForShard(ctx, wr.ts, keyspace, shard)
	if err != nil {
		return err
	}

	// Check corner cases we're going to depend on
	masterElectTabletInfo, ok := tabletMap[*masterElectTabletAlias]
	if !ok {
		return fmt.Errorf("master-elect tablet %v is not in the shard", masterElectTabletAlias)
	}
	ev.NewMaster = *masterElectTabletInfo.Tablet
	if topo.TabletAliasEqual(shardInfo.MasterAlias, masterElectTabletAlias) {
		return fmt.Errorf("master-elect tablet %v is already the master", masterElectTabletAlias)
	}
	oldMasterTabletInfo, ok := tabletMap[*shardInfo.MasterAlias]
	if !ok {
		return fmt.Errorf("old master tablet %v is not in the shard", shardInfo.MasterAlias)
	}
	ev.OldMaster = *oldMasterTabletInfo.Tablet

	// Demote the current master, get its replication position
	wr.logger.Infof("demote current master %v", shardInfo.MasterAlias)
	event.DispatchUpdate(ev, "demoting old master")
	rp, err := wr.tmc.DemoteMaster(ctx, oldMasterTabletInfo)
	if err != nil {
		return fmt.Errorf("old master tablet %v DemoteMaster failed: %v", shardInfo.MasterAlias, err)
	}

	// Wait on the master-elect tablet until it reaches that position,
	// then promote it
	wr.logger.Infof("promote slave %v", masterElectTabletAlias)
	event.DispatchUpdate(ev, "promoting slave")
	rp, err = wr.tmc.PromoteSlaveWhenCaughtUp(ctx, masterElectTabletInfo, rp)
	if err != nil {
		return fmt.Errorf("master-elect tablet %v failed to catch up with replication or be upgraded to master: %v", masterElectTabletAlias, err)
	}

	// Go through all the tablets:
	// - new master: populate the reparent journal
	// - everybody else: reparent to new master, wait for row
	event.DispatchUpdate(ev, "reparenting all tablets")
	now := time.Now().UnixNano()
	wgMaster := sync.WaitGroup{}
	wgSlaves := sync.WaitGroup{}
	rec := concurrency.AllErrorRecorder{}
	var masterErr error
	for alias, tabletInfo := range tabletMap {
		if topo.TabletAliasEqual(&alias, masterElectTabletAlias) {
			wgMaster.Add(1)
			go func(alias pb.TabletAlias, tabletInfo *topo.TabletInfo) {
				defer wgMaster.Done()
				wr.logger.Infof("populating reparent journal on new master %v", alias)
				masterErr = wr.TabletManagerClient().PopulateReparentJournal(ctx, tabletInfo, now, plannedReparentShardOperation, &alias, rp)
			}(alias, tabletInfo)
		} else {
			wgSlaves.Add(1)
			go func(alias pb.TabletAlias, tabletInfo *topo.TabletInfo) {
				defer wgSlaves.Done()
				wr.logger.Infof("setting new master on slave %v", alias)
				// also restart replication on old master
				forceStartSlave := topo.TabletAliasEqual(&alias, oldMasterTabletInfo.Alias)
				if err := wr.TabletManagerClient().SetMaster(ctx, tabletInfo, masterElectTabletAlias, now, forceStartSlave); err != nil {
					rec.RecordError(fmt.Errorf("Tablet %v SetMaster failed: %v", alias, err))
					return
				}
			}(alias, tabletInfo)
		}
	}

	// After the master is done, we can update the shard record
	// (note with semi-sync, it also means at least one slave is done)
	wgMaster.Wait()
	if masterErr != nil {
		wgSlaves.Wait()
		return fmt.Errorf("failed to PopulateReparentJournal on master: %v", masterErr)
	}
	wr.logger.Infof("updating shard record with new master %v", masterElectTabletAlias)
	shardInfo.MasterAlias = masterElectTabletAlias
	if err := topo.UpdateShard(ctx, wr.ts, shardInfo); err != nil {
		wgSlaves.Wait()
		return fmt.Errorf("failed to update shard master record: %v", err)
	}

	// Wait for the slaves to complete. If some of them fail, we
	// will rebuild the shard serving graph anyway
	wgSlaves.Wait()
	if err := rec.Error(); err != nil {
		wr.Logger().Errorf("Some slaves failed to reparent: %v", err)
		return err
	}

	// Then we rebuild the entire serving graph for the shard,
//.........这里部分代码省略.........

func (wr *Wrangler) initShardMasterLocked(ctx context.Context, ev *events.Reparent, keyspace, shard string, masterElectTabletAlias *pb.TabletAlias, force bool, waitSlaveTimeout time.Duration) error {
	shardInfo, err := wr.ts.GetShard(ctx, keyspace, shard)
	if err != nil {
		return err
	}
	ev.ShardInfo = *shardInfo

	event.DispatchUpdate(ev, "reading tablet map")
	tabletMap, err := topo.GetTabletMapForShard(ctx, wr.ts, keyspace, shard)
	if err != nil {
		return err
	}

	// Check the master elect is in tabletMap
	masterElectTabletInfo, ok := tabletMap[*masterElectTabletAlias]
	if !ok {
		return fmt.Errorf("master-elect tablet %v is not in the shard", masterElectTabletAlias)
	}
	ev.NewMaster = *masterElectTabletInfo.Tablet

	// Check the master is the only master is the shard, or -force was used.
	_, masterTabletMap := topotools.SortedTabletMap(tabletMap)
	if !topo.TabletAliasEqual(shardInfo.MasterAlias, masterElectTabletAlias) {
		if !force {
			return fmt.Errorf("master-elect tablet %v is not the shard master, use -force to proceed anyway", masterElectTabletAlias)
		}
		wr.logger.Warningf("master-elect tablet %v is not the shard master, proceeding anyway as -force was used", masterElectTabletAlias)
	}
	if _, ok := masterTabletMap[*masterElectTabletAlias]; !ok {
		if !force {
			return fmt.Errorf("master-elect tablet %v is not a master in the shard, use -force to proceed anyway", masterElectTabletAlias)
		}
		wr.logger.Warningf("master-elect tablet %v is not a master in the shard, proceeding anyway as -force was used", masterElectTabletAlias)
	}
	haveOtherMaster := false
	for alias, ti := range masterTabletMap {
		if !topo.TabletAliasEqual(&alias, masterElectTabletAlias) && ti.Type != pb.TabletType_SCRAP {
			haveOtherMaster = true
		}
	}
	if haveOtherMaster {
		if !force {
			return fmt.Errorf("master-elect tablet %v is not the only master in the shard, use -force to proceed anyway", masterElectTabletAlias)
		}
		wr.logger.Warningf("master-elect tablet %v is not the only master in the shard, proceeding anyway as -force was used", masterElectTabletAlias)
	}

	// First phase: reset replication on all tablets. If anyone fails,
	// we stop. It is probably because it is unreachable, and may leave
	// an unstable database process in the mix, with a database daemon
	// at a wrong replication spot.
	event.DispatchUpdate(ev, "resetting replication on all tablets")
	wg := sync.WaitGroup{}
	rec := concurrency.AllErrorRecorder{}
	for alias, tabletInfo := range tabletMap {
		wg.Add(1)
		go func(alias pb.TabletAlias, tabletInfo *topo.TabletInfo) {
			defer wg.Done()
			wr.logger.Infof("resetting replication on tablet %v", alias)
			if err := wr.TabletManagerClient().ResetReplication(ctx, tabletInfo); err != nil {
				rec.RecordError(fmt.Errorf("Tablet %v ResetReplication failed (either fix it, or Scrap it): %v", alias, err))
			}
		}(alias, tabletInfo)
	}
	wg.Wait()
	if err := rec.Error(); err != nil {
		return err
	}

	// Tell the new master to break its slaves, return its replication
	// position
	wr.logger.Infof("initializing master on %v", masterElectTabletAlias)
	event.DispatchUpdate(ev, "initializing master")
	rp, err := wr.TabletManagerClient().InitMaster(ctx, masterElectTabletInfo)
	if err != nil {
		return err
	}

	// Now tell the new master to insert the reparent_journal row,
	// and tell everybody else to become a slave of the new master,
	// and wait for the row in the reparent_journal table.
	// We start all these in parallel, to handle the semi-sync
	// case: for the master to be able to commit its row in the
	// reparent_journal table, it needs connected slaves.
	event.DispatchUpdate(ev, "reparenting all tablets")
	now := time.Now().UnixNano()
	wgMaster := sync.WaitGroup{}
	wgSlaves := sync.WaitGroup{}
	var masterErr error
	for alias, tabletInfo := range tabletMap {
		if topo.TabletAliasEqual(&alias, masterElectTabletAlias) {
			wgMaster.Add(1)
			go func(alias pb.TabletAlias, tabletInfo *topo.TabletInfo) {
				defer wgMaster.Done()
				wr.logger.Infof("populating reparent journal on new master %v", alias)
				masterErr = wr.TabletManagerClient().PopulateReparentJournal(ctx, tabletInfo, now, initShardMasterOperation, &alias, rp)
			}(alias, tabletInfo)
		} else {
			wgSlaves.Add(1)
			go func(alias pb.TabletAlias, tabletInfo *topo.TabletInfo) {
//.........这里部分代码省略.........

// TabletExternallyReparented updates all topo records so the current
// tablet is the new master for this shard.
// Should be called under RPCWrapLock.
func (agent *ActionAgent) TabletExternallyReparented(ctx context.Context, externalID string) error {
	startTime := time.Now()

	// If there is a finalize step running, wait for it to finish or time out
	// before checking the global shard record again.
	if agent.finalizeReparentCtx != nil {
		select {
		case <-agent.finalizeReparentCtx.Done():
			agent.finalizeReparentCtx = nil
		case <-ctx.Done():
			return ctx.Err()
		}
	}

	tablet := agent.Tablet()

	// Check the global shard record.
	si, err := topo.GetShard(ctx, agent.TopoServer, tablet.Keyspace, tablet.Shard)
	if err != nil {
		log.Warningf("fastTabletExternallyReparented: failed to read global shard record for %v/%v: %v", tablet.Keyspace, tablet.Shard, err)
		return err
	}
	if topo.TabletAliasEqual(si.MasterAlias, tablet.Alias) {
		// We may get called on the current master even when nothing has changed.
		// If the global shard record is already updated, it means we successfully
		// finished a previous reparent to this tablet.
		return nil
	}

	// Create a reusable Reparent event with available info.
	ev := &events.Reparent{
		ShardInfo: *si,
		NewMaster: *tablet.Tablet,
		OldMaster: pb.Tablet{
			Alias: si.MasterAlias,
			Type:  pb.TabletType_MASTER,
		},
		ExternalID: externalID,
	}
	defer func() {
		if err != nil {
			event.DispatchUpdate(ev, "failed: "+err.Error())
		}
	}()
	event.DispatchUpdate(ev, "starting external from tablet (fast)")

	// Execute state change to master by force-updating only the local copy of the
	// tablet record. The actual record in topo will be updated later.
	log.Infof("fastTabletExternallyReparented: executing change callback for state change to MASTER")
	oldTablet := *tablet.Tablet
	newTablet := oldTablet
	newTablet.Type = pb.TabletType_MASTER
	newTablet.HealthMap = nil
	agent.setTablet(topo.NewTabletInfo(&newTablet, -1))
	if err := agent.updateState(ctx, &oldTablet, "fastTabletExternallyReparented"); err != nil {
		return fmt.Errorf("fastTabletExternallyReparented: failed to change tablet state to MASTER: %v", err)
	}

	agent.mutex.Lock()
	agent._tabletExternallyReparentedTime = time.Now()
	agent.mutex.Unlock()

	// Directly write the new master endpoint in the serving graph.
	// We will do a true rebuild in the background soon, but in the meantime,
	// this will be enough for clients to re-resolve the new master.
	event.DispatchUpdate(ev, "writing new master endpoint")
	log.Infof("fastTabletExternallyReparented: writing new master endpoint to serving graph")
	ep, err := topo.TabletEndPoint(tablet.Tablet)
	if err != nil {
		return fmt.Errorf("fastTabletExternallyReparented: failed to generate EndPoint for tablet %v: %v", tablet.Alias, err)
	}
	err = topo.UpdateEndPoints(ctx, agent.TopoServer, tablet.Alias.Cell,
		si.Keyspace(), si.ShardName(), pb.TabletType_MASTER,
		&pb.EndPoints{Entries: []*pb.EndPoint{ep}}, -1)
	if err != nil {
		return fmt.Errorf("fastTabletExternallyReparented: failed to update master endpoint: %v", err)
	}
	externalReparentStats.Record("NewMasterVisible", startTime)

	// Start the finalize stage with a background context, but connect the trace.
	bgCtx, cancel := context.WithTimeout(agent.batchCtx, *finalizeReparentTimeout)
	bgCtx = trace.CopySpan(bgCtx, ctx)
	agent.finalizeReparentCtx = bgCtx
	go func() {
		err := agent.finalizeTabletExternallyReparented(bgCtx, si, ev)
		cancel()

		if err != nil {
			log.Warningf("finalizeTabletExternallyReparented error: %v", err)
			event.DispatchUpdate(ev, "failed: "+err.Error())
			return
		}
		externalReparentStats.Record("FullRebuild", startTime)
	}()

	return nil
}