Golang Reparent.NewMaster方法代码示例

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 := wr.ts.GetTabletMapForShard(ctx, 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", topoproto.TabletAliasString(masterElectTabletAlias))
	}
	ev.NewMaster = *masterElectTabletInfo.Tablet
	if topoproto.TabletAliasEqual(shardInfo.MasterAlias, masterElectTabletAlias) {
		return fmt.Errorf("master-elect tablet %v is already the master", topoproto.TabletAliasString(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 shardInfo.HasMaster() {
		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", topoproto.TabletAliasString(shardInfo.MasterAlias), err)
				scrapOldMaster = false
			}
		}

		if scrapOldMaster {
			ev.OldMaster = *oldMasterTabletInfo.Tablet
			wr.logger.Infof("scrapping old master %v", topoproto.TabletAliasString(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", topoproto.TabletAliasString(&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", topoproto.TabletAliasString(&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", topoproto.TabletAliasString(masterElectTabletAlias))
	}
	for alias, status := range statusMap {
		if topoproto.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", topoproto.TabletAliasString(&alias), topoproto.TabletAliasString(masterElectTabletAlias), status.Position, masterElectStatus)
		}
	}

	// Promote the masterElect
	wr.logger.Infof("promote slave %v", topoproto.TabletAliasString(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", topoproto.TabletAliasString(masterElectTabletAlias), err)
//.........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 := wr.ts.GetTabletMapForShard(ctx, 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", topoproto.TabletAliasString(masterElectTabletAlias))
	}
	ev.NewMaster = *masterElectTabletInfo.Tablet

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

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 := wr.ts.GetTabletMapForShard(ctx, 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", topoproto.TabletAliasString(masterElectTabletAlias))
	}
	ev.NewMaster = *masterElectTabletInfo.Tablet
	if topoproto.TabletAliasEqual(shardInfo.MasterAlias, masterElectTabletAlias) {
		return fmt.Errorf("master-elect tablet %v is already the master", topoproto.TabletAliasString(masterElectTabletAlias))
	}
	oldMasterTabletInfo, ok := tabletMap[*shardInfo.MasterAlias]
	if !ok {
		return fmt.Errorf("old master tablet %v is not in the shard", topoproto.TabletAliasString(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", topoproto.TabletAliasString(shardInfo.MasterAlias), err)
	}

	// Wait on the master-elect tablet until it reaches that position,
	// then promote it
	wr.logger.Infof("promote slave %v", topoproto.TabletAliasString(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", topoproto.TabletAliasString(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 topoproto.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", topoproto.TabletAliasString(&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", topoproto.TabletAliasString(&alias))
				// also restart replication on old master
				forceStartSlave := topoproto.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", topoproto.TabletAliasString(&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 := wr.ts.UpdateShard(ctx, 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 *topodatapb.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 := wr.ts.GetTabletMapForShard(ctx, 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", topoproto.TabletAliasString(masterElectTabletAlias))
	}
	ev.NewMaster = *masterElectTabletInfo.Tablet

	// Check the master is the only master is the shard, or -force was used.
	_, masterTabletMap := topotools.SortedTabletMap(tabletMap)
	if !topoproto.TabletAliasEqual(shardInfo.MasterAlias, masterElectTabletAlias) {
		if !force {
			return fmt.Errorf("master-elect tablet %v is not the shard master, use -force to proceed anyway", topoproto.TabletAliasString(masterElectTabletAlias))
		}
		wr.logger.Warningf("master-elect tablet %v is not the shard master, proceeding anyway as -force was used", topoproto.TabletAliasString(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", topoproto.TabletAliasString(masterElectTabletAlias))
		}
		wr.logger.Warningf("master-elect tablet %v is not a master in the shard, proceeding anyway as -force was used", topoproto.TabletAliasString(masterElectTabletAlias))
	}
	haveOtherMaster := false
	for alias := range masterTabletMap {
		if !topoproto.TabletAliasEqual(&alias, masterElectTabletAlias) {
			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", topoproto.TabletAliasString(masterElectTabletAlias))
		}
		wr.logger.Warningf("master-elect tablet %v is not the only master in the shard, proceeding anyway as -force was used", topoproto.TabletAliasString(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 topodatapb.TabletAlias, tabletInfo *topo.TabletInfo) {
			defer wg.Done()
			wr.logger.Infof("resetting replication on tablet %v", topoproto.TabletAliasString(&alias))
			if err := wr.tmc.ResetReplication(ctx, tabletInfo.Tablet); err != nil {
				rec.RecordError(fmt.Errorf("Tablet %v ResetReplication failed (either fix it, or Scrap it): %v", topoproto.TabletAliasString(&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", topoproto.TabletAliasString(masterElectTabletAlias))
	event.DispatchUpdate(ev, "initializing master")
	rp, err := wr.tmc.InitMaster(ctx, masterElectTabletInfo.Tablet)
	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 topoproto.TabletAliasEqual(&alias, masterElectTabletAlias) {
			wgMaster.Add(1)
			go func(alias topodatapb.TabletAlias, tabletInfo *topo.TabletInfo) {
				defer wgMaster.Done()
				wr.logger.Infof("populating reparent journal on new master %v", topoproto.TabletAliasString(&alias))
				masterErr = wr.tmc.PopulateReparentJournal(ctx, tabletInfo.Tablet, now, initShardMasterOperation, &alias, rp)
			}(alias, tabletInfo)
		} else {
			wgSlaves.Add(1)
			go func(alias topodatapb.TabletAlias, tabletInfo *topo.TabletInfo) {
//.........这里部分代码省略.........