Golang Logger.Infof方法代码示例

// FixShardReplication will fix the first problem it encounters within
// a ShardReplication object
func FixShardReplication(ctx context.Context, ts Server, logger logutil.Logger, cell, keyspace, shard string) error {
	sri, err := ts.GetShardReplication(ctx, cell, keyspace, shard)
	if err != nil {
		return err
	}

	for _, node := range sri.Nodes {
		ti, err := ts.GetTablet(ctx, node.TabletAlias)
		if err == ErrNoNode {
			logger.Warningf("Tablet %v is in the replication graph, but does not exist, removing it", node.TabletAlias)
			return RemoveShardReplicationRecord(ctx, ts, cell, keyspace, shard, node.TabletAlias)
		}
		if err != nil {
			// unknown error, we probably don't want to continue
			return err
		}

		if ti.Type == pb.TabletType_SCRAP {
			logger.Warningf("Tablet %v is in the replication graph, but is scrapped, removing it", node.TabletAlias)
			return RemoveShardReplicationRecord(ctx, ts, cell, keyspace, shard, node.TabletAlias)
		}

		logger.Infof("Keeping tablet %v in the replication graph", node.TabletAlias)
	}

	logger.Infof("All entries in replication graph are valid")
	return nil
}

// NewRestartableResultReader creates a new RestartableResultReader for
// the provided tablet and chunk.
// It will automatically create the necessary query to read all rows within
// the chunk.
// NOTE: We assume that the Columns field in "td" was ordered by a preceding
// call to reorderColumnsPrimaryKeyFirst().
func NewRestartableResultReader(ctx context.Context, logger logutil.Logger, ts topo.Server, tabletAlias *topodatapb.TabletAlias, td *tabletmanagerdatapb.TableDefinition, chunk chunk) (*RestartableResultReader, error) {
	shortCtx, cancel := context.WithTimeout(ctx, *remoteActionsTimeout)
	tablet, err := ts.GetTablet(shortCtx, tabletAlias)
	cancel()
	if err != nil {
		return nil, fmt.Errorf("tablet=%v table=%v chunk=%v: Failed to resolve tablet alias: %v", topoproto.TabletAliasString(tabletAlias), td.Name, chunk, err)
	}

	conn, err := tabletconn.GetDialer()(tablet.Tablet, *remoteActionsTimeout)
	if err != nil {
		return nil, fmt.Errorf("tablet=%v table=%v chunk=%v: Failed to get dialer for tablet: %v", topoproto.TabletAliasString(tabletAlias), td.Name, chunk, err)
	}

	r := &RestartableResultReader{
		ctx:    ctx,
		logger: logger,
		tablet: tablet.Tablet,
		td:     td,
		chunk:  chunk,
		conn:   conn,
	}

	if err := r.startStream(); err != nil {
		return nil, err
	}
	logger.Infof("tablet=%v table=%v chunk=%v: Starting to stream rows using query '%v'.", topoproto.TabletAliasString(tabletAlias), td.Name, chunk, r.query)
	return r, nil
}

// RebuildShard updates the SrvShard objects and underlying serving graph.
//
// Re-read from TopologyServer to make sure we are using the side
// effects of all actions.
//
// This function will start each cell over from the beginning on ErrBadVersion,
// so it doesn't need a lock on the shard.
func RebuildShard(ctx context.Context, log logutil.Logger, ts topo.Server, keyspace, shard string, cells []string, lockTimeout time.Duration) (*topo.ShardInfo, error) {
	log.Infof("RebuildShard %v/%v", keyspace, shard)

	span := trace.NewSpanFromContext(ctx)
	span.StartLocal("topotools.RebuildShard")
	defer span.Finish()
	ctx = trace.NewContext(ctx, span)

	// read the existing shard info. It has to exist.
	shardInfo, err := ts.GetShard(ctx, keyspace, shard)
	if err != nil {
		return nil, err
	}

	// rebuild all cells in parallel
	wg := sync.WaitGroup{}
	rec := concurrency.AllErrorRecorder{}
	for _, cell := range shardInfo.Cells {
		// skip this cell if we shouldn't rebuild it
		if !topo.InCellList(cell, cells) {
			continue
		}

		wg.Add(1)
		go func(cell string) {
			defer wg.Done()
			rec.RecordError(rebuildCellSrvShard(ctx, log, ts, shardInfo, cell))
		}(cell)
	}
	wg.Wait()

	return shardInfo, rec.Error()
}

// MakeSplitCreateTableSql returns a table creation statement
// that is modified to be faster, and the associated optional
// 'alter table' to modify the table at the end.
// - If the strategy contains the string 'skipAutoIncrement(NNN)' then
// we do not re-add the auto_increment on that table.
// - If the strategy contains the string 'delaySecondaryIndexes',
// then non-primary key indexes will be added afterwards.
// - If the strategy contains the string 'useMyIsam' we load
// the data into a myisam table and we then convert to innodb
// - If the strategy contains the string 'delayPrimaryKey',
// then the primary key index will be added afterwards (use with useMyIsam)
func MakeSplitCreateTableSql(logger logutil.Logger, schema, databaseName, tableName string, strategy string) (string, string, error) {
	alters := make([]string, 0, 5)
	lines := strings.Split(schema, "\n")
	delayPrimaryKey := strings.Contains(strategy, "delayPrimaryKey")
	delaySecondaryIndexes := strings.Contains(strategy, "delaySecondaryIndexes")
	useMyIsam := strings.Contains(strategy, "useMyIsam")

	for i, line := range lines {
		if strings.HasPrefix(line, "CREATE TABLE `") {
			lines[i] = strings.Replace(line, "CREATE TABLE `", "CREATE TABLE `"+databaseName+"`.`", 1)
			continue
		}

		if strings.Contains(line, " AUTO_INCREMENT") {
			// only add to the final ALTER TABLE if we're not
			// dropping the AUTO_INCREMENT on the table
			if strings.Contains(strategy, "skipAutoIncrement("+tableName+")") {
				logger.Infof("Will not add AUTO_INCREMENT back on table %v", tableName)
			} else {
				alters = append(alters, "MODIFY "+line[:len(line)-1])
			}
			lines[i] = strings.Replace(line, " AUTO_INCREMENT", "", 1)
			continue
		}

		isPrimaryKey := strings.Contains(line, " PRIMARY KEY")
		isSecondaryIndex := !isPrimaryKey && strings.Contains(line, " KEY")
		if (isPrimaryKey && delayPrimaryKey) || (isSecondaryIndex && delaySecondaryIndexes) {

			// remove the comma at the end of the previous line,
			lines[i-1] = lines[i-1][:len(lines[i-1])-1]

			// keep our comma if any (so the next index
			// might remove it)
			// also add the key definition to the alters
			if strings.HasSuffix(line, ",") {
				lines[i] = ","
				alters = append(alters, "ADD "+line[:len(line)-1])
			} else {
				lines[i] = ""
				alters = append(alters, "ADD "+line)
			}
		}

		if useMyIsam && strings.Contains(line, " ENGINE=InnoDB") {
			lines[i] = strings.Replace(line, " ENGINE=InnoDB", " ENGINE=MyISAM", 1)
			alters = append(alters, "ENGINE=InnoDB")
		}
	}

	alter := ""
	if len(alters) > 0 {
		alter = "ALTER TABLE `" + databaseName + "`.`" + tableName + "` " + strings.Join(alters, ", ")
	}
	return strings.Join(lines, "\n"), alter, nil
}

// TableScanByKeyRange returns a QueryResultReader that gets all the
// rows from a table that match the supplied KeyRange, ordered by
// Primary Key. The returned columns are ordered with the Primary Key
// columns in front.
func TableScanByKeyRange(ctx context.Context, log logutil.Logger, ts topo.Server, tabletAlias *topodatapb.TabletAlias, tableDefinition *tabletmanagerdatapb.TableDefinition, keyRange *topodatapb.KeyRange, shardingColumnName string, shardingColumnType topodatapb.KeyspaceIdType) (*QueryResultReader, error) {
	where := ""
	// TODO(aaijazi): this currently only works with V2 style sharding keys.
	// We should add FilteredQueryResultReader that will do a full table scan, with client-side
	// filtering to remove rows that don't map to the keyrange that we want.
	if keyRange != nil {
		switch shardingColumnType {
		case topodatapb.KeyspaceIdType_UINT64:
			if len(keyRange.Start) > 0 {
				if len(keyRange.End) > 0 {
					// have start & end
					where = fmt.Sprintf("WHERE %v >= %v AND %v < %v ", shardingColumnName, uint64FromKeyspaceID(keyRange.Start), shardingColumnName, uint64FromKeyspaceID(keyRange.End))
				} else {
					// have start only
					where = fmt.Sprintf("WHERE %v >= %v ", shardingColumnName, uint64FromKeyspaceID(keyRange.Start))
				}
			} else {
				if len(keyRange.End) > 0 {
					// have end only
					where = fmt.Sprintf("WHERE %v < %v ", shardingColumnName, uint64FromKeyspaceID(keyRange.End))
				}
			}
		case topodatapb.KeyspaceIdType_BYTES:
			if len(keyRange.Start) > 0 {
				if len(keyRange.End) > 0 {
					// have start & end
					where = fmt.Sprintf("WHERE HEX(%v) >= '%v' AND HEX(%v) < '%v' ", shardingColumnName, hex.EncodeToString(keyRange.Start), shardingColumnName, hex.EncodeToString(keyRange.End))
				} else {
					// have start only
					where = fmt.Sprintf("WHERE HEX(%v) >= '%v' ", shardingColumnName, hex.EncodeToString(keyRange.Start))
				}
			} else {
				if len(keyRange.End) > 0 {
					// have end only
					where = fmt.Sprintf("WHERE HEX(%v) < '%v' ", shardingColumnName, hex.EncodeToString(keyRange.End))
				}
			}
		default:
			return nil, fmt.Errorf("Unsupported ShardingColumnType: %v", shardingColumnType)
		}
	}

	sql := fmt.Sprintf("SELECT %v FROM %v %vORDER BY %v", strings.Join(orderedColumns(tableDefinition), ", "), tableDefinition.Name, where, strings.Join(tableDefinition.PrimaryKeyColumns, ", "))
	log.Infof("SQL query for %v/%v: %v", topoproto.TabletAliasString(tabletAlias), tableDefinition.Name, sql)
	return NewQueryResultReaderForTablet(ctx, ts, tabletAlias, sql)
}

// TableScanByKeyRange returns a QueryResultReader that gets all the
// rows from a table that match the supplied KeyRange, ordered by
// Primary Key. The returned columns are ordered with the Primary Key
// columns in front.
func TableScanByKeyRange(ctx context.Context, log logutil.Logger, ts topo.Server, tabletAlias *topodatapb.TabletAlias, tableDefinition *tabletmanagerdatapb.TableDefinition, keyRange *topodatapb.KeyRange, keyspaceIDType topodatapb.KeyspaceIdType) (*QueryResultReader, error) {
	where := ""
	if keyRange != nil {
		switch keyspaceIDType {
		case topodatapb.KeyspaceIdType_UINT64:
			if len(keyRange.Start) > 0 {
				if len(keyRange.End) > 0 {
					// have start & end
					where = fmt.Sprintf("WHERE keyspace_id >= %v AND keyspace_id < %v ", uint64FromKeyspaceID(keyRange.Start), uint64FromKeyspaceID(keyRange.End))
				} else {
					// have start only
					where = fmt.Sprintf("WHERE keyspace_id >= %v ", uint64FromKeyspaceID(keyRange.Start))
				}
			} else {
				if len(keyRange.End) > 0 {
					// have end only
					where = fmt.Sprintf("WHERE keyspace_id < %v ", uint64FromKeyspaceID(keyRange.End))
				}
			}
		case topodatapb.KeyspaceIdType_BYTES:
			if len(keyRange.Start) > 0 {
				if len(keyRange.End) > 0 {
					// have start & end
					where = fmt.Sprintf("WHERE HEX(keyspace_id) >= '%v' AND HEX(keyspace_id) < '%v' ", hex.EncodeToString(keyRange.Start), hex.EncodeToString(keyRange.End))
				} else {
					// have start only
					where = fmt.Sprintf("WHERE HEX(keyspace_id) >= '%v' ", hex.EncodeToString(keyRange.Start))
				}
			} else {
				if len(keyRange.End) > 0 {
					// have end only
					where = fmt.Sprintf("WHERE HEX(keyspace_id) < '%v' ", hex.EncodeToString(keyRange.End))
				}
			}
		default:
			return nil, fmt.Errorf("Unsupported KeyspaceIdType: %v", keyspaceIDType)
		}
	}

	sql := fmt.Sprintf("SELECT %v FROM %v %vORDER BY %v", strings.Join(orderedColumns(tableDefinition), ", "), tableDefinition.Name, where, strings.Join(tableDefinition.PrimaryKeyColumns, ", "))
	log.Infof("SQL query for %v/%v: %v", topoproto.TabletAliasString(tabletAlias), tableDefinition.Name, sql)
	return NewQueryResultReaderForTablet(ctx, ts, tabletAlias, sql)
}

// RestartSlavesExternal will tell all the slaves in the provided list
// that they have a new master, and also tell all the masters. The
// masters will be scrapped if they don't answer.
// We execute all the actions in parallel.
func RestartSlavesExternal(ts topo.Server, log logutil.Logger, slaveTabletMap, masterTabletMap map[topo.TabletAlias]*topo.TabletInfo, masterElectTabletAlias topo.TabletAlias, slaveWasRestarted func(*topo.TabletInfo, *actionnode.SlaveWasRestartedArgs) error) {
	wg := sync.WaitGroup{}

	swrd := actionnode.SlaveWasRestartedArgs{
		Parent: masterElectTabletAlias,
	}

	log.Infof("Updating individual tablets with the right master...")

	// do all the slaves
	for _, ti := range slaveTabletMap {
		wg.Add(1)
		go func(ti *topo.TabletInfo) {
			if err := slaveWasRestarted(ti, &swrd); err != nil {
				log.Warningf("Slave %v had an error: %v", ti.Alias, err)
			}
			wg.Done()
		}(ti)
	}

	// and do the old master and any straggler, if possible.
	for _, ti := range masterTabletMap {
		wg.Add(1)
		go func(ti *topo.TabletInfo) {
			err := slaveWasRestarted(ti, &swrd)
			if err != nil {
				// the old master can be annoying if left
				// around in the replication graph, so if we
				// can't restart it, we just scrap it.
				// We don't rebuild the Shard just yet though.
				log.Warningf("Old master %v is not restarting in time, forcing it to spare: %v", ti.Alias, err)

				ti.Type = topo.TYPE_SPARE
				ti.Parent = masterElectTabletAlias
				if err := topo.UpdateTablet(ts, ti); err != nil {
					log.Warningf("Failed to change old master %v to spare: %v", ti.Alias, err)
				}
			}
			wg.Done()
		}(ti)
	}
	wg.Wait()
}

// TableScanByKeyRange returns a QueryResultReader that gets all the
// rows from a table that match the supplied KeyRange, ordered by
// Primary Key. The returned columns are ordered with the Primary Key
// columns in front.
func TableScanByKeyRange(ctx context.Context, log logutil.Logger, ts topo.Server, tabletAlias topo.TabletAlias, tableDefinition *myproto.TableDefinition, keyRange key.KeyRange, keyspaceIdType key.KeyspaceIdType) (*QueryResultReader, error) {
	where := ""
	switch keyspaceIdType {
	case key.KIT_UINT64:
		if keyRange.Start != key.MinKey {
			if keyRange.End != key.MaxKey {
				// have start & end
				where = fmt.Sprintf("WHERE keyspace_id >= %v AND keyspace_id < %v ", uint64FromKeyspaceId(keyRange.Start), uint64FromKeyspaceId(keyRange.End))
			} else {
				// have start only
				where = fmt.Sprintf("WHERE keyspace_id >= %v ", uint64FromKeyspaceId(keyRange.Start))
			}
		} else {
			if keyRange.End != key.MaxKey {
				// have end only
				where = fmt.Sprintf("WHERE keyspace_id < %v ", uint64FromKeyspaceId(keyRange.End))
			}
		}
	case key.KIT_BYTES:
		if keyRange.Start != key.MinKey {
			if keyRange.End != key.MaxKey {
				// have start & end
				where = fmt.Sprintf("WHERE HEX(keyspace_id) >= '%v' AND HEX(keyspace_id) < '%v' ", keyRange.Start.Hex(), keyRange.End.Hex())
			} else {
				// have start only
				where = fmt.Sprintf("WHERE HEX(keyspace_id) >= '%v' ", keyRange.Start.Hex())
			}
		} else {
			if keyRange.End != key.MaxKey {
				// have end only
				where = fmt.Sprintf("WHERE HEX(keyspace_id) < '%v' ", keyRange.End.Hex())
			}
		}
	default:
		return nil, fmt.Errorf("Unsupported KeyspaceIdType: %v", keyspaceIdType)
	}

	sql := fmt.Sprintf("SELECT %v FROM %v %vORDER BY %v", strings.Join(orderedColumns(tableDefinition), ", "), tableDefinition.Name, where, strings.Join(tableDefinition.PrimaryKeyColumns, ", "))
	log.Infof("SQL query for %v/%v: %v", tabletAlias, tableDefinition.Name, sql)
	return NewQueryResultReaderForTablet(ctx, ts, tabletAlias, sql)
}

// This piece runs on the presumably empty machine acting as the target in the
// create replica action.
//
// validate target (self)
// shutdown_mysql()
// create temp data directory /vt/target/vt_<keyspace>
// copy compressed data files via HTTP
// verify hash of compressed files
// uncompress into /vt/vt_<target-uid>/data/vt_<keyspace>
// start_mysql()
// clean up compressed files
func (mysqld *Mysqld) RestoreFromSnapshot(logger logutil.Logger, snapshotManifest *SnapshotManifest, fetchConcurrency, fetchRetryCount int, dontWaitForSlaveStart bool, hookExtraEnv map[string]string) error {
	if snapshotManifest == nil {
		return errors.New("RestoreFromSnapshot: nil snapshotManifest")
	}

	logger.Infof("ValidateCloneTarget")
	if err := mysqld.ValidateCloneTarget(hookExtraEnv); err != nil {
		return err
	}

	logger.Infof("Shutdown mysqld")
	if err := mysqld.Shutdown(true, MysqlWaitTime); err != nil {
		return err
	}

	logger.Infof("Fetch snapshot")
	if err := mysqld.fetchSnapshot(snapshotManifest, fetchConcurrency, fetchRetryCount); err != nil {
		return err
	}

	logger.Infof("Restart mysqld")
	if err := mysqld.Start(MysqlWaitTime); err != nil {
		return err
	}

	cmdList, err := mysqld.StartReplicationCommands(snapshotManifest.ReplicationStatus)
	if err != nil {
		return err
	}
	if err := mysqld.ExecuteSuperQueryList(cmdList); err != nil {
		return err
	}

	if !dontWaitForSlaveStart {
		if err := mysqld.WaitForSlaveStart(SlaveStartDeadline); err != nil {
			return err
		}
	}

	h := hook.NewSimpleHook("postflight_restore")
	h.ExtraEnv = hookExtraEnv
	if err := h.ExecuteOptional(); err != nil {
		return err
	}

	return nil
}

// RestartSlavesExternal will tell all the slaves in the provided list
// that they have a new master, and also tell all the masters. The
// masters will be scrapped if they don't answer.
// We execute all the actions in parallel.
func RestartSlavesExternal(ts topo.Server, log logutil.Logger, slaveTabletMap, masterTabletMap map[topodatapb.TabletAlias]*topo.TabletInfo, masterElectTabletAlias *topodatapb.TabletAlias, slaveWasRestarted func(*topo.TabletInfo, *topodatapb.TabletAlias) error) {
	wg := sync.WaitGroup{}

	log.Infof("Updating individual tablets with the right master...")

	// do all the slaves
	for _, ti := range slaveTabletMap {
		wg.Add(1)
		go func(ti *topo.TabletInfo) {
			if err := slaveWasRestarted(ti, masterElectTabletAlias); err != nil {
				log.Warningf("Slave %v had an error: %v", ti.Alias, err)
			}
			wg.Done()
		}(ti)
	}

	// and do the old master and any straggler, if possible.
	for _, ti := range masterTabletMap {
		wg.Add(1)
		go func(ti *topo.TabletInfo) {
			err := slaveWasRestarted(ti, masterElectTabletAlias)
			if err != nil {
				// the old master can be annoying if left
				// around in the replication graph, so if we
				// can't restart it, we just make it spare.
				log.Warningf("Old master %v is not restarting in time, forcing it to spare: %v", ti.Alias, err)

				ti.Type = topodatapb.TabletType_SPARE
				if err := ts.UpdateTablet(context.TODO(), ti); err != nil {
					log.Warningf("Failed to change old master %v to spare: %v", ti.Alias, err)
				}
			}
			wg.Done()
		}(ti)
	}
	wg.Wait()
}

func (mysqld *Mysqld) restoreAfterSnapshot(logger logutil.Logger, slaveStartRequired, readOnly bool, hookExtraEnv map[string]string, connToRelease dbconnpool.PoolConnection) (err error) {
	// Try to fix mysqld regardless of snapshot success..
	logger.Infof("exec UNLOCK TABLES")
	_, err = connToRelease.ExecuteFetch("UNLOCK TABLES", 10000, false)
	connToRelease.Recycle()
	if err != nil {
		return fmt.Errorf("failed to UNLOCK TABLES: %v", err)
	}

	// restore original mysqld state that we saved above
	if slaveStartRequired {
		if err = mysqld.StartSlave(hookExtraEnv); err != nil {
			return
		}
		// this should be quick, but we might as well just wait
		if err = mysqld.WaitForSlaveStart(5); err != nil {
			return
		}
	}
	if err = mysqld.SetReadOnly(readOnly); err != nil {
		return
	}
	return nil
}

func (mysqld *Mysqld) prepareToSnapshot(logger logutil.Logger, allowHierarchicalReplication bool, hookExtraEnv map[string]string) (slaveStartRequired, readOnly bool, replicationPosition, myMasterPosition proto.ReplicationPosition, masterAddr string, connToRelease dbconnpool.PoolConnection, err error) {
	// save initial state so we can restore on Start()
	if slaveStatus, slaveErr := mysqld.SlaveStatus(); slaveErr == nil {
		slaveStartRequired = slaveStatus.SlaveRunning()
	}

	// For masters, set read-only so we don't write anything during snapshot
	readOnly = true
	if readOnly, err = mysqld.IsReadOnly(); err != nil {
		return
	}

	logger.Infof("Set Read Only")
	if !readOnly {
		mysqld.SetReadOnly(true)
	}
	logger.Infof("Stop Slave")
	if err = mysqld.StopSlave(hookExtraEnv); err != nil {
		return
	}

	// Get the replication position and master addr
	replicationPosition, masterAddr, err = mysqld.getReplicationPositionForClones(allowHierarchicalReplication)
	if err != nil {
		return
	}

	// get our master position, some targets may use it
	myMasterPosition, err = mysqld.MasterPosition()
	if err != nil && err != ErrNotMaster {
		// this is a real error
		return
	}

	logger.Infof("Flush tables")
	if connToRelease, err = mysqld.dbaPool.Get(0); err != nil {
		return
	}
	logger.Infof("exec FLUSH TABLES WITH READ LOCK")
	if _, err = connToRelease.ExecuteFetch("FLUSH TABLES WITH READ LOCK", 10000, false); err != nil {
		connToRelease.Recycle()
		return
	}

	return
}

// Update shard file with new master, replicas, etc.
//
// Re-read from TopologyServer to make sure we are using the side
// effects of all actions.
//
// This function locks individual SvrShard paths, so it doesn't need a lock
// on the shard.
func RebuildShard(log logutil.Logger, ts topo.Server, keyspace, shard string, cells []string, timeout time.Duration, interrupted chan struct{}) error {
	log.Infof("RebuildShard %v/%v", keyspace, shard)

	// read the existing shard info. It has to exist.
	shardInfo, err := ts.GetShard(keyspace, shard)
	if err != nil {
		return err
	}

	// rebuild all cells in parallel
	wg := sync.WaitGroup{}
	rec := concurrency.AllErrorRecorder{}
	for _, cell := range shardInfo.Cells {
		// skip this cell if we shouldn't rebuild it
		if !topo.InCellList(cell, cells) {
			continue
		}

		// start with the master if it's in the current cell
		tabletsAsMap := make(map[topo.TabletAlias]bool)
		if shardInfo.MasterAlias.Cell == cell {
			tabletsAsMap[shardInfo.MasterAlias] = true
		}

		wg.Add(1)
		go func(cell string) {
			defer wg.Done()

			// read the ShardReplication object to find tablets
			sri, err := ts.GetShardReplication(cell, keyspace, shard)
			if err != nil {
				rec.RecordError(fmt.Errorf("GetShardReplication(%v, %v, %v) failed: %v", cell, keyspace, shard, err))
				return
			}

			// add all relevant tablets to the map
			for _, rl := range sri.ReplicationLinks {
				tabletsAsMap[rl.TabletAlias] = true
				if rl.Parent.Cell == cell {
					tabletsAsMap[rl.Parent] = true
				}
			}

			// convert the map to a list
			aliases := make([]topo.TabletAlias, 0, len(tabletsAsMap))
			for a := range tabletsAsMap {
				aliases = append(aliases, a)
			}

			// read all the Tablet records
			tablets, err := topo.GetTabletMap(ts, aliases)
			switch err {
			case nil:
				// keep going, we're good
			case topo.ErrPartialResult:
				log.Warningf("Got ErrPartialResult from topo.GetTabletMap in cell %v, some tablets may not be added properly to serving graph", cell)
			default:
				rec.RecordError(fmt.Errorf("GetTabletMap in cell %v failed: %v", cell, err))
				return
			}

			// Lock the SrvShard so we write a consistent data set.
			actionNode := actionnode.RebuildSrvShard()
			lockPath, err := actionNode.LockSrvShard(ts, cell, keyspace, shard, timeout, interrupted)
			if err != nil {
				rec.RecordError(err)
				return
			}

			// write the data we need to
			rebuildErr := rebuildCellSrvShard(log, ts, shardInfo, cell, tablets)

			// and unlock
			if err := actionNode.UnlockSrvShard(ts, cell, keyspace, shard, lockPath, rebuildErr); err != nil {
				rec.RecordError(err)
			}
		}(cell)
	}
	wg.Wait()

	return rec.Error()
}

func (mysqld *Mysqld) createSnapshot(logger logutil.Logger, concurrency int, serverMode bool) ([]SnapshotFile, error) {
	sources := make([]string, 0, 128)
	destinations := make([]string, 0, 128)

	// clean out and start fresh
	logger.Infof("removing previous snapshots: %v", mysqld.SnapshotDir)
	if err := os.RemoveAll(mysqld.SnapshotDir); err != nil {
		return nil, err
	}

	// FIXME(msolomon) innodb paths must match patterns in mycnf -
	// probably belongs as a derived path.
	type snapPair struct{ srcDir, dstDir string }
	dps := []snapPair{
		{mysqld.config.InnodbDataHomeDir, path.Join(mysqld.SnapshotDir, innodbDataSubdir)},
		{mysqld.config.InnodbLogGroupHomeDir, path.Join(mysqld.SnapshotDir, innodbLogSubdir)},
	}

	dataDirEntries, err := ioutil.ReadDir(mysqld.config.DataDir)
	if err != nil {
		return nil, err
	}

	for _, de := range dataDirEntries {
		dbDirPath := path.Join(mysqld.config.DataDir, de.Name())
		// If this is not a directory, try to eval it as a syslink.
		if !de.IsDir() {
			dbDirPath, err = filepath.EvalSymlinks(dbDirPath)
			if err != nil {
				return nil, err
			}
			de, err = os.Stat(dbDirPath)
			if err != nil {
				return nil, err
			}
		}
		if de.IsDir() {
			// Copy anything that defines a db.opt file - that includes empty databases.
			_, err := os.Stat(path.Join(dbDirPath, "db.opt"))
			if err == nil {
				dps = append(dps, snapPair{dbDirPath, path.Join(mysqld.SnapshotDir, dataDir, de.Name())})
			} else {
				// Look for at least one .frm file
				dbDirEntries, err := ioutil.ReadDir(dbDirPath)
				if err == nil {
					for _, dbEntry := range dbDirEntries {
						if strings.HasSuffix(dbEntry.Name(), ".frm") {
							dps = append(dps, snapPair{dbDirPath, path.Join(mysqld.SnapshotDir, dataDir, de.Name())})
							break
						}
					}
				} else {
					return nil, err
				}
			}
		}
	}

	for _, dp := range dps {
		if err := os.MkdirAll(dp.dstDir, 0775); err != nil {
			return nil, err
		}
		if s, d, err := findFilesToServe(dp.srcDir, dp.dstDir, !serverMode); err != nil {
			return nil, err
		} else {
			sources = append(sources, s...)
			destinations = append(destinations, d...)
		}
	}

	return newSnapshotFiles(sources, destinations, mysqld.SnapshotDir, concurrency, !serverMode)
}

//.........这里部分代码省略.........
		return
	}

	// save initial state so we can restore on Start()
	slaveStartRequired = false
	sourceIsMaster := false
	readOnly = true

	slaveStatus, err := mysqld.SlaveStatus()
	if err == nil {
		slaveStartRequired = slaveStatus.SlaveRunning()
	} else if err == ErrNotSlave {
		sourceIsMaster = true
	} else {
		// If we can't get any data, just fail.
		return
	}

	readOnly, err = mysqld.IsReadOnly()
	if err != nil {
		return
	}

	// Stop sources of writes so we can get a consistent replication position.
	// If the source is a slave use the master replication position
	// unless we are allowing hierachical replicas.
	masterAddr := ""
	var replicationPosition proto.ReplicationPosition
	if sourceIsMaster {
		if err = mysqld.SetReadOnly(true); err != nil {
			return
		}
		replicationPosition, err = mysqld.MasterPosition()
		if err != nil {
			return
		}
		masterAddr = mysqld.IpAddr()
	} else {
		if err = mysqld.StopSlave(hookExtraEnv); err != nil {
			return
		}
		var slaveStatus *proto.ReplicationStatus
		slaveStatus, err = mysqld.SlaveStatus()
		if err != nil {
			return
		}
		replicationPosition = slaveStatus.Position

		// We are a slave, check our replication strategy before
		// choosing the master address.
		if allowHierarchicalReplication {
			masterAddr = mysqld.IpAddr()
		} else {
			masterAddr, err = mysqld.GetMasterAddr()
			if err != nil {
				return
			}
		}
	}

	if err = mysqld.Shutdown(true, MysqlWaitTime); err != nil {
		return
	}

	var smFile string
	dataFiles, snapshotErr := mysqld.createSnapshot(logger, concurrency, serverMode)
	if snapshotErr != nil {
		logger.Errorf("CreateSnapshot failed: %v", snapshotErr)
	} else {
		var sm *SnapshotManifest
		sm, snapshotErr = newSnapshotManifest(sourceAddr, mysqld.IpAddr(),
			masterAddr, dbName, dataFiles, replicationPosition, proto.ReplicationPosition{})
		if snapshotErr != nil {
			logger.Errorf("CreateSnapshot failed: %v", snapshotErr)
		} else {
			smFile = path.Join(mysqld.SnapshotDir, SnapshotManifestFile)
			if snapshotErr = writeJson(smFile, sm); snapshotErr != nil {
				logger.Errorf("CreateSnapshot failed: %v", snapshotErr)
			}
		}
	}

	// restore our state if required
	if serverMode && snapshotErr == nil {
		logger.Infof("server mode snapshot worked, not restarting mysql")
	} else {
		if err = mysqld.SnapshotSourceEnd(slaveStartRequired, readOnly, false /*deleteSnapshot*/, hookExtraEnv); err != nil {
			return
		}
	}

	if snapshotErr != nil {
		return "", slaveStartRequired, readOnly, snapshotErr
	}
	relative, err := filepath.Rel(mysqld.SnapshotDir, smFile)
	if err != nil {
		return "", slaveStartRequired, readOnly, nil
	}
	return path.Join(SnapshotURLPath, relative), slaveStartRequired, readOnly, nil
}