Golang concurrency.AllErrorRecorder类代码示例

func (wr *Wrangler) getMastersPosition(shards []*topo.ShardInfo) (map[*topo.ShardInfo]*mysqlctl.ReplicationPosition, error) {
	mu := sync.Mutex{}
	result := make(map[*topo.ShardInfo]*mysqlctl.ReplicationPosition)

	wg := sync.WaitGroup{}
	rec := concurrency.AllErrorRecorder{}
	for _, si := range shards {
		wg.Add(1)
		go func(si *topo.ShardInfo) {
			log.Infof("Gathering master position for %v", si.MasterAlias)
			pos, err := wr.getMasterPosition(si.MasterAlias)
			if err != nil {
				rec.RecordError(err)
			} else {
				log.Infof("Got master position for %v", si.MasterAlias)
				mu.Lock()
				result[si] = pos
				mu.Unlock()
			}
			wg.Done()
		}(si)
	}
	wg.Wait()
	return result, rec.Error()
}

// DeleteKeyspaceShards implements topo.Server.
func (s *Server) DeleteKeyspaceShards(ctx context.Context, keyspace string) error {
	shards, err := s.GetShardNames(ctx, keyspace)
	if err != nil {
		return err
	}

	wg := sync.WaitGroup{}
	rec := concurrency.AllErrorRecorder{}
	global := s.getGlobal()
	for _, shard := range shards {
		wg.Add(1)
		go func(shard string) {
			defer wg.Done()
			_, err := global.Delete(shardDirPath(keyspace, shard), true /* recursive */)
			rec.RecordError(convertError(err))
		}(shard)
	}
	wg.Wait()

	if err = rec.Error(); err != nil {
		return err
	}

	event.Dispatch(&events.KeyspaceChange{
		KeyspaceInfo: *topo.NewKeyspaceInfo(keyspace, nil, -1),
		Status:       "deleted all shards",
	})
	return nil
}

// CleanUp will run the recorded actions.
// If an action on a target fails, it will not run the next action on
// the same target.
// We return the aggregate errors for all cleanups.
// TODO(alainjobart) Actions should run concurrently on a per target
// basis. They are then serialized on each target.
func (cleaner *Cleaner) CleanUp(wr *Wrangler) error {
	actionMap := make(map[string]*cleanUpHelper)
	rec := concurrency.AllErrorRecorder{}
	cleaner.mu.Lock()
	for i := len(cleaner.actions) - 1; i >= 0; i-- {
		actionReference := cleaner.actions[i]
		helper, ok := actionMap[actionReference.target]
		if !ok {
			helper = &cleanUpHelper{
				err: nil,
			}
			actionMap[actionReference.target] = helper
		}
		if helper.err != nil {
			log.Warningf("previous action failed on target %v, no running %v", actionReference.target, actionReference.name)
			continue
		}
		err := actionReference.action.CleanUp(wr)
		if err != nil {
			helper.err = err
			rec.RecordError(err)
			log.Errorf("action %v failed on %v: %v", actionReference.name, actionReference.target, err)
		} else {
			log.Infof("action %v successfull on %v", actionReference.name, actionReference.target)
		}
	}
	cleaner.mu.Unlock()
	return rec.Error()
}

// FindAllTabletAliasesInShardByCell uses the replication graph to find all the
// tablet aliases in the given shard.
//
// It can return ErrPartialResult if some cells were not fetched,
// in which case the result only contains the cells that were fetched.
//
// The tablet aliases are sorted by cell, then by UID.
func FindAllTabletAliasesInShardByCell(ctx context.Context, ts Server, keyspace, shard string, cells []string) ([]TabletAlias, error) {
	span := trace.NewSpanFromContext(ctx)
	span.StartLocal("topo.FindAllTabletAliasesInShardbyCell")
	span.Annotate("keyspace", keyspace)
	span.Annotate("shard", shard)
	span.Annotate("num_cells", len(cells))
	defer span.Finish()
	ctx = trace.NewContext(ctx, span)

	// read the shard information to find the cells
	si, err := GetShard(ctx, ts, keyspace, shard)
	if err != nil {
		return nil, err
	}

	resultAsMap := make(map[TabletAlias]bool)
	if si.MasterAlias != nil && !TabletAliasIsZero(si.MasterAlias) {
		if InCellList(si.MasterAlias.Cell, cells) {
			resultAsMap[ProtoToTabletAlias(si.MasterAlias)] = true
		}
	}

	// read the replication graph in each cell and add all found tablets
	wg := sync.WaitGroup{}
	mutex := sync.Mutex{}
	rec := concurrency.AllErrorRecorder{}
	for _, cell := range si.Cells {
		if !InCellList(cell, cells) {
			continue
		}
		wg.Add(1)
		go func(cell string) {
			defer wg.Done()
			sri, err := ts.GetShardReplication(ctx, cell, keyspace, shard)
			if err != nil {
				rec.RecordError(fmt.Errorf("GetShardReplication(%v, %v, %v) failed: %v", cell, keyspace, shard, err))
				return
			}

			mutex.Lock()
			for _, node := range sri.Nodes {
				resultAsMap[ProtoToTabletAlias(node.TabletAlias)] = true
			}
			mutex.Unlock()
		}(cell)
	}
	wg.Wait()
	err = nil
	if rec.HasErrors() {
		log.Warningf("FindAllTabletAliasesInShard(%v,%v): got partial result: %v", keyspace, shard, rec.Error())
		err = ErrPartialResult
	}

	result := make([]TabletAlias, 0, len(resultAsMap))
	for a := range resultAsMap {
		result = append(result, a)
	}
	sort.Sort(TabletAliasList(result))
	return result, err
}

// execShardAction executes the action on a particular shard.
// If the action fails, it determines whether the keyspace/shard
// have moved, re-resolves the topology and tries again, if it is
// not executing a transaction.
func (stc *ScatterConn) execShardAction(
	context interface{},
	keyspace string,
	shard string,
	tabletType topo.TabletType,
	session *SafeSession,
	action shardActionFunc,
	allErrors *concurrency.AllErrorRecorder,
	results chan interface{},
) {
	for {
		sdc := stc.getConnection(keyspace, shard, tabletType)
		transactionId, err := stc.updateSession(context, sdc, keyspace, shard, tabletType, session)
		if err != nil {
			allErrors.RecordError(err)
			return
		}
		err = action(sdc, transactionId, results)
		// Determine whether keyspace can be re-resolved
		if shouldResolveKeyspace(err, transactionId) {
			newKeyspace, err := getKeyspaceAlias(stc.toposerv, stc.cell, keyspace, tabletType)
			if err == nil && newKeyspace != keyspace {
				sdc.Close()
				stc.cleanupShardConn(keyspace, shard, tabletType)
				keyspace = newKeyspace
				continue
			}
		}
		if err != nil {
			allErrors.RecordError(err)
			return
		}
		break
	}
}

// WaitForDrain blocks until the selected tablets (cells/keyspace/shard/tablet_type)
// have reported a QPS rate of 0.0.
// NOTE: This is just an observation of one point in time and no guarantee that
// the tablet was actually drained. At later times, a QPS rate > 0.0 could still
// be observed.
func (wr *Wrangler) WaitForDrain(ctx context.Context, cells []string, keyspace, shard string, servedType topodatapb.TabletType,
	retryDelay, healthCheckTopologyRefresh, healthcheckRetryDelay, healthCheckTimeout time.Duration) error {
	if len(cells) == 0 {
		// Retrieve list of cells for the shard from the topology.
		shardInfo, err := wr.ts.GetShard(ctx, keyspace, shard)
		if err != nil {
			return fmt.Errorf("failed to retrieve list of all cells. GetShard() failed: %v", err)
		}
		cells = shardInfo.Cells
	}

	// Check all cells in parallel.
	wg := sync.WaitGroup{}
	rec := concurrency.AllErrorRecorder{}
	for _, cell := range cells {
		wg.Add(1)
		go func(cell string) {
			defer wg.Done()
			rec.RecordError(wr.waitForDrainInCell(ctx, cell, keyspace, shard, servedType,
				retryDelay, healthCheckTopologyRefresh, healthcheckRetryDelay, healthCheckTimeout))
		}(cell)
	}
	wg.Wait()

	return rec.Error()
}

// 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()
}

func (wr *Wrangler) waitForFilteredReplication(sourcePositions map[*topo.ShardInfo]myproto.ReplicationPosition, destinationShards []*topo.ShardInfo) error {
	wg := sync.WaitGroup{}
	rec := concurrency.AllErrorRecorder{}
	for _, si := range destinationShards {
		wg.Add(1)
		go func(si *topo.ShardInfo) {
			for _, sourceShard := range si.SourceShards {
				// we're waiting on this guy
				blpPosition := blproto.BlpPosition{
					Uid: sourceShard.Uid,
				}

				// find the position it should be at
				for s, pos := range sourcePositions {
					if s.Keyspace() == sourceShard.Keyspace && s.ShardName() == sourceShard.Shard {
						blpPosition.Position = pos
					}
				}

				log.Infof("Waiting for %v to catch up", si.MasterAlias)
				if err := wr.ai.WaitBlpPosition(si.MasterAlias, blpPosition, wr.ActionTimeout()); err != nil {
					rec.RecordError(err)
				} else {
					log.Infof("%v caught up", si.MasterAlias)
				}
				wg.Done()
			}
		}(si)
	}
	wg.Wait()
	return rec.Error()
}

// execShardAction executes the action on a particular shard.
// If the action fails, it determines whether the keyspace/shard
// have moved, re-resolves the topology and tries again, if it is
// not executing a transaction.
func (stc *ScatterConn) execShardAction(
	context context.Context,
	keyspace string,
	shard string,
	tabletType topo.TabletType,
	session *SafeSession,
	action shardActionFunc,
	allErrors *concurrency.AllErrorRecorder,
	results chan interface{},
) {
	for {
		sdc := stc.getConnection(context, keyspace, shard, tabletType)
		transactionId, err := stc.updateSession(context, sdc, keyspace, shard, tabletType, session)
		if err != nil {
			allErrors.RecordError(err)
			return
		}
		err = action(sdc, transactionId, results)
		if err != nil {
			allErrors.RecordError(err)
			return
		}
		break
	}
}

// CleanUp will run the recorded actions.
// If an action on a target fails, it will not run the next action on
// the same target.
// We return the aggregate errors for all cleanups.
// CleanUp uses its own context, with a timeout of 5 minutes, so that clean up action will run even if the original context times out.
// TODO(alainjobart) Actions should run concurrently on a per target
// basis. They are then serialized on each target.
func (cleaner *Cleaner) CleanUp(wr *Wrangler) error {
	// we use a background context so we're not dependent on the original context timeout
	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
	actionMap := make(map[string]*cleanUpHelper)
	rec := concurrency.AllErrorRecorder{}
	cleaner.mu.Lock()
	for i := len(cleaner.actions) - 1; i >= 0; i-- {
		actionReference := cleaner.actions[i]
		helper, ok := actionMap[actionReference.target]
		if !ok {
			helper = &cleanUpHelper{
				err: nil,
			}
			actionMap[actionReference.target] = helper
		}
		if helper.err != nil {
			wr.Logger().Warningf("previous action failed on target %v, no running %v", actionReference.target, actionReference.name)
			continue
		}
		err := actionReference.action.CleanUp(ctx, wr)
		if err != nil {
			helper.err = err
			rec.RecordError(err)
			wr.Logger().Errorf("action %v failed on %v: %v", actionReference.name, actionReference.target, err)
		} else {
			wr.Logger().Infof("action %v successful on %v", actionReference.name, actionReference.target)
		}
	}
	cleaner.mu.Unlock()
	cancel()
	return rec.Error()
}

// DiffSchemaToArray diffs two schemas and return the schema diffs if there is any.
func DiffSchemaToArray(leftName string, left *SchemaDefinition, rightName string, right *SchemaDefinition) (result []string) {
	er := concurrency.AllErrorRecorder{}
	DiffSchema(leftName, left, rightName, right, &er)
	if er.HasErrors() {
		return er.ErrorStrings()
	}
	return nil
}

// DiffPermissionsToArray difs two sets of permissions, and returns the difference
func DiffPermissionsToArray(leftName string, left *tabletmanagerdatapb.Permissions, rightName string, right *tabletmanagerdatapb.Permissions) (result []string) {
	er := concurrency.AllErrorRecorder{}
	DiffPermissions(leftName, left, rightName, right, &er)
	if er.HasErrors() {
		return er.ErrorStrings()
	}
	return nil
}

func DiffPermissionsToArray(leftName string, left *Permissions, rightName string, right *Permissions) (result []string) {
	er := concurrency.AllErrorRecorder{}
	DiffPermissions(leftName, left, rightName, right, &er)
	if er.HasErrors() {
		return er.Errors
	} else {
		return nil
	}
}

func (wr *Wrangler) ShardMultiRestore(keyspace, shard string, sources []topo.TabletAlias, tables []string, concurrency, fetchConcurrency, insertTableConcurrency, fetchRetryCount int, strategy string) error {

	// check parameters
	if len(tables) > 0 && len(sources) > 1 {
		return fmt.Errorf("ShardMultiRestore can only handle one source when tables are specified")
	}

	// lock the shard to perform the changes we need done
	actionNode := actionnode.ShardMultiRestore(&actionnode.MultiRestoreArgs{
		SrcTabletAliases:       sources,
		Concurrency:            concurrency,
		FetchConcurrency:       fetchConcurrency,
		InsertTableConcurrency: insertTableConcurrency,
		FetchRetryCount:        fetchRetryCount,
		Strategy:               strategy})
	lockPath, err := wr.lockShard(keyspace, shard, actionNode)
	if err != nil {
		return err
	}

	mrErr := wr.SetSourceShards(keyspace, shard, sources, tables)
	err = wr.unlockShard(keyspace, shard, actionNode, lockPath, mrErr)
	if err != nil {
		if mrErr != nil {
			log.Errorf("unlockShard got error back: %v", err)
			return mrErr
		}
		return err
	}
	if mrErr != nil {
		return mrErr
	}

	// find all tablets in the shard
	destTablets, err := topo.FindAllTabletAliasesInShard(wr.ts, keyspace, shard)
	if err != nil {
		return err
	}

	// now launch MultiRestore on all tablets we need to do
	rec := cc.AllErrorRecorder{}
	wg := sync.WaitGroup{}
	for _, tabletAlias := range destTablets {
		wg.Add(1)
		go func(tabletAlias topo.TabletAlias) {
			log.Infof("Starting multirestore on tablet %v", tabletAlias)
			err := wr.MultiRestore(tabletAlias, sources, concurrency, fetchConcurrency, insertTableConcurrency, fetchRetryCount, strategy)
			log.Infof("Multirestore on tablet %v is done (err=%v)", tabletAlias, err)
			rec.RecordError(err)
			wg.Done()
		}(tabletAlias)
	}
	wg.Wait()

	return rec.Error()
}

func (wr *Wrangler) makeMastersReadOnly(shards []*topo.ShardInfo) error {
	wg := sync.WaitGroup{}
	rec := concurrency.AllErrorRecorder{}
	for _, si := range shards {
		if si.MasterAlias.IsZero() {
			rec.RecordError(fmt.Errorf("Shard %v/%v has no master?", si.Keyspace(), si.ShardName()))
			continue
		}

		wg.Add(1)
		go func(si *topo.ShardInfo) {
			defer wg.Done()

			wr.Logger().Infof("Making master %v read-only", si.MasterAlias)
			ti, err := wr.ts.GetTablet(si.MasterAlias)
			if err != nil {
				rec.RecordError(err)
				return
			}

			if err = wr.tmc.SetReadOnly(ti, wr.ActionTimeout()); err != nil {
				rec.RecordError(err)
				return
			}
			wr.Logger().Infof("Master %v is now read-only", si.MasterAlias)
		}(si)
	}
	wg.Wait()
	return rec.Error()
}