func (wr *Wrangler) shardMultiRestore(keyspace, shard string, sources []topo.TabletAlias, concurrency, fetchConcurrency, insertTableConcurrency, fetchRetryCount int, strategy string) error {
// read the shard
shardInfo, err := wr.ts.GetShard(keyspace, shard)
if err != nil {
return err
}
// read the source tablets
sourceTablets, err := GetTabletMap(wr.TopoServer(), sources)
if err != nil {
return err
}
// insert their KeyRange / shard in the SourceShards array
shardInfo.SourceShards = make([]topo.SourceShard, 0, len(sourceTablets))
for _, ti := range sourceTablets {
overlap, err := key.KeyRangesOverlap(shardInfo.KeyRange, ti.KeyRange)
if err != nil {
return fmt.Errorf("Source shard %v doesn't overlap destination shard %v", ti.KeyRange, shardInfo.KeyRange)
}
ss := topo.SourceShard{
KeyRange: overlap,
}
shardInfo.SourceShards = append(shardInfo.SourceShards, ss)
}
// and write the shard
if err = wr.ts.UpdateShard(shardInfo); err != nil {
return err
}
return nil
}
func (bpc *BinlogPlayerController) Iteration() (err error) {
defer func() {
if x := recover(); x != nil {
log.Errorf("%v: caught panic: %v", bpc, x)
err = fmt.Errorf("panic: %v", x)
}
}()
// Enable any user to set the timestamp.
// We do it on every iteration to be sure, in case mysql was
// restarted.
bpc.DisableSuperToSetTimestamp()
// create the db connection, connect it
vtClient := mysqlctl.NewDbClient(bpc.dbConfig)
if err := vtClient.Connect(); err != nil {
return fmt.Errorf("can't connect to database: %v", err)
}
defer vtClient.Close()
// Read the start position
startPosition, err := mysqlctl.ReadStartPosition(vtClient, bpc.sourceShard.Uid)
if err != nil {
return fmt.Errorf("can't read startPosition: %v", err)
}
// Find the server list for the source shard in our cell
addrs, err := bpc.ts.GetEndPoints(bpc.cell, bpc.sourceShard.Keyspace, bpc.sourceShard.Shard, topo.TYPE_REPLICA)
if err != nil {
return fmt.Errorf("can't find any source tablet for %v %v %v: %v", bpc.cell, bpc.sourceShard.String(), topo.TYPE_REPLICA, err)
}
if len(addrs.Entries) == 0 {
return fmt.Errorf("empty source tablet list for %v %v %v", bpc.cell, bpc.sourceShard.String(), topo.TYPE_REPLICA)
}
newServerIndex := rand.Intn(len(addrs.Entries))
addr := fmt.Sprintf("%v:%v", addrs.Entries[newServerIndex].Host, addrs.Entries[newServerIndex].NamedPortMap["_vtocc"])
// check which kind of replication we're doing, tables or keyrange
if len(bpc.sourceShard.Tables) > 0 {
// tables, just get them
player := mysqlctl.NewBinlogPlayerTables(vtClient, addr, bpc.sourceShard.Tables, startPosition, bpc.stopAtGroupId)
return player.ApplyBinlogEvents(bpc.interrupted)
} else {
// the data we have to replicate is the intersection of the
// source keyrange and our keyrange
overlap, err := key.KeyRangesOverlap(bpc.sourceShard.KeyRange, bpc.keyRange)
if err != nil {
return fmt.Errorf("Source shard %v doesn't overlap destination shard %v", bpc.sourceShard.KeyRange, bpc.keyRange)
}
player := mysqlctl.NewBinlogPlayerKeyRange(vtClient, addr, overlap, startPosition, bpc.stopAtGroupId)
return player.ApplyBinlogEvents(bpc.interrupted)
}
}
func (ta *TabletActor) multiRestore(actionNode *actionnode.ActionNode) (err error) {
args := actionNode.Args.(*actionnode.MultiRestoreArgs)
// read our current tablet, verify its state
// we only support restoring to the master or spare replicas
tablet, err := ta.ts.GetTablet(ta.tabletAlias)
if err != nil {
return err
}
if tablet.Type != topo.TYPE_MASTER && tablet.Type != topo.TYPE_SPARE && tablet.Type != topo.TYPE_REPLICA && tablet.Type != topo.TYPE_RDONLY {
return fmt.Errorf("expected master, spare replica or rdonly type, not %v: %v", tablet.Type, ta.tabletAlias)
}
// get source tablets addresses
sourceAddrs := make([]*url.URL, len(args.SrcTabletAliases))
keyRanges := make([]key.KeyRange, len(args.SrcTabletAliases))
for i, alias := range args.SrcTabletAliases {
t, e := ta.ts.GetTablet(alias)
if e != nil {
return e
}
sourceAddrs[i] = &url.URL{Host: t.GetAddr(), Path: "/" + t.DbName()}
keyRanges[i], e = key.KeyRangesOverlap(tablet.KeyRange, t.KeyRange)
if e != nil {
return e
}
}
// change type to restore, no change to replication graph
originalType := tablet.Type
tablet.Type = topo.TYPE_RESTORE
err = topo.UpdateTablet(ta.ts, tablet)
if err != nil {
return err
}
// run the action, scrap if it fails
if err := ta.mysqld.MultiRestore(tablet.DbName(), keyRanges, sourceAddrs, args.Concurrency, args.FetchConcurrency, args.InsertTableConcurrency, args.FetchRetryCount, args.Strategy); err != nil {
if e := Scrap(ta.ts, ta.tabletAlias, false); e != nil {
log.Errorf("Failed to Scrap after failed RestoreFromMultiSnapshot: %v", e)
}
return err
}
// restore type back
tablet.Type = originalType
return topo.UpdateTablet(ta.ts, tablet)
}
func (sdw *SplitDiffWorker) diff(ctx context.Context) error {
sdw.SetState(WorkerStateDiff)
sdw.wr.Logger().Infof("Gathering schema information...")
sdw.sourceSchemaDefinitions = make([]*myproto.SchemaDefinition, len(sdw.sourceAliases))
wg := sync.WaitGroup{}
rec := concurrency.AllErrorRecorder{}
wg.Add(1)
go func() {
var err error
shortCtx, cancel := context.WithTimeout(ctx, *remoteActionsTimeout)
sdw.destinationSchemaDefinition, err = sdw.wr.GetSchema(
shortCtx, sdw.destinationAlias, nil /* tables */, sdw.excludeTables, false /* includeViews */)
cancel()
rec.RecordError(err)
sdw.wr.Logger().Infof("Got schema from destination %v", sdw.destinationAlias)
wg.Done()
}()
for i, sourceAlias := range sdw.sourceAliases {
wg.Add(1)
go func(i int, sourceAlias pb.TabletAlias) {
var err error
shortCtx, cancel := context.WithTimeout(ctx, *remoteActionsTimeout)
sdw.sourceSchemaDefinitions[i], err = sdw.wr.GetSchema(
shortCtx, &sourceAlias, nil /* tables */, sdw.excludeTables, false /* includeViews */)
cancel()
rec.RecordError(err)
sdw.wr.Logger().Infof("Got schema from source[%v] %v", i, sourceAlias)
wg.Done()
}(i, *sourceAlias)
}
wg.Wait()
if rec.HasErrors() {
return rec.Error()
}
// TODO(alainjobart) Checking against each source may be
// overkill, if all sources have the same schema?
sdw.wr.Logger().Infof("Diffing the schema...")
rec = concurrency.AllErrorRecorder{}
for i, sourceSchemaDefinition := range sdw.sourceSchemaDefinitions {
sourceName := fmt.Sprintf("source[%v]", i)
myproto.DiffSchema("destination", sdw.destinationSchemaDefinition, sourceName, sourceSchemaDefinition, &rec)
}
if rec.HasErrors() {
sdw.wr.Logger().Warningf("Different schemas: %v", rec.Error().Error())
} else {
sdw.wr.Logger().Infof("Schema match, good.")
}
// run the diffs, 8 at a time
sdw.wr.Logger().Infof("Running the diffs...")
sem := sync2.NewSemaphore(8, 0)
for _, tableDefinition := range sdw.destinationSchemaDefinition.TableDefinitions {
wg.Add(1)
go func(tableDefinition *myproto.TableDefinition) {
defer wg.Done()
sem.Acquire()
defer sem.Release()
sdw.wr.Logger().Infof("Starting the diff on table %v", tableDefinition.Name)
if len(sdw.sourceAliases) != 1 {
err := fmt.Errorf("Don't support more than one source for table yet: %v", tableDefinition.Name)
rec.RecordError(err)
sdw.wr.Logger().Errorf(err.Error())
return
}
overlap, err := key.KeyRangesOverlap(sdw.shardInfo.KeyRange, sdw.shardInfo.SourceShards[0].KeyRange)
if err != nil {
newErr := fmt.Errorf("Source shard doesn't overlap with destination????: %v", err)
rec.RecordError(newErr)
sdw.wr.Logger().Errorf(newErr.Error())
return
}
sourceQueryResultReader, err := TableScanByKeyRange(ctx, sdw.wr.Logger(), sdw.wr.TopoServer(), sdw.sourceAliases[0], tableDefinition, overlap, key.ProtoToKeyspaceIdType(sdw.keyspaceInfo.ShardingColumnType))
if err != nil {
newErr := fmt.Errorf("TableScanByKeyRange(source) failed: %v", err)
rec.RecordError(newErr)
sdw.wr.Logger().Errorf(newErr.Error())
return
}
defer sourceQueryResultReader.Close()
destinationQueryResultReader, err := TableScanByKeyRange(ctx, sdw.wr.Logger(), sdw.wr.TopoServer(), sdw.destinationAlias, tableDefinition, nil, key.ProtoToKeyspaceIdType(sdw.keyspaceInfo.ShardingColumnType))
if err != nil {
newErr := fmt.Errorf("TableScanByKeyRange(destination) failed: %v", err)
rec.RecordError(newErr)
sdw.wr.Logger().Errorf(newErr.Error())
return
}
defer destinationQueryResultReader.Close()
differ, err := NewRowDiffer(sourceQueryResultReader, destinationQueryResultReader, tableDefinition)
if err != nil {
newErr := fmt.Errorf("NewRowDiffer() failed: %v", err)
rec.RecordError(newErr)
sdw.wr.Logger().Errorf(newErr.Error())
return
}
report, err := differ.Go(sdw.wr.Logger())
if err != nil {
newErr := fmt.Errorf("Differ.Go failed: %v", err.Error())
rec.RecordError(newErr)
sdw.wr.Logger().Errorf(newErr.Error())
} else {
if report.HasDifferences() {
err := fmt.Errorf("Table %v has differences: %v", tableDefinition.Name, report.String())
rec.RecordError(err)
sdw.wr.Logger().Warningf(err.Error())
} else {
sdw.wr.Logger().Infof("Table %v checks out (%v rows processed, %v qps)", tableDefinition.Name, report.processedRows, report.processingQPS)
}
}
}(tableDefinition)
}
wg.Wait()
return rec.Error()
}
// Iteration is a single iteration for the player: get the current status,
// try to play, and plays until interrupted, or until an error occurs.
func (bpc *BinlogPlayerController) Iteration() (err error) {
defer func() {
if x := recover(); x != nil {
log.Errorf("%v: caught panic: %v\n%s", bpc, x, tb.Stack(4))
err = fmt.Errorf("panic: %v", x)
}
}()
// Apply any special settings necessary for playback of binlogs.
// We do it on every iteration to be sure, in case MySQL was restarted.
if err := bpc.mysqld.EnableBinlogPlayback(); err != nil {
// We failed to apply the required settings, so we shouldn't keep going.
return err
}
// create the db connection, connect it
vtClient := bpc.vtClientFactory()
if err := vtClient.Connect(); err != nil {
return fmt.Errorf("can't connect to database: %v", err)
}
defer vtClient.Close()
// Read the start position
startPosition, flags, err := binlogplayer.ReadStartPosition(vtClient, bpc.sourceShard.Uid)
if err != nil {
return fmt.Errorf("can't read startPosition: %v", err)
}
// if we shouldn't start, we just error out and try again later
if strings.Index(flags, binlogplayer.BlpFlagDontStart) != -1 {
return fmt.Errorf("not starting because flag '%v' is set", binlogplayer.BlpFlagDontStart)
}
// wait for the endpoint set (usefull for the first run at least, fast for next runs)
if err := discovery.WaitForEndPoints(bpc.healthCheck, bpc.cell, bpc.sourceShard.Keyspace, bpc.sourceShard.Shard, []topodatapb.TabletType{topodatapb.TabletType_REPLICA}); err != nil {
return fmt.Errorf("error waiting for endpoints for %v %v %v: %v", bpc.cell, bpc.sourceShard.String(), topodatapb.TabletType_REPLICA, err)
}
// Find the server list from the health check
addrs := bpc.healthCheck.GetEndPointStatsFromTarget(bpc.sourceShard.Keyspace, bpc.sourceShard.Shard, topodatapb.TabletType_REPLICA)
if len(addrs) == 0 {
return fmt.Errorf("can't find any source tablet for %v %v %v", bpc.cell, bpc.sourceShard.String(), topodatapb.TabletType_REPLICA)
}
newServerIndex := rand.Intn(len(addrs))
endPoint := addrs[newServerIndex].EndPoint
// save our current server
bpc.playerMutex.Lock()
bpc.sourceTablet = &topodatapb.TabletAlias{
Cell: bpc.cell,
Uid: endPoint.Uid,
}
bpc.lastError = nil
bpc.playerMutex.Unlock()
// check which kind of replication we're doing, tables or keyrange
if len(bpc.sourceShard.Tables) > 0 {
// tables, first resolve wildcards
tables, err := mysqlctl.ResolveTables(bpc.mysqld, bpc.dbName, bpc.sourceShard.Tables)
if err != nil {
return fmt.Errorf("failed to resolve table names: %v", err)
}
// tables, just get them
player, err := binlogplayer.NewBinlogPlayerTables(vtClient, endPoint, tables, bpc.sourceShard.Uid, startPosition, bpc.stopPosition, bpc.binlogPlayerStats)
if err != nil {
return fmt.Errorf("NewBinlogPlayerTables failed: %v", err)
}
return player.ApplyBinlogEvents(bpc.ctx)
}
// the data we have to replicate is the intersection of the
// source keyrange and our keyrange
overlap, err := key.KeyRangesOverlap(bpc.sourceShard.KeyRange, bpc.keyRange)
if err != nil {
return fmt.Errorf("Source shard %v doesn't overlap destination shard %v", bpc.sourceShard.KeyRange, bpc.keyRange)
}
player, err := binlogplayer.NewBinlogPlayerKeyRange(vtClient, endPoint, overlap, bpc.sourceShard.Uid, startPosition, bpc.stopPosition, bpc.binlogPlayerStats)
if err != nil {
return fmt.Errorf("NewBinlogPlayerKeyRange failed: %v", err)
}
return player.ApplyBinlogEvents(bpc.ctx)
}
func (ta *TabletActor) multiRestore(actionNode *actionnode.ActionNode) (err error) {
args := actionNode.Args.(*actionnode.MultiRestoreArgs)
// read our current tablet, verify its state
// we only support restoring to the master or active replicas
tablet, err := ta.ts.GetTablet(ta.tabletAlias)
if err != nil {
return err
}
if tablet.Type != topo.TYPE_MASTER && !topo.IsSlaveType(tablet.Type) {
return fmt.Errorf("expected master, or slave type, not %v: %v", tablet.Type, ta.tabletAlias)
}
// get source tablets addresses
sourceAddrs := make([]*url.URL, len(args.SrcTabletAliases))
keyRanges := make([]key.KeyRange, len(args.SrcTabletAliases))
fromStoragePaths := make([]string, len(args.SrcTabletAliases))
for i, alias := range args.SrcTabletAliases {
t, e := ta.ts.GetTablet(alias)
if e != nil {
return e
}
sourceAddrs[i] = &url.URL{
Host: t.Addr(),
Path: "/" + t.DbName(),
}
keyRanges[i], e = key.KeyRangesOverlap(tablet.KeyRange, t.KeyRange)
if e != nil {
return e
}
fromStoragePaths[i] = path.Join(ta.mysqld.SnapshotDir, "from-storage", fmt.Sprintf("from-%v-%v", keyRanges[i].Start.Hex(), keyRanges[i].End.Hex()))
}
// change type to restore, no change to replication graph
originalType := tablet.Type
tablet.Type = topo.TYPE_RESTORE
err = topo.UpdateTablet(ta.ts, tablet)
if err != nil {
return err
}
// first try to get the data from a remote storage
wg := sync.WaitGroup{}
rec := concurrency.AllErrorRecorder{}
for i, alias := range args.SrcTabletAliases {
wg.Add(1)
go func(i int, alias topo.TabletAlias) {
defer wg.Done()
h := hook.NewSimpleHook("copy_snapshot_from_storage")
h.ExtraEnv = make(map[string]string)
for k, v := range ta.hookExtraEnv() {
h.ExtraEnv[k] = v
}
h.ExtraEnv["KEYRANGE"] = fmt.Sprintf("%v-%v", keyRanges[i].Start.Hex(), keyRanges[i].End.Hex())
h.ExtraEnv["SNAPSHOT_PATH"] = fromStoragePaths[i]
h.ExtraEnv["SOURCE_TABLET_ALIAS"] = alias.String()
hr := h.Execute()
if hr.ExitStatus != hook.HOOK_SUCCESS {
rec.RecordError(fmt.Errorf("%v hook failed(%v): %v", h.Name, hr.ExitStatus, hr.Stderr))
}
}(i, alias)
}
wg.Wait()
// run the action, scrap if it fails
if rec.HasErrors() {
log.Infof("Got errors trying to get snapshots from storage, trying to get them from original tablets: %v", rec.Error())
err = ta.mysqld.MultiRestore(tablet.DbName(), keyRanges, sourceAddrs, nil, args.Concurrency, args.FetchConcurrency, args.InsertTableConcurrency, args.FetchRetryCount, args.Strategy)
} else {
log.Infof("Got snapshots from storage, reading them from disk directly")
err = ta.mysqld.MultiRestore(tablet.DbName(), keyRanges, nil, fromStoragePaths, args.Concurrency, args.FetchConcurrency, args.InsertTableConcurrency, args.FetchRetryCount, args.Strategy)
}
if err != nil {
if e := topotools.Scrap(ta.ts, ta.tabletAlias, false); e != nil {
log.Errorf("Failed to Scrap after failed RestoreFromMultiSnapshot: %v", e)
}
return err
}
// restore type back
tablet.Type = originalType
return topo.UpdateTablet(ta.ts, tablet)
}
func (bpc *BinlogPlayerController) Iteration() (err error) {
defer func() {
if x := recover(); x != nil {
log.Errorf("%v: Caught panic: %v", bpc, x)
err = fmt.Errorf("panic: %v", x)
}
}()
// Enable any user to set the timestamp.
// We do it on every iteration to be sure, in case mysql was
// restarted.
bpc.DisableSuperToSetTimestamp()
// create the db connection, connect it
vtClient := mysqlctl.NewDbClient(bpc.dbConfig)
if err := vtClient.Connect(); err != nil {
return fmt.Errorf("can't connect to database: %v", err)
}
defer vtClient.Close()
// Read the start position
startPosition, err := mysqlctl.ReadStartPosition(vtClient, bpc.sourceShard.Uid)
if err != nil {
return fmt.Errorf("can't read startPosition: %v", err)
}
// Find the server list for the source shard in our cell
addrs, err := bpc.ts.GetEndPoints(bpc.cell, bpc.sourceShard.Keyspace, bpc.sourceShard.Shard, topo.TYPE_REPLICA)
if err != nil {
return fmt.Errorf("can't find any source tablet for %v %v %v: %v", bpc.cell, bpc.sourceShard.String(), topo.TYPE_REPLICA, err)
}
if len(addrs.Entries) == 0 {
return fmt.Errorf("empty source tablet list for %v %v %v", bpc.cell, bpc.sourceShard.String(), topo.TYPE_REPLICA)
}
// if the server we were using before is in the list, just keep using it
usePreviousServer := false
for _, addr := range addrs.Entries {
vtAddr := fmt.Sprintf("%v:%v", addr.Host, addr.NamedPortMap["_vtocc"])
if vtAddr == startPosition.Addr {
log.Infof("%v: Previous server %s still healthy, using it", bpc, vtAddr)
usePreviousServer = true
}
}
// if we can't use the previous server, pick a new one randomly
if !usePreviousServer {
newServerIndex := rand.Intn(len(addrs.Entries))
startPosition.Addr = fmt.Sprintf("%v:%v", addrs.Entries[newServerIndex].Host, addrs.Entries[newServerIndex].NamedPortMap["_vtocc"])
startPosition.Position.MasterFilename = ""
startPosition.Position.MasterPosition = 0
log.Infof("%v: Connecting to different server: %s", bpc, startPosition.Addr)
}
// the data we have to replicate is the intersection of the
// source keyrange and our keyrange
overlap, err := key.KeyRangesOverlap(bpc.sourceShard.KeyRange, bpc.keyRange)
if err != nil {
return fmt.Errorf("Source shard %v doesn't overlap destination shard %v", bpc.sourceShard.KeyRange, bpc.keyRange)
}
// Create the player.
player, err := mysqlctl.NewBinlogPlayer(vtClient, overlap, bpc.sourceShard.Uid, startPosition, nil /*tables*/, 1 /*txnBatch*/, 30*time.Second /*maxTxnInterval*/, false /*execDdl*/)
if err != nil {
return fmt.Errorf("can't create player: %v", err)
}
// Run player loop until it's done.
return player.ApplyBinlogEvents(bpc.interrupted)
}
// Iteration is a single iteration for the player: get the current status,
// try to play, and plays until interrupted, or until an error occurs.
func (bpc *BinlogPlayerController) Iteration() (err error) {
defer func() {
if x := recover(); x != nil {
log.Errorf("%v: caught panic: %v\n%s", bpc, x, tb.Stack(4))
err = fmt.Errorf("panic: %v", x)
}
}()
// Check if the context is still good.
select {
case <-bpc.ctx.Done():
if bpc.ctx.Err() == context.Canceled {
// We were stopped. Break out of Loop().
return nil
}
return fmt.Errorf("giving up since the context is done: %v", bpc.ctx.Err())
default:
}
// Apply any special settings necessary for playback of binlogs.
// We do it on every iteration to be sure, in case MySQL was restarted.
if err := bpc.mysqld.EnableBinlogPlayback(); err != nil {
// We failed to apply the required settings, so we shouldn't keep going.
return err
}
// create the db connection, connect it
vtClient := bpc.vtClientFactory()
if err := vtClient.Connect(); err != nil {
return fmt.Errorf("can't connect to database: %v", err)
}
defer vtClient.Close()
// Read the start position
startPosition, flags, err := binlogplayer.ReadStartPosition(vtClient, bpc.sourceShard.Uid)
if err != nil {
return fmt.Errorf("can't read startPosition: %v", err)
}
// if we shouldn't start, we just error out and try again later
if strings.Index(flags, binlogplayer.BlpFlagDontStart) != -1 {
return fmt.Errorf("not starting because flag '%v' is set", binlogplayer.BlpFlagDontStart)
}
// wait for the tablet set (usefull for the first run at least, fast for next runs)
if err := bpc.tabletStatsCache.WaitForTablets(bpc.ctx, bpc.cell, bpc.sourceShard.Keyspace, bpc.sourceShard.Shard, []topodatapb.TabletType{topodatapb.TabletType_REPLICA}); err != nil {
return fmt.Errorf("error waiting for tablets for %v %v %v: %v", bpc.cell, bpc.sourceShard.String(), topodatapb.TabletType_REPLICA, err)
}
// Find the server list from the health check.
// Note: We cannot use tsc.GetHealthyTabletStats() here because it does
// not return non-serving tablets. We must include non-serving tablets because
// REPLICA source tablets may not be serving anymore because their traffic was
// already migrated to the destination shards.
addrs := discovery.RemoveUnhealthyTablets(bpc.tabletStatsCache.GetTabletStats(bpc.sourceShard.Keyspace, bpc.sourceShard.Shard, topodatapb.TabletType_REPLICA))
if len(addrs) == 0 {
return fmt.Errorf("can't find any healthy source tablet for %v %v %v", bpc.cell, bpc.sourceShard.String(), topodatapb.TabletType_REPLICA)
}
newServerIndex := rand.Intn(len(addrs))
tablet := addrs[newServerIndex].Tablet
// save our current server
bpc.playerMutex.Lock()
bpc.sourceTablet = tablet.Alias
bpc.lastError = nil
bpc.playerMutex.Unlock()
// check which kind of replication we're doing, tables or keyrange
if len(bpc.sourceShard.Tables) > 0 {
// tables, first resolve wildcards
tables, err := mysqlctl.ResolveTables(bpc.mysqld, bpc.dbName, bpc.sourceShard.Tables)
if err != nil {
return fmt.Errorf("failed to resolve table names: %v", err)
}
// tables, just get them
player, err := binlogplayer.NewBinlogPlayerTables(vtClient, tablet, tables, bpc.sourceShard.Uid, startPosition, bpc.stopPosition, bpc.binlogPlayerStats)
if err != nil {
return fmt.Errorf("NewBinlogPlayerTables failed: %v", err)
}
return player.ApplyBinlogEvents(bpc.ctx)
}
// the data we have to replicate is the intersection of the
// source keyrange and our keyrange
overlap, err := key.KeyRangesOverlap(bpc.sourceShard.KeyRange, bpc.keyRange)
if err != nil {
return fmt.Errorf("Source shard %v doesn't overlap destination shard %v", bpc.sourceShard.KeyRange, bpc.keyRange)
}
player, err := binlogplayer.NewBinlogPlayerKeyRange(vtClient, tablet, overlap, bpc.sourceShard.Uid, startPosition, bpc.stopPosition, bpc.binlogPlayerStats)
if err != nil {
return fmt.Errorf("NewBinlogPlayerKeyRange failed: %v", err)
}
return player.ApplyBinlogEvents(bpc.ctx)
}
// Iteration is a single iteration for the player: get the current status,
// try to play, and plays until interrupted, or until an error occurs.
func (bpc *BinlogPlayerController) Iteration(ctx context.Context) (err error) {
defer func() {
if x := recover(); x != nil {
log.Errorf("%v: caught panic: %v", bpc, x)
err = fmt.Errorf("panic: %v", x)
}
}()
// Apply any special settings necessary for playback of binlogs.
// We do it on every iteration to be sure, in case MySQL was restarted.
if err := bpc.mysqld.EnableBinlogPlayback(); err != nil {
// We failed to apply the required settings, so we shouldn't keep going.
return err
}
// create the db connection, connect it
vtClient := binlogplayer.NewDbClient(bpc.dbConfig)
if err := vtClient.Connect(); err != nil {
return fmt.Errorf("can't connect to database: %v", err)
}
defer vtClient.Close()
// Read the start position
startPosition, flags, err := binlogplayer.ReadStartPosition(vtClient, bpc.sourceShard.Uid)
if err != nil {
return fmt.Errorf("can't read startPosition: %v", err)
}
// if we shouldn't start, we just error out and try again later
if strings.Index(flags, binlogplayer.BlpFlagDontStart) != -1 {
return fmt.Errorf("not starting because flag '%v' is set", binlogplayer.BlpFlagDontStart)
}
// Find the server list for the source shard in our cell
addrs, _, err := bpc.ts.GetEndPoints(ctx, bpc.cell, bpc.sourceShard.Keyspace, bpc.sourceShard.Shard, topo.TYPE_REPLICA)
if err != nil {
return fmt.Errorf("can't find any source tablet for %v %v %v: %v", bpc.cell, bpc.sourceShard.String(), topo.TYPE_REPLICA, err)
}
if len(addrs.Entries) == 0 {
return fmt.Errorf("empty source tablet list for %v %v %v", bpc.cell, bpc.sourceShard.String(), topo.TYPE_REPLICA)
}
newServerIndex := rand.Intn(len(addrs.Entries))
port, _ := addrs.Entries[newServerIndex].NamedPortMap["vt"]
addr := netutil.JoinHostPort(addrs.Entries[newServerIndex].Host, port)
// save our current server
bpc.playerMutex.Lock()
bpc.sourceTablet = topo.TabletAlias{
Cell: bpc.cell,
Uid: addrs.Entries[newServerIndex].Uid,
}
bpc.lastError = nil
bpc.playerMutex.Unlock()
// check which kind of replication we're doing, tables or keyrange
if len(bpc.sourceShard.Tables) > 0 {
// tables, first resolve wildcards
tables, err := mysqlctl.ResolveTables(bpc.mysqld, bpc.dbName, bpc.sourceShard.Tables)
if err != nil {
return fmt.Errorf("failed to resolve table names: %v", err)
}
// tables, just get them
player := binlogplayer.NewBinlogPlayerTables(vtClient, addr, tables, startPosition, bpc.stopPosition, bpc.binlogPlayerStats)
return player.ApplyBinlogEvents(bpc.interrupted)
}
// the data we have to replicate is the intersection of the
// source keyrange and our keyrange
overlap, err := key.KeyRangesOverlap(bpc.sourceShard.KeyRange, bpc.keyRange)
if err != nil {
return fmt.Errorf("Source shard %v doesn't overlap destination shard %v", bpc.sourceShard.KeyRange, bpc.keyRange)
}
player := binlogplayer.NewBinlogPlayerKeyRange(vtClient, addr, bpc.keyspaceIDType, overlap, startPosition, bpc.stopPosition, bpc.binlogPlayerStats)
return player.ApplyBinlogEvents(bpc.interrupted)
}
func (sdw *SplitDiffWorker) diff(ctx context.Context) error {
sdw.SetState(WorkerStateDiff)
sdw.wr.Logger().Infof("Gathering schema information...")
wg := sync.WaitGroup{}
rec := &concurrency.AllErrorRecorder{}
wg.Add(1)
go func() {
var err error
shortCtx, cancel := context.WithTimeout(ctx, *remoteActionsTimeout)
sdw.destinationSchemaDefinition, err = sdw.wr.GetSchema(
shortCtx, sdw.destinationAlias, nil /* tables */, sdw.excludeTables, false /* includeViews */)
cancel()
rec.RecordError(err)
sdw.wr.Logger().Infof("Got schema from destination %v", sdw.destinationAlias)
wg.Done()
}()
wg.Add(1)
go func() {
var err error
shortCtx, cancel := context.WithTimeout(ctx, *remoteActionsTimeout)
sdw.sourceSchemaDefinition, err = sdw.wr.GetSchema(
shortCtx, sdw.sourceAlias, nil /* tables */, sdw.excludeTables, false /* includeViews */)
cancel()
rec.RecordError(err)
sdw.wr.Logger().Infof("Got schema from source %v", sdw.sourceAlias)
wg.Done()
}()
wg.Wait()
if rec.HasErrors() {
return rec.Error()
}
sdw.wr.Logger().Infof("Diffing the schema...")
rec = &concurrency.AllErrorRecorder{}
tmutils.DiffSchema("destination", sdw.destinationSchemaDefinition, "source", sdw.sourceSchemaDefinition, rec)
if rec.HasErrors() {
sdw.wr.Logger().Warningf("Different schemas: %v", rec.Error().Error())
} else {
sdw.wr.Logger().Infof("Schema match, good.")
}
// read the vschema if needed
var keyspaceSchema *vindexes.KeyspaceSchema
if *useV3ReshardingMode {
kschema, err := sdw.wr.TopoServer().GetVSchema(ctx, sdw.keyspace)
if err != nil {
return fmt.Errorf("cannot load VSchema for keyspace %v: %v", sdw.keyspace, err)
}
if kschema == nil {
return fmt.Errorf("no VSchema for keyspace %v", sdw.keyspace)
}
keyspaceSchema, err = vindexes.BuildKeyspaceSchema(kschema, sdw.keyspace)
if err != nil {
return fmt.Errorf("cannot build vschema for keyspace %v: %v", sdw.keyspace, err)
}
}
// Compute the overlap keyrange. Later, we'll compare it with
// source or destination keyrange. If it matches either,
// we'll just ask for all the data. If the overlap is a subset,
// we'll filter.
overlap, err := key.KeyRangesOverlap(sdw.shardInfo.KeyRange, sdw.shardInfo.SourceShards[sdw.sourceUID].KeyRange)
if err != nil {
return fmt.Errorf("Source shard doesn't overlap with destination: %v", err)
}
// run the diffs, 8 at a time
sdw.wr.Logger().Infof("Running the diffs...")
// TODO(mberlin): Parameterize the hard coded value 8.
sem := sync2.NewSemaphore(8, 0)
for _, tableDefinition := range sdw.destinationSchemaDefinition.TableDefinitions {
wg.Add(1)
go func(tableDefinition *tabletmanagerdatapb.TableDefinition) {
defer wg.Done()
sem.Acquire()
defer sem.Release()
sdw.wr.Logger().Infof("Starting the diff on table %v", tableDefinition.Name)
// On the source, see if we need a full scan
// or a filtered scan.
var sourceQueryResultReader *QueryResultReader
if key.KeyRangeEqual(overlap, sdw.shardInfo.SourceShards[sdw.sourceUID].KeyRange) {
sourceQueryResultReader, err = TableScan(ctx, sdw.wr.Logger(), sdw.wr.TopoServer(), sdw.sourceAlias, tableDefinition)
} else {
sourceQueryResultReader, err = TableScanByKeyRange(ctx, sdw.wr.Logger(), sdw.wr.TopoServer(), sdw.sourceAlias, tableDefinition, overlap, keyspaceSchema, sdw.keyspaceInfo.ShardingColumnName, sdw.keyspaceInfo.ShardingColumnType)
}
if err != nil {
newErr := fmt.Errorf("TableScan(ByKeyRange?)(source) failed: %v", err)
rec.RecordError(newErr)
sdw.wr.Logger().Errorf("%v", newErr)
return
}
defer sourceQueryResultReader.Close(ctx)
// On the destination, see if we need a full scan
// or a filtered scan.
var destinationQueryResultReader *QueryResultReader
if key.KeyRangeEqual(overlap, sdw.shardInfo.KeyRange) {
destinationQueryResultReader, err = TableScan(ctx, sdw.wr.Logger(), sdw.wr.TopoServer(), sdw.destinationAlias, tableDefinition)
} else {
destinationQueryResultReader, err = TableScanByKeyRange(ctx, sdw.wr.Logger(), sdw.wr.TopoServer(), sdw.destinationAlias, tableDefinition, overlap, keyspaceSchema, sdw.keyspaceInfo.ShardingColumnName, sdw.keyspaceInfo.ShardingColumnType)
}
if err != nil {
newErr := fmt.Errorf("TableScan(ByKeyRange?)(destination) failed: %v", err)
rec.RecordError(newErr)
sdw.wr.Logger().Errorf("%v", newErr)
return
}
defer destinationQueryResultReader.Close(ctx)
// Create the row differ.
differ, err := NewRowDiffer(sourceQueryResultReader, destinationQueryResultReader, tableDefinition)
if err != nil {
newErr := fmt.Errorf("NewRowDiffer() failed: %v", err)
rec.RecordError(newErr)
sdw.wr.Logger().Errorf("%v", newErr)
return
}
// And run the diff.
report, err := differ.Go(sdw.wr.Logger())
if err != nil {
newErr := fmt.Errorf("Differ.Go failed: %v", err.Error())
rec.RecordError(newErr)
sdw.wr.Logger().Errorf("%v", newErr)
} else {
if report.HasDifferences() {
err := fmt.Errorf("Table %v has differences: %v", tableDefinition.Name, report.String())
rec.RecordError(err)
sdw.wr.Logger().Warningf(err.Error())
} else {
sdw.wr.Logger().Infof("Table %v checks out (%v rows processed, %v qps)", tableDefinition.Name, report.processedRows, report.processingQPS)
}
}
}(tableDefinition)
}
wg.Wait()
return rec.Error()
}
// Iteration is a single iteration for the player: get the current status,
// try to play, and plays until interrupted, or until an error occurs.
func (bpc *BinlogPlayerController) Iteration() (err error) {
defer func() {
if x := recover(); x != nil {
log.Errorf("%v: caught panic: %v", bpc, x)
err = fmt.Errorf("panic: %v", x)
}
}()
// Enable any user to set the timestamp.
// We do it on every iteration to be sure, in case mysql was
// restarted.
bpc.DisableSuperToSetTimestamp()
// create the db connection, connect it
vtClient := binlogplayer.NewDbClient(bpc.dbConfig)
if err := vtClient.Connect(); err != nil {
return fmt.Errorf("can't connect to database: %v", err)
}
defer vtClient.Close()
// Read the start position
startPosition, flags, err := binlogplayer.ReadStartPosition(vtClient, bpc.sourceShard.Uid)
if err != nil {
return fmt.Errorf("can't read startPosition: %v", err)
}
// if we shouldn't start, we just error out and try again later
if strings.Index(flags, binlogplayer.BLP_FLAG_DONT_START) != -1 {
return fmt.Errorf("not starting because flag '%v' is set", binlogplayer.BLP_FLAG_DONT_START)
}
// Find the server list for the source shard in our cell
addrs, err := bpc.ts.GetEndPoints(bpc.cell, bpc.sourceShard.Keyspace, bpc.sourceShard.Shard, topo.TYPE_REPLICA)
if err != nil {
return fmt.Errorf("can't find any source tablet for %v %v %v: %v", bpc.cell, bpc.sourceShard.String(), topo.TYPE_REPLICA, err)
}
if len(addrs.Entries) == 0 {
return fmt.Errorf("empty source tablet list for %v %v %v", bpc.cell, bpc.sourceShard.String(), topo.TYPE_REPLICA)
}
newServerIndex := rand.Intn(len(addrs.Entries))
addr := fmt.Sprintf("%v:%v", addrs.Entries[newServerIndex].Host, addrs.Entries[newServerIndex].NamedPortMap["_vtocc"])
// save our current server
bpc.playerMutex.Lock()
bpc.sourceTablet = topo.TabletAlias{
Cell: bpc.cell,
Uid: addrs.Entries[newServerIndex].Uid,
}
bpc.lastError = nil
bpc.playerMutex.Unlock()
// check which kind of replication we're doing, tables or keyrange
if len(bpc.sourceShard.Tables) > 0 {
// tables, first resolve wildcards
tables, err := bpc.mysqld.ResolveTables(bpc.dbName, bpc.sourceShard.Tables)
if err != nil {
return fmt.Errorf("failed to resolve table names: %v", err)
}
// tables, just get them
player := binlogplayer.NewBinlogPlayerTables(vtClient, addr, tables, startPosition, bpc.stopAtGTID, bpc.binlogPlayerStats)
return player.ApplyBinlogEvents(bpc.interrupted)
} else {
// the data we have to replicate is the intersection of the
// source keyrange and our keyrange
overlap, err := key.KeyRangesOverlap(bpc.sourceShard.KeyRange, bpc.keyRange)
if err != nil {
return fmt.Errorf("Source shard %v doesn't overlap destination shard %v", bpc.sourceShard.KeyRange, bpc.keyRange)
}
player := binlogplayer.NewBinlogPlayerKeyRange(vtClient, addr, bpc.keyspaceIdType, overlap, startPosition, bpc.stopAtGTID, bpc.binlogPlayerStats)
return player.ApplyBinlogEvents(bpc.interrupted)
}
}