// RunUntil will run all the players until they reach the given position.
// Holds the map lock during that exercise, shouldn't take long at all.
func (blm *BinlogPlayerMap) RunUntil(ctx context.Context, blpPositionList []*tabletmanagerdatapb.BlpPosition, waitTimeout time.Duration) error {
// lock and check state
blm.mu.Lock()
defer blm.mu.Unlock()
if blm.state != BpmStateStopped {
return fmt.Errorf("RunUntil: player not stopped: %v", blm.state)
}
log.Infof("Starting map of binlog players until position")
// find the exact stop position for all players, to be sure
// we're not doing anything wrong
posMap := make(map[uint32]string)
for _, bpc := range blm.players {
blpPos := tmutils.FindBlpPositionByID(blpPositionList, bpc.sourceShard.Uid)
if blpPos == nil {
return fmt.Errorf("No binlog position passed in for player Uid %v", bpc.sourceShard.Uid)
}
posMap[bpc.sourceShard.Uid] = blpPos.Position
}
// start all the players giving them where to stop
for _, bpc := range blm.players {
if err := bpc.StartUntil(ctx, posMap[bpc.sourceShard.Uid]); err != nil {
return err
}
}
// wait for all players to be stopped, or timeout
wg := sync.WaitGroup{}
rec := concurrency.AllErrorRecorder{}
for _, bpc := range blm.players {
wg.Add(1)
go func(bpc *BinlogPlayerController) {
if err := bpc.WaitForStop(waitTimeout); err != nil {
rec.RecordError(err)
}
wg.Done()
}(bpc)
}
wg.Wait()
return rec.Error()
}
func (sdw *SplitDiffWorker) synchronizeReplication(ctx context.Context) error {
sdw.SetState(WorkerStateSyncReplication)
shortCtx, cancel := context.WithTimeout(ctx, *remoteActionsTimeout)
masterInfo, err := sdw.wr.TopoServer().GetTablet(shortCtx, sdw.shardInfo.MasterAlias)
cancel()
if err != nil {
return fmt.Errorf("synchronizeReplication: cannot get Tablet record for master %v: %v", sdw.shardInfo.MasterAlias, err)
}
// 1 - stop the master binlog replication, get its current position
sdw.wr.Logger().Infof("Stopping master binlog replication on %v", sdw.shardInfo.MasterAlias)
shortCtx, cancel = context.WithTimeout(ctx, *remoteActionsTimeout)
blpPositionList, err := sdw.wr.TabletManagerClient().StopBlp(shortCtx, masterInfo)
cancel()
if err != nil {
return fmt.Errorf("StopBlp for %v failed: %v", sdw.shardInfo.MasterAlias, err)
}
wrangler.RecordStartBlpAction(sdw.cleaner, masterInfo)
// 2 - stop all the source tablets at a binlog position
// higher than the destination master
stopPositionList := make([]*pbt.BlpPosition, len(sdw.shardInfo.SourceShards))
for i, ss := range sdw.shardInfo.SourceShards {
// find where we should be stopping
blpPos := tmutils.FindBlpPositionByID(blpPositionList, ss.Uid)
if blpPos == nil {
return fmt.Errorf("no binlog position on the master for Uid %v", ss.Uid)
}
// read the tablet
shortCtx, cancel := context.WithTimeout(ctx, *remoteActionsTimeout)
sourceTablet, err := sdw.wr.TopoServer().GetTablet(shortCtx, sdw.sourceAliases[i])
cancel()
if err != nil {
return err
}
// stop replication
sdw.wr.Logger().Infof("Stopping slave[%v] %v at a minimum of %v", i, sdw.sourceAliases[i], blpPos.Position)
shortCtx, cancel = context.WithTimeout(ctx, *remoteActionsTimeout)
stoppedAt, err := sdw.wr.TabletManagerClient().StopSlaveMinimum(shortCtx, sourceTablet, blpPos.Position, *remoteActionsTimeout)
cancel()
if err != nil {
return fmt.Errorf("cannot stop slave %v at right binlog position %v: %v", sdw.sourceAliases[i], blpPos.Position, err)
}
stopPositionList[i] = &pbt.BlpPosition{
Uid: ss.Uid,
Position: stoppedAt,
}
// change the cleaner actions from ChangeSlaveType(rdonly)
// to StartSlave() + ChangeSlaveType(spare)
wrangler.RecordStartSlaveAction(sdw.cleaner, sourceTablet)
action, err := wrangler.FindChangeSlaveTypeActionByTarget(sdw.cleaner, sdw.sourceAliases[i])
if err != nil {
return fmt.Errorf("cannot find ChangeSlaveType action for %v: %v", sdw.sourceAliases[i], err)
}
action.TabletType = pb.TabletType_SPARE
}
// 3 - ask the master of the destination shard to resume filtered
// replication up to the new list of positions
sdw.wr.Logger().Infof("Restarting master %v until it catches up to %v", sdw.shardInfo.MasterAlias, stopPositionList)
shortCtx, cancel = context.WithTimeout(ctx, *remoteActionsTimeout)
masterPos, err := sdw.wr.TabletManagerClient().RunBlpUntil(shortCtx, masterInfo, stopPositionList, *remoteActionsTimeout)
cancel()
if err != nil {
return fmt.Errorf("RunBlpUntil for %v until %v failed: %v", sdw.shardInfo.MasterAlias, stopPositionList, err)
}
// 4 - wait until the destination tablet is equal or passed
// that master binlog position, and stop its replication.
sdw.wr.Logger().Infof("Waiting for destination tablet %v to catch up to %v", sdw.destinationAlias, masterPos)
shortCtx, cancel = context.WithTimeout(ctx, *remoteActionsTimeout)
destinationTablet, err := sdw.wr.TopoServer().GetTablet(shortCtx, sdw.destinationAlias)
cancel()
if err != nil {
return err
}
shortCtx, cancel = context.WithTimeout(ctx, *remoteActionsTimeout)
_, err = sdw.wr.TabletManagerClient().StopSlaveMinimum(shortCtx, destinationTablet, masterPos, *remoteActionsTimeout)
cancel()
if err != nil {
return fmt.Errorf("StopSlaveMinimum for %v at %v failed: %v", sdw.destinationAlias, masterPos, err)
}
wrangler.RecordStartSlaveAction(sdw.cleaner, destinationTablet)
action, err := wrangler.FindChangeSlaveTypeActionByTarget(sdw.cleaner, sdw.destinationAlias)
if err != nil {
return fmt.Errorf("cannot find ChangeSlaveType action for %v: %v", sdw.destinationAlias, err)
}
action.TabletType = pb.TabletType_SPARE
// 5 - restart filtered replication on destination master
sdw.wr.Logger().Infof("Restarting filtered replication on master %v", sdw.shardInfo.MasterAlias)
shortCtx, cancel = context.WithTimeout(ctx, *remoteActionsTimeout)
err = sdw.wr.TabletManagerClient().StartBlp(shortCtx, masterInfo)
if err := sdw.cleaner.RemoveActionByName(wrangler.StartBlpActionName, topoproto.TabletAliasString(sdw.shardInfo.MasterAlias)); err != nil {
sdw.wr.Logger().Warningf("Cannot find cleaning action %v/%v: %v", wrangler.StartBlpActionName, topoproto.TabletAliasString(sdw.shardInfo.MasterAlias), err)
}
cancel()
if err != nil {
return fmt.Errorf("StartBlp failed for %v: %v", sdw.shardInfo.MasterAlias, err)
}
return nil
}