// synchronizeReplication phase:
// 1 - ask the subset slave to stop replication
// 2 - sleep for 5 seconds
// 3 - ask the superset slave to stop replication
// Note this is not 100% correct, but good enough for now
func (worker *SQLDiffWorker) synchronizeReplication() error {
worker.setState(SQLDiffSynchronizeReplication)
// stop replication on subset slave
worker.wr.Logger().Infof("Stopping replication on subset slave %v", worker.subset.alias)
subsetTablet, err := worker.wr.TopoServer().GetTablet(worker.subset.alias)
if err != nil {
return err
}
ctx, cancel := context.WithTimeout(context.TODO(), 30*time.Second)
if err := worker.wr.TabletManagerClient().StopSlave(ctx, subsetTablet); err != nil {
return fmt.Errorf("Cannot stop slave %v: %v", worker.subset.alias, err)
}
cancel()
if worker.CheckInterrupted() {
return topo.ErrInterrupted
}
// change the cleaner actions from ChangeSlaveType(rdonly)
// to StartSlave() + ChangeSlaveType(spare)
wrangler.RecordStartSlaveAction(worker.cleaner, subsetTablet, 30*time.Second)
action, err := wrangler.FindChangeSlaveTypeActionByTarget(worker.cleaner, worker.subset.alias)
if err != nil {
return fmt.Errorf("cannot find ChangeSlaveType action for %v: %v", worker.subset.alias, err)
}
action.TabletType = topo.TYPE_SPARE
// sleep for a few seconds
time.Sleep(5 * time.Second)
if worker.CheckInterrupted() {
return topo.ErrInterrupted
}
// stop replication on superset slave
worker.wr.Logger().Infof("Stopping replication on superset slave %v", worker.superset.alias)
supersetTablet, err := worker.wr.TopoServer().GetTablet(worker.superset.alias)
if err != nil {
return err
}
ctx, cancel = context.WithTimeout(context.TODO(), 30*time.Second)
if err := worker.wr.TabletManagerClient().StopSlave(ctx, supersetTablet); err != nil {
return fmt.Errorf("Cannot stop slave %v: %v", worker.superset.alias, err)
}
cancel()
// change the cleaner actions from ChangeSlaveType(rdonly)
// to StartSlave() + ChangeSlaveType(spare)
wrangler.RecordStartSlaveAction(worker.cleaner, supersetTablet, 30*time.Second)
action, err = wrangler.FindChangeSlaveTypeActionByTarget(worker.cleaner, worker.superset.alias)
if err != nil {
return fmt.Errorf("cannot find ChangeSlaveType action for %v: %v", worker.superset.alias, err)
}
action.TabletType = topo.TYPE_SPARE
return nil
}
// findTargets phase:
// - find one rdonly in the source shard
// - mark it as 'checker' pointing back to us
// - get the aliases of all the targets
func (vscw *VerticalSplitCloneWorker) findTargets() error {
vscw.setState(stateVSCFindTargets)
// find an appropriate endpoint in the source shard
var err error
vscw.sourceAlias, err = findChecker(vscw.wr, vscw.cleaner, vscw.cell, vscw.sourceKeyspace, "0")
if err != nil {
return fmt.Errorf("cannot find checker for %v/%v/0: %v", vscw.cell, vscw.sourceKeyspace, err)
}
vscw.wr.Logger().Infof("Using tablet %v as the source", vscw.sourceAlias)
// get the tablet info for it
vscw.sourceTablet, err = vscw.wr.TopoServer().GetTablet(vscw.sourceAlias)
if err != nil {
return fmt.Errorf("cannot read tablet %v: %v", vscw.sourceTablet, err)
}
// stop replication on it
ctx, cancel := context.WithTimeout(context.TODO(), 30*time.Second)
if err := vscw.wr.TabletManagerClient().StopSlave(ctx, vscw.sourceTablet); err != nil {
return fmt.Errorf("cannot stop replication on tablet %v", vscw.sourceAlias)
}
cancel()
wrangler.RecordStartSlaveAction(vscw.cleaner, vscw.sourceTablet, 30*time.Second)
action, err := wrangler.FindChangeSlaveTypeActionByTarget(vscw.cleaner, vscw.sourceAlias)
if err != nil {
return fmt.Errorf("cannot find ChangeSlaveType action for %v: %v", vscw.sourceAlias, err)
}
action.TabletType = topo.TYPE_SPARE
return vscw.findMasterTargets()
}
// findTargets phase:
// - find one rdonly in the source shard
// - mark it as 'checker' pointing back to us
// - get the aliases of all the targets
func (scw *SplitCloneWorker) findTargets() error {
scw.setState(stateSCFindTargets)
var err error
// find an appropriate endpoint in the source shards
scw.sourceAliases = make([]topo.TabletAlias, len(scw.sourceShards))
for i, si := range scw.sourceShards {
scw.sourceAliases[i], err = findChecker(scw.wr, scw.cleaner, scw.cell, si.Keyspace(), si.ShardName())
if err != nil {
return fmt.Errorf("cannot find checker for %v/%v/%v: %v", scw.cell, si.Keyspace(), si.ShardName(), err)
}
scw.wr.Logger().Infof("Using tablet %v as source for %v/%v", scw.sourceAliases[i], si.Keyspace(), si.ShardName())
}
// get the tablet info for them, and stop their replication
scw.sourceTablets = make([]*topo.TabletInfo, len(scw.sourceAliases))
for i, alias := range scw.sourceAliases {
scw.sourceTablets[i], err = scw.wr.TopoServer().GetTablet(alias)
if err != nil {
return fmt.Errorf("cannot read tablet %v: %v", alias, err)
}
ctx, cancel := context.WithTimeout(context.TODO(), 30*time.Second)
if err := scw.wr.TabletManagerClient().StopSlave(ctx, scw.sourceTablets[i]); err != nil {
return fmt.Errorf("cannot stop replication on tablet %v", alias)
}
cancel()
wrangler.RecordStartSlaveAction(scw.cleaner, scw.sourceTablets[i], 30*time.Second)
action, err := wrangler.FindChangeSlaveTypeActionByTarget(scw.cleaner, alias)
if err != nil {
return fmt.Errorf("cannot find ChangeSlaveType action for %v: %v", alias, err)
}
action.TabletType = topo.TYPE_SPARE
}
return scw.findMasterTargets()
}
func (vsdw *VerticalSplitDiffWorker) synchronizeReplication() error {
vsdw.setState(stateVSDSynchronizeReplication)
masterInfo, err := vsdw.wr.TopoServer().GetTablet(vsdw.shardInfo.MasterAlias)
if err != nil {
return fmt.Errorf("synchronizeReplication: cannot get Tablet record for master %v: %v", vsdw.shardInfo.MasterAlias, err)
}
// 1 - stop the master binlog replication, get its current position
vsdw.wr.Logger().Infof("Stopping master binlog replication on %v", vsdw.shardInfo.MasterAlias)
ctx, cancel := context.WithTimeout(context.TODO(), 30*time.Second)
blpPositionList, err := vsdw.wr.TabletManagerClient().StopBlp(ctx, masterInfo)
if err != nil {
return fmt.Errorf("StopBlp on master %v failed: %v", vsdw.shardInfo.MasterAlias, err)
}
cancel()
wrangler.RecordStartBlpAction(vsdw.cleaner, masterInfo, 30*time.Second)
// 2 - stop the source 'checker' at a binlog position
// higher than the destination master
stopPositionList := blproto.BlpPositionList{
Entries: make([]blproto.BlpPosition, 1),
}
ss := vsdw.shardInfo.SourceShards[0]
// find where we should be stopping
pos, err := blpPositionList.FindBlpPositionById(ss.Uid)
if err != nil {
return fmt.Errorf("no binlog position on the master for Uid %v", ss.Uid)
}
// stop replication
vsdw.wr.Logger().Infof("Stopping slave %v at a minimum of %v", vsdw.sourceAlias, pos.Position)
sourceTablet, err := vsdw.wr.TopoServer().GetTablet(vsdw.sourceAlias)
if err != nil {
return err
}
stoppedAt, err := vsdw.wr.TabletManagerClient().StopSlaveMinimum(context.TODO(), sourceTablet, pos.Position, 30*time.Second)
if err != nil {
return fmt.Errorf("cannot stop slave %v at right binlog position %v: %v", vsdw.sourceAlias, pos.Position, err)
}
stopPositionList.Entries[0].Uid = ss.Uid
stopPositionList.Entries[0].Position = stoppedAt.Position
// change the cleaner actions from ChangeSlaveType(rdonly)
// to StartSlave() + ChangeSlaveType(spare)
wrangler.RecordStartSlaveAction(vsdw.cleaner, sourceTablet, 30*time.Second)
action, err := wrangler.FindChangeSlaveTypeActionByTarget(vsdw.cleaner, vsdw.sourceAlias)
if err != nil {
return fmt.Errorf("cannot find ChangeSlaveType action for %v: %v", vsdw.sourceAlias, err)
}
action.TabletType = topo.TYPE_SPARE
// 3 - ask the master of the destination shard to resume filtered
// replication up to the new list of positions
vsdw.wr.Logger().Infof("Restarting master %v until it catches up to %v", vsdw.shardInfo.MasterAlias, stopPositionList)
masterPos, err := vsdw.wr.TabletManagerClient().RunBlpUntil(context.TODO(), masterInfo, &stopPositionList, 30*time.Second)
if err != nil {
return fmt.Errorf("RunBlpUntil on %v until %v failed: %v", vsdw.shardInfo.MasterAlias, stopPositionList, err)
}
// 4 - wait until the destination checker is equal or passed
// that master binlog position, and stop its replication.
vsdw.wr.Logger().Infof("Waiting for destination checker %v to catch up to %v", vsdw.destinationAlias, masterPos)
destinationTablet, err := vsdw.wr.TopoServer().GetTablet(vsdw.destinationAlias)
if err != nil {
return err
}
_, err = vsdw.wr.TabletManagerClient().StopSlaveMinimum(context.TODO(), destinationTablet, masterPos, 30*time.Second)
if err != nil {
return fmt.Errorf("StopSlaveMinimum on %v at %v failed: %v", vsdw.destinationAlias, masterPos, err)
}
wrangler.RecordStartSlaveAction(vsdw.cleaner, destinationTablet, 30*time.Second)
action, err = wrangler.FindChangeSlaveTypeActionByTarget(vsdw.cleaner, vsdw.destinationAlias)
if err != nil {
return fmt.Errorf("cannot find ChangeSlaveType action for %v: %v", vsdw.destinationAlias, err)
}
action.TabletType = topo.TYPE_SPARE
// 5 - restart filtered replication on destination master
vsdw.wr.Logger().Infof("Restarting filtered replication on master %v", vsdw.shardInfo.MasterAlias)
ctx, cancel = context.WithTimeout(context.TODO(), 30*time.Second)
err = vsdw.wr.TabletManagerClient().StartBlp(ctx, masterInfo)
if err := vsdw.cleaner.RemoveActionByName(wrangler.StartBlpActionName, vsdw.shardInfo.MasterAlias.String()); err != nil {
vsdw.wr.Logger().Warningf("Cannot find cleaning action %v/%v: %v", wrangler.StartBlpActionName, vsdw.shardInfo.MasterAlias.String(), err)
}
cancel()
if err != nil {
return fmt.Errorf("StartBlp on %v failed: %v", vsdw.shardInfo.MasterAlias, err)
}
return nil
}