// ChangeType changes the type of the tablet and possibly also updates
// the health informaton for it. Make this external, since these
// transitions need to be forced from time to time.
//
// - if health is nil, we don't touch the Tablet's Health record.
// - if health is an empty map, we clear the Tablet's Health record.
// - if health has values, we overwrite the Tablet's Health record.
func ChangeType(ctx context.Context, ts topo.Server, tabletAlias *pb.TabletAlias, newType pb.TabletType, health map[string]string) error {
tablet, err := ts.GetTablet(ctx, tabletAlias)
if err != nil {
return err
}
if !topo.IsTrivialTypeChange(tablet.Type, newType) {
return fmt.Errorf("cannot change tablet type %v -> %v %v", tablet.Type, newType, tabletAlias)
}
tablet.Type = newType
if newType == pb.TabletType_IDLE {
tablet.Keyspace = ""
tablet.Shard = ""
tablet.KeyRange = nil
tablet.HealthMap = health
}
if health != nil {
if len(health) == 0 {
tablet.HealthMap = nil
} else {
tablet.HealthMap = health
}
}
return ts.UpdateTablet(ctx, tablet)
}
// CopyTablets will create the tablets in the destination topo
func CopyTablets(fromTS, toTS topo.Server) {
cells, err := fromTS.GetKnownCells()
if err != nil {
relog.Fatal("fromTS.GetKnownCells failed: %v", err)
}
wg := sync.WaitGroup{}
rec := concurrency.AllErrorRecorder{}
for _, cell := range cells {
wg.Add(1)
go func(cell string) {
defer wg.Done()
tabletAliases, err := fromTS.GetTabletsByCell(cell)
if err != nil {
rec.RecordError(err)
} else {
for _, tabletAlias := range tabletAliases {
wg.Add(1)
go func(tabletAlias topo.TabletAlias) {
defer wg.Done()
// read the source tablet
ti, err := fromTS.GetTablet(tabletAlias)
if err != nil {
rec.RecordError(err)
return
}
// try to create the destination
err = toTS.CreateTablet(ti.Tablet)
if err == topo.ErrNodeExists {
// update the destination tablet
_, err = toTS.UpdateTablet(ti, -1)
}
if err != nil {
rec.RecordError(err)
return
}
// create the replication paths
// for masters only here
if ti.Type == topo.TYPE_MASTER {
if err = toTS.CreateReplicationPath(ti.Keyspace, ti.Shard, ti.Alias().String()); err != nil && err != topo.ErrNodeExists {
rec.RecordError(err)
}
}
}(tabletAlias)
}
}
}(cell)
}
wg.Wait()
if rec.HasErrors() {
relog.Fatal("copyTablets failed: %v", rec.Error())
}
}
// Scrap will update the tablet type to 'Scrap', and remove it from
// the serving graph.
//
// 'force' means we are not on the tablet being scrapped, so it is
// probably dead. So if 'force' is true, we will also remove pending
// remote actions. And if 'force' is false, we also run an optional
// hook.
func Scrap(ctx context.Context, ts topo.Server, tabletAlias *pb.TabletAlias, force bool) error {
tablet, err := ts.GetTablet(ctx, tabletAlias)
if err != nil {
return err
}
// If you are already scrap, skip updating replication data. It won't
// be there anyway.
wasAssigned := tablet.IsAssigned()
tablet.Type = pb.TabletType_SCRAP
// Update the tablet first, since that is canonical.
err = ts.UpdateTablet(ctx, tablet)
if err != nil {
return err
}
if wasAssigned {
err = topo.DeleteTabletReplicationData(ctx, ts, tablet.Tablet)
if err != nil {
if err == topo.ErrNoNode {
log.V(6).Infof("no ShardReplication object for cell %v", tablet.Alias.Cell)
err = nil
}
if err != nil {
log.Warningf("remove replication data for %v failed: %v", tablet.Alias, err)
}
}
}
// run a hook for final cleanup, only in non-force mode.
// (force mode executes on the vtctl side, not on the vttablet side)
if !force {
hk := hook.NewSimpleHook("postflight_scrap")
ConfigureTabletHook(hk, tablet.Alias)
if hookErr := hk.ExecuteOptional(); hookErr != nil {
// we don't want to return an error, the server
// is already in bad shape probably.
log.Warningf("Scrap: postflight_scrap failed: %v", hookErr)
}
}
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, *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 make it spare.
// 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 = 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()
}