func createSourceTablet(t *testing.T, ts topo.Server, keyspace, shard string) {
vshard, kr, err := topo.ValidateShardName(shard)
if err != nil {
t.Fatalf("ValidateShardName(%v) failed: %v", shard, err)
}
ctx := context.Background()
tablet := &pb.Tablet{
Alias: &pb.TabletAlias{
Cell: "cell1",
Uid: 100,
},
Type: pb.TabletType_REPLICA,
KeyRange: kr,
Keyspace: keyspace,
Shard: vshard,
PortMap: map[string]int32{
"vt": 80,
},
}
if err := ts.CreateTablet(ctx, tablet); err != nil {
t.Fatalf("CreateTablet failed: %v", err)
}
if err := topotools.UpdateTabletEndpoints(ctx, ts, tablet); err != nil {
t.Fatalf("topotools.UpdateTabletEndpoints failed: %v", err)
}
}
// updateServingGraph will update the serving graph if we need to.
func (agent *ActionAgent) updateServingGraph(tablet *topo.TabletInfo, targetTabletType pbt.TabletType) error {
if topo.IsInServingGraph(targetTabletType) {
if err := topotools.UpdateTabletEndpoints(agent.batchCtx, agent.TopoServer, tablet.Tablet); err != nil {
return fmt.Errorf("UpdateTabletEndpoints failed: %v", err)
}
}
return nil
}
func (agent *ActionAgent) verifyServingAddrs(ctx context.Context) error {
ti := agent.Tablet()
if !ti.IsRunningQueryService() {
return nil
}
// Check to see our address is registered in the right place.
return topotools.UpdateTabletEndpoints(ctx, agent.TopoServer, ti.Tablet)
}
// createSourceTablet is a helper method to create the source tablet
// in the given keyspace/shard.
func createSourceTablet(t *testing.T, name string, ts topo.Server, keyspace, shard string) {
vshard, kr, err := topo.ValidateShardName(shard)
if err != nil {
t.Fatalf("ValidateShardName(%v) failed: %v", shard, err)
}
ctx := context.Background()
tablet := &topodatapb.Tablet{
Alias: &topodatapb.TabletAlias{
Cell: "cell1",
Uid: 100,
},
Type: topodatapb.TabletType_REPLICA,
KeyRange: kr,
Keyspace: keyspace,
Shard: vshard,
PortMap: map[string]int32{
"vt": 80,
},
}
if err := ts.CreateTablet(ctx, tablet); err != nil {
t.Fatalf("CreateTablet failed: %v", err)
}
if err := topotools.UpdateTabletEndpoints(ctx, ts, tablet); err != nil {
t.Fatalf("topotools.UpdateTabletEndpoints failed: %v", err)
}
// register a tablet conn dialer that will return the instance
// we want
tabletconn.RegisterDialer(name, func(ctx context.Context, endPoint *topodatapb.EndPoint, k, s string, tabletType topodatapb.TabletType, timeout time.Duration) (tabletconn.TabletConn, error) {
return &fakeTabletConn{
endPoint: endPoint,
keyspace: keyspace,
shard: vshard,
tabletType: topodatapb.TabletType_REPLICA,
}, nil
})
flag.Lookup("tablet_protocol").Value.Set(name)
}
// InitTablet initializes the tablet record if necessary.
func (agent *ActionAgent) InitTablet(port, gRPCPort int32) error {
// only enabled if one of init_tablet_type (when healthcheck
// is disabled) or init_keyspace (when healthcheck is enabled)
// is passed in, then check other parameters
if *initTabletType == "" && *initKeyspace == "" {
return nil
}
// figure out our default target type
var tabletType topodatapb.TabletType
if *initTabletType != "" {
if *targetTabletType != "" {
log.Fatalf("cannot specify both target_tablet_type and init_tablet_type parameters (as they might conflict)")
}
// use the type specified on the command line
var err error
tabletType, err = topoproto.ParseTabletType(*initTabletType)
if err != nil {
log.Fatalf("Invalid init tablet type %v: %v", *initTabletType, err)
}
if tabletType == topodatapb.TabletType_MASTER {
// We disallow MASTER, so we don't have to change
// shard.MasterAlias, and deal with the corner cases.
log.Fatalf("init_tablet_type cannot be %v", tabletType)
}
} else if *targetTabletType != "" {
if strings.ToUpper(*targetTabletType) == topodatapb.TabletType_name[int32(topodatapb.TabletType_MASTER)] {
log.Fatalf("target_tablet_type cannot be '%v'. Use '%v' instead.", tabletType, topodatapb.TabletType_REPLICA)
}
// use spare, the healthcheck will turn us into what
// we need to be eventually
tabletType = topodatapb.TabletType_SPARE
} else {
log.Fatalf("if init tablet is enabled, one of init_tablet_type or target_tablet_type needs to be specified")
}
// create a context for this whole operation
ctx, cancel := context.WithTimeout(agent.batchCtx, *initTimeout)
defer cancel()
// since we're assigned to a shard, make sure it exists, see if
// we are its master, and update its cells list if necessary
if *initKeyspace == "" || *initShard == "" {
log.Fatalf("if init tablet is enabled and the target type is not idle, init_keyspace and init_shard also need to be specified")
}
shard, _, err := topo.ValidateShardName(*initShard)
if err != nil {
log.Fatalf("cannot validate shard name: %v", err)
}
log.Infof("Reading shard record %v/%v", *initKeyspace, shard)
// read the shard, create it if necessary
si, err := topotools.GetOrCreateShard(ctx, agent.TopoServer, *initKeyspace, shard)
if err != nil {
return fmt.Errorf("InitTablet cannot GetOrCreateShard shard: %v", err)
}
if si.MasterAlias != nil && topoproto.TabletAliasEqual(si.MasterAlias, agent.TabletAlias) {
// we are the current master for this shard (probably
// means the master tablet process was just restarted),
// so InitTablet as master.
tabletType = topodatapb.TabletType_MASTER
}
// See if we need to add the tablet's cell to the shard's cell
// list. If we do, it has to be under the shard lock.
if !si.HasCell(agent.TabletAlias.Cell) {
actionNode := actionnode.UpdateShard()
lockPath, err := actionNode.LockShard(ctx, agent.TopoServer, *initKeyspace, shard)
if err != nil {
return fmt.Errorf("LockShard(%v/%v) failed: %v", *initKeyspace, shard, err)
}
// re-read the shard with the lock
si, err = agent.TopoServer.GetShard(ctx, *initKeyspace, shard)
if err != nil {
return actionNode.UnlockShard(ctx, agent.TopoServer, *initKeyspace, shard, lockPath, err)
}
// see if we really need to update it now
if !si.HasCell(agent.TabletAlias.Cell) {
si.Cells = append(si.Cells, agent.TabletAlias.Cell)
// write it back
if err := agent.TopoServer.UpdateShard(ctx, si); err != nil {
return actionNode.UnlockShard(ctx, agent.TopoServer, *initKeyspace, shard, lockPath, err)
}
}
// and unlock
if err := actionNode.UnlockShard(ctx, agent.TopoServer, *initKeyspace, shard, lockPath, nil); err != nil {
return err
}
}
log.Infof("Initializing the tablet for type %v", tabletType)
// figure out the hostname
hostname := *tabletHostname
if hostname != "" {
log.Infof("Using hostname: %v from -tablet_hostname flag.", hostname)
} else {
hostname, err := netutil.FullyQualifiedHostname()
if err != nil {
return err
}
log.Infof("Using detected machine hostname: %v To change this, fix your machine network configuration or override it with -tablet_hostname.", hostname)
}
// create and populate tablet record
tablet := &topodatapb.Tablet{
Alias: agent.TabletAlias,
Hostname: hostname,
PortMap: make(map[string]int32),
Keyspace: *initKeyspace,
Shard: *initShard,
Type: tabletType,
DbNameOverride: *initDbNameOverride,
Tags: initTags,
}
if port != 0 {
tablet.PortMap["vt"] = port
}
if gRPCPort != 0 {
tablet.PortMap["grpc"] = gRPCPort
}
if err := topo.TabletComplete(tablet); err != nil {
return fmt.Errorf("InitTablet TabletComplete failed: %v", err)
}
// now try to create the record
err = agent.TopoServer.CreateTablet(ctx, tablet)
switch err {
case nil:
// it worked, we're good, can update the replication graph
if err := topo.UpdateTabletReplicationData(ctx, agent.TopoServer, tablet); err != nil {
return fmt.Errorf("UpdateTabletReplicationData failed: %v", err)
}
case topo.ErrNodeExists:
// The node already exists, will just try to update
// it. So we read it first.
oldTablet, err := agent.TopoServer.GetTablet(ctx, tablet.Alias)
if err != nil {
return fmt.Errorf("InitTablet failed to read existing tablet record: %v", err)
}
// Sanity check the keyspace and shard
if oldTablet.Keyspace != tablet.Keyspace || oldTablet.Shard != tablet.Shard {
return fmt.Errorf("InitTablet failed because existing tablet keyspace and shard %v/%v differ from the provided ones %v/%v", oldTablet.Keyspace, oldTablet.Shard, tablet.Keyspace, tablet.Shard)
}
// And overwrite the rest
*(oldTablet.Tablet) = *tablet
if err := agent.TopoServer.UpdateTablet(ctx, oldTablet); err != nil {
return fmt.Errorf("UpdateTablet failed: %v", err)
}
// Note we don't need to UpdateTabletReplicationData
// as the tablet already existed with the right data
// in the replication graph
default:
return fmt.Errorf("CreateTablet failed: %v", err)
}
// and now update the serving graph. Note we do that in any case,
// to clean any inaccurate record from any part of the serving graph.
if err := topotools.UpdateTabletEndpoints(ctx, agent.TopoServer, tablet); err != nil {
return fmt.Errorf("UpdateTabletEndpoints failed: %v", err)
}
return nil
}
// finalizeTabletExternallyReparented performs slow, synchronized reconciliation
// tasks that ensure topology is self-consistent, and then marks the reparent as
// finished by updating the global shard record.
func (agent *ActionAgent) finalizeTabletExternallyReparented(ctx context.Context, si *topo.ShardInfo, ev *events.Reparent) (err error) {
var wg sync.WaitGroup
var errs concurrency.AllErrorRecorder
oldMasterAlias := si.MasterAlias
// Update the tablet records and serving graph for the old and new master concurrently.
event.DispatchUpdate(ev, "updating old and new master tablet records")
log.Infof("finalizeTabletExternallyReparented: updating tablet records")
wg.Add(1)
go func() {
defer wg.Done()
// Update our own record to master.
updatedTablet, err := agent.TopoServer.UpdateTabletFields(ctx, agent.TabletAlias,
func(tablet *topodatapb.Tablet) error {
tablet.Type = topodatapb.TabletType_MASTER
tablet.HealthMap = nil
return nil
})
if err != nil {
errs.RecordError(err)
return
}
// Update the serving graph for the tablet.
if updatedTablet != nil {
errs.RecordError(
topotools.UpdateTabletEndpoints(ctx, agent.TopoServer, updatedTablet))
}
}()
if !topoproto.TabletAliasIsZero(oldMasterAlias) {
wg.Add(1)
go func() {
// Forcibly demote the old master in topology, since we can't rely on the
// old master to be up to change its own record.
oldMasterTablet, err := agent.TopoServer.UpdateTabletFields(ctx, oldMasterAlias,
func(tablet *topodatapb.Tablet) error {
tablet.Type = topodatapb.TabletType_SPARE
return nil
})
if err != nil {
errs.RecordError(err)
wg.Done()
return
}
// We now know more about the old master, so add it to event data.
ev.OldMaster = *oldMasterTablet
// Update the serving graph.
errs.RecordError(
topotools.UpdateTabletEndpoints(ctx, agent.TopoServer, oldMasterTablet))
wg.Done()
// Tell the old master to re-read its tablet record and change its state.
// We don't need to wait for it.
tmc := tmclient.NewTabletManagerClient()
tmc.RefreshState(ctx, topo.NewTabletInfo(oldMasterTablet, -1))
}()
}
tablet := agent.Tablet()
// Wait for the tablet records to be updated. At that point, any rebuild will
// see the new master, so we're ready to mark the reparent as done in the
// global shard record.
wg.Wait()
if errs.HasErrors() {
return errs.Error()
}
// Update the master field in the global shard record. We don't use a lock
// here anymore. The lock was only to ensure that the global shard record
// didn't get modified between the time when we read it and the time when we
// write it back. Now we use an update loop pattern to do that instead.
event.DispatchUpdate(ev, "updating global shard record")
log.Infof("finalizeTabletExternallyReparented: updating global shard record")
si, err = agent.TopoServer.UpdateShardFields(ctx, tablet.Keyspace, tablet.Shard, func(shard *topodatapb.Shard) error {
shard.MasterAlias = tablet.Alias
return nil
})
if err != nil {
return err
}
// We already took care of updating the serving graph for the old and new masters.
// All that's left now is in case of a cross-cell reparent, we need to update the
// master cell setting in the SrvShard records of all cells.
if oldMasterAlias == nil || oldMasterAlias.Cell != tablet.Alias.Cell {
event.DispatchUpdate(ev, "rebuilding shard serving graph")
log.Infof("finalizeTabletExternallyReparented: updating SrvShard in all cells for cross-cell reparent")
if err := topotools.UpdateAllSrvShards(ctx, agent.TopoServer, si); err != nil {
return err
}
}
event.DispatchUpdate(ev, "finished")
return nil
}
