// InitMaster breaks slaves replication, get the current MySQL replication
// position, insert a row in the reparent_journal table, and returns
// the replication position
func (agent *ActionAgent) InitMaster(ctx context.Context) (myproto.ReplicationPosition, error) {
// we need to insert something in the binlogs, so we can get the
// current position. Let's just use the mysqlctl.CreateReparentJournal commands.
cmds := mysqlctl.CreateReparentJournal()
if err := agent.MysqlDaemon.ExecuteSuperQueryList(cmds); err != nil {
return myproto.ReplicationPosition{}, err
}
// get the current replication position
rp, err := agent.MysqlDaemon.MasterPosition()
if err != nil {
return myproto.ReplicationPosition{}, err
}
// Set the server read-write, from now on we can accept real
// client writes. Note that if semi-sync replication is enabled,
// we'll still need some slaves to be able to commit
// transactions.
if err := agent.MysqlDaemon.SetReadOnly(false); err != nil {
return myproto.ReplicationPosition{}, err
}
// Change our type to master if not already
if err := topo.UpdateTabletFields(ctx, agent.TopoServer, agent.TabletAlias, func(tablet *topo.Tablet) error {
tablet.Type = topo.TYPE_MASTER
tablet.Health = nil
return nil
}); err != nil {
return myproto.ReplicationPosition{}, err
}
agent.initReplication = true
return rp, nil
}
// CheckTablet verifies the topo server API is correct for managing tablets.
func CheckTablet(ctx context.Context, t *testing.T, ts topo.Server) {
cell := getLocalCell(ctx, t, ts)
tablet := &topo.Tablet{
Alias: topo.TabletAlias{Cell: cell, Uid: 1},
Hostname: "localhost",
IPAddr: "10.11.12.13",
Portmap: map[string]int{
"vt": 3333,
"mysql": 3334,
},
Tags: map[string]string{"tag": "value"},
Keyspace: "test_keyspace",
Type: topo.TYPE_MASTER,
KeyRange: newKeyRange("-10"),
}
if err := ts.CreateTablet(ctx, tablet); err != nil {
t.Errorf("CreateTablet: %v", err)
}
if err := ts.CreateTablet(ctx, tablet); err != topo.ErrNodeExists {
t.Errorf("CreateTablet(again): %v", err)
}
if _, err := ts.GetTablet(ctx, topo.TabletAlias{Cell: cell, Uid: 666}); err != topo.ErrNoNode {
t.Errorf("GetTablet(666): %v", err)
}
ti, err := ts.GetTablet(ctx, tablet.Alias)
if err != nil {
t.Errorf("GetTablet %v: %v", tablet.Alias, err)
}
if eq, err := tabletEqual(ti.Tablet, tablet); err != nil {
t.Errorf("cannot compare tablets: %v", err)
} else if !eq {
t.Errorf("put and got tablets are not identical:\n%#v\n%#v", tablet, ti.Tablet)
}
if _, err := ts.GetTabletsByCell(ctx, "666"); err != topo.ErrNoNode {
t.Errorf("GetTabletsByCell(666): %v", err)
}
inCell, err := ts.GetTabletsByCell(ctx, cell)
if err != nil {
t.Errorf("GetTabletsByCell: %v", err)
}
if len(inCell) != 1 || inCell[0] != tablet.Alias {
t.Errorf("GetTabletsByCell: want [%v], got %v", tablet.Alias, inCell)
}
ti.Hostname = "remotehost"
if err := topo.UpdateTablet(ctx, ts, ti); err != nil {
t.Errorf("UpdateTablet: %v", err)
}
ti, err = ts.GetTablet(ctx, tablet.Alias)
if err != nil {
t.Errorf("GetTablet %v: %v", tablet.Alias, err)
}
if want := "remotehost"; ti.Hostname != want {
t.Errorf("ti.Hostname: want %v, got %v", want, ti.Hostname)
}
if err := topo.UpdateTabletFields(ctx, ts, tablet.Alias, func(t *topo.Tablet) error {
t.Hostname = "anotherhost"
return nil
}); err != nil {
t.Errorf("UpdateTabletFields: %v", err)
}
ti, err = ts.GetTablet(ctx, tablet.Alias)
if err != nil {
t.Errorf("GetTablet %v: %v", tablet.Alias, err)
}
if want := "anotherhost"; ti.Hostname != want {
t.Errorf("ti.Hostname: want %v, got %v", want, ti.Hostname)
}
if err := ts.DeleteTablet(ctx, tablet.Alias); err != nil {
t.Errorf("DeleteTablet: %v", err)
}
if err := ts.DeleteTablet(ctx, tablet.Alias); err != topo.ErrNoNode {
t.Errorf("DeleteTablet(again): %v", err)
}
if _, err := ts.GetTablet(ctx, tablet.Alias); err != topo.ErrNoNode {
t.Errorf("GetTablet: expected error, tablet was deleted: %v", err)
}
}
// 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.
var updatedTablet *topo.Tablet
err := topo.UpdateTabletFields(ctx, agent.TopoServer, agent.TabletAlias,
func(tablet *topo.Tablet) error {
tablet.Type = topo.TYPE_MASTER
tablet.Health = nil
updatedTablet = tablet
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 !oldMasterAlias.IsZero() {
wg.Add(1)
go func() {
// Force the old master to spare.
var oldMasterTablet *topo.Tablet
err := topo.UpdateTabletFields(ctx, agent.TopoServer, oldMasterAlias,
func(tablet *topo.Tablet) error {
tablet.Type = topo.TYPE_SPARE
oldMasterTablet = tablet
return nil
})
if err != nil {
errs.RecordError(err)
wg.Done()
return
}
if oldMasterTablet != nil {
// Update the serving graph.
errs.RecordError(
topotools.UpdateTabletEndpoints(ctx, agent.TopoServer, oldMasterTablet))
wg.Done()
// Tell the old master to refresh 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 = topo.UpdateShardFields(ctx, agent.TopoServer, tablet.Keyspace, tablet.Shard, func(shard *topo.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.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
}