Now reparent doesn't interfere with slave replication status.

If they were replicating, we stop / set master / start, if they were
not, we just set master. Side effect is ReparentTablet may not
start replication if it was stopped, but it's better that way.

go/vt/mysqlctl/mysql_flavor.go

@@ -21,8 +21,9 @@ This file handles the differences between flavors of mysql.
 
 // MysqlFlavor is the abstract interface for a flavor.
 type MysqlFlavor interface {
-	// VersionMatch returns true if the version string (from SELECT VERSION())
-	// represents a server that this flavor knows how to talk to.
+	// VersionMatch returns true if the version string (from
+	// SELECT VERSION()) represents a server that this flavor
+	// knows how to talk to.
 	VersionMatch(version string) bool
 
 	// MasterPosition returns the ReplicationPosition of a master.
@@ -39,40 +40,48 @@ type MysqlFlavor interface {
 	// a slave into a master.
 	PromoteSlaveCommands() []string
 
-	// StartReplicationCommands returns the commands to start replicating from
-	// a given master and position as specified in a ReplicationStatus.
+	// StartReplicationCommands returns the commands to start
+	// replicating from a given master and position as specified
+	// in a ReplicationStatus. It should start replication.
 	StartReplicationCommands(params *sqldb.ConnParams, status *proto.ReplicationStatus) ([]string, error)
 
 	// SetMasterCommands returns the commands to use the provided master
 	// as the new master (without changing any GTID position).
+	// It is guaranteed to be called with replication stopped.
+	// It should not start or stop replication.
 	SetMasterCommands(params *sqldb.ConnParams, masterHost string, masterPort int, masterConnectRetry int) ([]string, error)
 
-	// ParseGTID parses a GTID in the canonical format of this MySQL flavor into
-	// a proto.GTID interface value.
+	// ParseGTID parses a GTID in the canonical format of this
+	// MySQL flavor into a proto.GTID interface value.
 	ParseGTID(string) (proto.GTID, error)
 
-	// ParseReplicationPosition parses a replication position in the canonical
-	// format of this MySQL flavor into a proto.ReplicationPosition struct.
+	// ParseReplicationPosition parses a replication position in
+	// the canonical format of this MySQL flavor into a
+	// proto.ReplicationPosition struct.
 	ParseReplicationPosition(string) (proto.ReplicationPosition, error)
 
-	// SendBinlogDumpCommand sends the flavor-specific version of the
-	// COM_BINLOG_DUMP command to start dumping raw binlog events over a slave
-	// connection, starting at a given GTID.
+	// SendBinlogDumpCommand sends the flavor-specific version of
+	// the COM_BINLOG_DUMP command to start dumping raw binlog
+	// events over a slave connection, starting at a given GTID.
 	SendBinlogDumpCommand(mysqld *Mysqld, conn *SlaveConnection, startPos proto.ReplicationPosition) error
 
-	// MakeBinlogEvent takes a raw packet from the MySQL binlog stream connection
-	// and returns a BinlogEvent through which the packet can be examined.
+	// MakeBinlogEvent takes a raw packet from the MySQL binlog
+	// stream connection and returns a BinlogEvent through which
+	// the packet can be examined.
 	MakeBinlogEvent(buf []byte) blproto.BinlogEvent
 
-	// WaitMasterPos waits until slave replication reaches at least targetPos.
+	// WaitMasterPos waits until slave replication reaches at
+	// least targetPos.
 	WaitMasterPos(mysqld *Mysqld, targetPos proto.ReplicationPosition, waitTimeout time.Duration) error
 
-	// EnableBinlogPlayback prepares the server to play back events from a binlog stream.
-	// Whatever it does for a given flavor, it must be idempotent.
+	// EnableBinlogPlayback prepares the server to play back
+	// events from a binlog stream.  Whatever it does for a given
+	// flavor, it must be idempotent.
 	EnableBinlogPlayback(mysqld *Mysqld) error
 
-	// DisableBinlogPlayback returns the server to the normal state after playback is done.
-	// Whatever it does for a given flavor, it must be idempotent.
+	// DisableBinlogPlayback returns the server to the normal
+	// state after playback is done.  Whatever it does for a given
+	// flavor, it must be idempotent.
 	DisableBinlogPlayback(mysqld *Mysqld) error
 }
 

go/vt/mysqlctl/mysql_flavor_mariadb.go

@@ -125,9 +125,7 @@ func (*mariaDB10) SetMasterCommands(params *sqldb.ConnParams, masterHost string,
 	changeMasterTo := "CHANGE MASTER TO\n  " + strings.Join(args, ",\n  ")
 
 	return []string{
-		"STOP SLAVE",
 		changeMasterTo,
-		"START SLAVE",
 	}, nil
 }
 

go/vt/tabletmanager/agent_rpc_actions.go

@@ -542,11 +542,24 @@ func (agent *ActionAgent) SetMaster(ctx context.Context, parent topo.TabletAlias
 		return err
 	}
 
+	// See if we are replicating at all
+	replicating := false
+	rs, err := agent.MysqlDaemon.SlaveStatus()
+	if err == nil && (rs.SlaveIORunning || rs.SlaveSQLRunning) {
+		replicating = true
+	}
+
 	// TODO(alainjobart) fix the hardcoding of MasterConnectRetry
 	cmds, err := agent.MysqlDaemon.SetMasterCommands(ti.Hostname, ti.Portmap["mysql"], 10)
 	if err != nil {
 		return err
 	}
+	if replicating {
+		newCmds := []string{"STOP SLAVE"}
+		newCmds = append(newCmds, cmds...)
+		newCmds = append(newCmds, "START SLAVE")
+		cmds = newCmds
+	}
 
 	if err := agent.MysqlDaemon.ExecuteSuperQueryList(cmds); err != nil {
 		return err
@@ -567,7 +580,7 @@ func (agent *ActionAgent) SetMaster(ctx context.Context, parent topo.TabletAlias
 
 	// if needed, wait until we get the replicated row, or our
 	// context times out
-	if timeCreatedNS == 0 {
+	if !replicating || timeCreatedNS == 0 {
 		return nil
 	}
 	return agent.MysqlDaemon.WaitForReparentJournal(ctx, timeCreatedNS)

go/vt/vtctl/vtctl.go

@@ -74,6 +74,12 @@ var commands = []commandGroup{
 			command{"SetReadWrite", commandSetReadWrite,
 				"[<tablet alias>]",
 				"Sets the tablet as ReadWrite."},
+			command{"StartSlave", commandStartSlave,
+				"[<tablet alias>]",
+				"Starts replication on the slave."},
+			command{"StopSlave", commandStopSlave,
+				"[<tablet alias>]",
+				"Stops replication on the slave."},
 			command{"ChangeSlaveType", commandChangeSlaveType,
 				"[-force] [-dry-run] <tablet alias> <tablet type>",
 				"Change the db type for this tablet if possible. This is mostly for arranging replicas - it will not convert a master.\n" +
@@ -715,6 +721,44 @@ func commandSetReadWrite(ctx context.Context, wr *wrangler.Wrangler, subFlags *f
 	return wr.TabletManagerClient().SetReadWrite(ctx, ti)
 }
 
+func commandStartSlave(ctx context.Context, wr *wrangler.Wrangler, subFlags *flag.FlagSet, args []string) error {
+	if err := subFlags.Parse(args); err != nil {
+		return err
+	}
+	if subFlags.NArg() != 1 {
+		return fmt.Errorf("action StartSlave requires <tablet alias>")
+	}
+
+	tabletAlias, err := topo.ParseTabletAliasString(subFlags.Arg(0))
+	if err != nil {
+		return err
+	}
+	ti, err := wr.TopoServer().GetTablet(tabletAlias)
+	if err != nil {
+		return fmt.Errorf("failed reading tablet %v: %v", tabletAlias, err)
+	}
+	return wr.TabletManagerClient().StartSlave(ctx, ti)
+}
+
+func commandStopSlave(ctx context.Context, wr *wrangler.Wrangler, subFlags *flag.FlagSet, args []string) error {
+	if err := subFlags.Parse(args); err != nil {
+		return err
+	}
+	if subFlags.NArg() != 1 {
+		return fmt.Errorf("action StopSlave requires <tablet alias>")
+	}
+
+	tabletAlias, err := topo.ParseTabletAliasString(subFlags.Arg(0))
+	if err != nil {
+		return err
+	}
+	ti, err := wr.TopoServer().GetTablet(tabletAlias)
+	if err != nil {
+		return fmt.Errorf("failed reading tablet %v: %v", tabletAlias, err)
+	}
+	return wr.TabletManagerClient().StopSlave(ctx, ti)
+}
+
 func commandChangeSlaveType(ctx context.Context, wr *wrangler.Wrangler, subFlags *flag.FlagSet, args []string) error {
 	force := subFlags.Bool("force", false, "will change the type in zookeeper, and not run hooks")
 	dryRun := subFlags.Bool("dry-run", false, "just list the proposed change")

go/vt/wrangler/testlib/emergency_reparent_shard_test.go

@@ -52,7 +52,7 @@ func TestEmergencyReparentShard(t *testing.T) {
 	oldMaster.StartActionLoop(t, wr)
 	defer oldMaster.StopActionLoop(t)
 
-	// good slave 1
+	// good slave 1 is replicating
 	goodSlave1.FakeMysqlDaemon.ReadOnly = true
 	goodSlave1.FakeMysqlDaemon.CurrentMasterPosition = myproto.ReplicationPosition{
 		GTIDSet: myproto.MariadbGTID{
@@ -61,15 +61,21 @@ func TestEmergencyReparentShard(t *testing.T) {
 			Sequence: 455,
 		},
 	}
+	goodSlave1.FakeMysqlDaemon.CurrentSlaveStatus = &myproto.ReplicationStatus{
+		SlaveIORunning:  true,
+		SlaveSQLRunning: true,
+	}
 	goodSlave1.FakeMysqlDaemon.SetMasterCommandsInput = fmt.Sprintf("%v:%v,%v", newMaster.Tablet.Hostname, newMaster.Tablet.Portmap["mysql"], 10)
 	goodSlave1.FakeMysqlDaemon.SetMasterCommandsResult = []string{"set master cmd 1"}
 	goodSlave1.FakeMysqlDaemon.ExpectedExecuteSuperQueryList = []string{
+		"STOP SLAVE",
 		"set master cmd 1",
+		"START SLAVE",
 	}
 	goodSlave1.StartActionLoop(t, wr)
 	defer goodSlave1.StopActionLoop(t)
 
-	// good slave 2
+	// good slave 2 is not replicating
 	goodSlave2.FakeMysqlDaemon.ReadOnly = true
 	goodSlave2.FakeMysqlDaemon.CurrentMasterPosition = myproto.ReplicationPosition{
 		GTIDSet: myproto.MariadbGTID{

go/vt/wrangler/testlib/planned_reparent_shard_test.go

@@ -67,17 +67,23 @@ func TestPlannedReparentShard(t *testing.T) {
 	defer oldMaster.StopActionLoop(t)
 	oldMaster.Agent.QueryServiceControl.(*tabletserver.TestQueryServiceControl).QueryServiceEnabled = true
 
-	// good slave 1
+	// good slave 1 is replicating
 	goodSlave1.FakeMysqlDaemon.ReadOnly = true
+	goodSlave1.FakeMysqlDaemon.CurrentSlaveStatus = &myproto.ReplicationStatus{
+		SlaveIORunning:  true,
+		SlaveSQLRunning: true,
+	}
 	goodSlave1.FakeMysqlDaemon.SetMasterCommandsInput = fmt.Sprintf("%v:%v,%v", newMaster.Tablet.Hostname, newMaster.Tablet.Portmap["mysql"], 10)
 	goodSlave1.FakeMysqlDaemon.SetMasterCommandsResult = []string{"set master cmd 1"}
 	goodSlave1.FakeMysqlDaemon.ExpectedExecuteSuperQueryList = []string{
+		"STOP SLAVE",
 		"set master cmd 1",
+		"START SLAVE",
 	}
 	goodSlave1.StartActionLoop(t, wr)
 	defer goodSlave1.StopActionLoop(t)
 
-	// good slave 2
+	// good slave 2 is not replicating
 	goodSlave2.FakeMysqlDaemon.ReadOnly = true
 	goodSlave2.FakeMysqlDaemon.SetMasterCommandsInput = fmt.Sprintf("%v:%v,%v", newMaster.Tablet.Hostname, newMaster.Tablet.Portmap["mysql"], 10)
 	goodSlave2.FakeMysqlDaemon.SetMasterCommandsResult = []string{"set master cmd 1"}

test/reparent.py

@@ -613,8 +613,10 @@ class TestReparent(unittest.TestCase):
 
     utils.pause('check orphan')
 
-    # reparent the tablet, should catch up on replication really quickly
+    # reparent the tablet (will not start replication, so we have to
+    # do it ourselves), then it should catch up on replication really quickly
     utils.run_vtctl(['ReparentTablet', tablet_41983.tablet_alias])
+    utils.run_vtctl(['StartSlave', tablet_41983.tablet_alias])
 
     # wait until it gets the data
     timeout = 10.0