Return SourceAlias in FullStatus RPC, use to optimize VTOrc

Signed-off-by: Tim Vaillancourt <[email protected]>

go/vt/vtorc/db/generate_base.go

@@ -53,6 +53,7 @@ CREATE TABLE database_instance (
 	log_replica_updates tinyint NOT NULL,
 	binary_log_file varchar(128) NOT NULL,
 	binary_log_pos bigint NOT NULL,
+	source_alias varchar(256) NOT NULL,
 	source_host varchar(128) NOT NULL,
 	source_port smallint NOT NULL,
 	replica_net_timeout int NOT NULL,

go/vt/vtorc/inst/instance.go

@@ -40,6 +40,7 @@ type Instance struct {
 	LogBinEnabled                bool
 	LogReplicationUpdatesEnabled bool
 	SelfBinlogCoordinates        BinlogCoordinates
+	SourceAlias                  string
 	SourceHost                   string
 	SourcePort                   int
 	SourceUUID                   string

go/vt/vtorc/inst/instance_dao.go

@@ -301,6 +301,7 @@ func ReadTopologyInstanceBufferable(tabletAlias string, latency *stopwatch.Named
 
 		instance.SourceHost = fs.ReplicationStatus.SourceHost
 		instance.SourcePort = int(fs.ReplicationStatus.SourcePort)
+		instance.SourceAlias = topoproto.TabletAliasString(fs.SourceAlias)
 
 		if fs.ReplicationStatus.ReplicationLagUnknown {
 			instance.SecondsBehindPrimary.Valid = false
@@ -478,21 +479,34 @@ func ReadInstanceClusterAttributes(instance *Instance) (err error) {
 	var primaryExecutedGtidSet string
 	primaryDataFound := false
 
-	query := `SELECT
+	primaryAlias := instance.SourceAlias
+	primaryHostname := instance.SourceHost
+	primaryPort := instance.SourcePort
+
+	whereCond := `alias = ?`
+	args := sqlutils.Args(primaryAlias)
+
+	// Fallback to hostname + port query if no primaryAlias defined (added in v22)
+	if primaryAlias == "" {
+		whereCond = `hostname = ? AND port = ?`
+		args = sqlutils.Args(primaryHostname, primaryPort)
+	}
+
+	query := fmt.Sprintf(`SELECT
 		replication_depth,
 		source_host,
 		source_port,
 		ancestry_uuid,
 		executed_gtid_set
 	FROM database_instance
 	WHERE
-		hostname = ?
-		AND port = ?`
-	primaryHostname := instance.SourceHost
-	primaryPort := instance.SourcePort
-	args := sqlutils.Args(primaryHostname, primaryPort)
+		%s`,
+		whereCond,
+	)
+
 	err = db.QueryVTOrc(query, args, func(m sqlutils.RowMap) error {
 		primaryReplicationDepth = m.GetUint("replication_depth")
+		primaryAlias = m.GetString("alias")
 		primaryHostname = m.GetString("source_host")
 		primaryPort = m.GetInt("source_port")
 		ancestryUUID = m.GetString("ancestry_uuid")
@@ -536,6 +550,7 @@ func readInstanceRow(m sqlutils.RowMap) *Instance {
 	instance.BinlogRowImage = m.GetString("binlog_row_image")
 	instance.LogBinEnabled = m.GetBool("log_bin")
 	instance.LogReplicationUpdatesEnabled = m.GetBool("log_replica_updates")
+	instance.SourceAlias = m.GetString("source_alias")
 	instance.SourceHost = m.GetString("source_host")
 	instance.SourcePort = m.GetInt("source_port")
 	instance.ReplicaNetTimeout = m.GetInt32("replica_net_timeout")
@@ -836,6 +851,7 @@ func mkInsertForInstances(instances []*Instance, instanceWasActuallyFound bool,
 		"log_replica_updates",
 		"binary_log_file",
 		"binary_log_pos",
+		"source_alias",
 		"source_host",
 		"source_port",
 		"replica_net_timeout",
@@ -916,6 +932,7 @@ func mkInsertForInstances(instances []*Instance, instanceWasActuallyFound bool,
 		args = append(args, instance.LogReplicationUpdatesEnabled)
 		args = append(args, instance.SelfBinlogCoordinates.LogFile)
 		args = append(args, instance.SelfBinlogCoordinates.LogPos)
+		args = append(args, instance.SourceAlias)
 		args = append(args, instance.SourceHost)
 		args = append(args, instance.SourcePort)
 		args = append(args, instance.ReplicaNetTimeout)

go/vt/vttablet/tabletmanager/rpc_replication.go

@@ -155,17 +155,24 @@ func (tm *TabletManager) FullStatus(ctx context.Context) (*replicationdatapb.Ful
 	// Semi sync status - "show status like 'Rpl_semi_sync_%_status'"
 	primarySemiSyncStatus, replicaSemiSyncStatus := tm.MysqlDaemon.SemiSyncStatus(ctx)
 
-	//  Semi sync clients count - "show status like 'semi_sync_source_clients'"
+	// Semi sync clients count - "show status like 'semi_sync_source_clients'"
 	semiSyncClients := tm.MysqlDaemon.SemiSyncClients(ctx)
 
 	// Semi sync settings - "show status like 'rpl_semi_sync_%'
 	semiSyncTimeout, semiSyncNumReplicas := tm.MysqlDaemon.SemiSyncSettings(ctx)
 
+	// Replication configuration
 	replConfiguration, err := tm.MysqlDaemon.ReplicationConfiguration(ctx)
 	if err != nil {
 		return nil, err
 	}
 
+	// Replication source tablet alias
+	sourceAlias, err := tm.getReplicationSourceAlias(ctx)
+	if err != nil {
+		return nil, err
+	}
+
 	return &replicationdatapb.FullStatus{
 		ServerId:                    serverID,
 		ServerUuid:                  serverUUID,
@@ -189,6 +196,7 @@ func (tm *TabletManager) FullStatus(ctx context.Context) (*replicationdatapb.Ful
 		SemiSyncWaitForReplicaCount: semiSyncNumReplicas,
 		SuperReadOnly:               superReadOnly,
 		ReplicationConfiguration:    replConfiguration,
+		SourceAlias:                 sourceAlias,
 	}, nil
 }
 
@@ -731,6 +739,14 @@ func (tm *TabletManager) SetReplicationSource(ctx context.Context, parentAlias *
 	return tm.setReplicationSourceLocked(ctx, parentAlias, timeCreatedNS, waitPosition, forceStartReplication, semiSyncAction, heartbeatInterval)
 }
 
+func (tm *TabletManager) getReplicationSourceAlias(ctx context.Context) (*topodatapb.TabletAlias, error) {
+	if err := tm.lock(ctx); err != nil {
+		return nil, err
+	}
+	defer tm.unlock()
+	return tm.sourceAlias, nil
+}
+
 func (tm *TabletManager) setReplicationSourceSemiSyncNoAction(ctx context.Context, parentAlias *topodatapb.TabletAlias, timeCreatedNS int64, waitPosition string, forceStartReplication bool) error {
 	log.Infof("SetReplicationSource: parent: %v  position: %v force: %v", parentAlias, waitPosition, forceStartReplication)
 	if err := tm.lock(ctx); err != nil {
@@ -883,6 +899,9 @@ func (tm *TabletManager) setReplicationSourceLocked(ctx context.Context, parentA
 		}
 	}
 
+	// Store the current primary alias
+	tm.sourceAlias = parentAlias
+
 	return nil
 }
 

go/vt/vttablet/tabletmanager/tm_init.go

@@ -174,6 +174,9 @@ type TabletManager struct {
 	// tabletAlias is saved away from tablet for read-only access
 	tabletAlias *topodatapb.TabletAlias
 
+	// sourceAlias is the current replication source.
+	sourceAlias *topodatapb.TabletAlias
+
 	// baseTabletType is the tablet type we revert back to
 	// when we transition back from something like PRIMARY.
 	baseTabletType topodatapb.TabletType