Initial support for multimaster

Author: Guillaume Lefranc

failover.go

@@ -41,7 +41,9 @@ func masterFailover(fail bool) {
 	}
 	master = servers[skey]
 	master.State = stateMaster
-	slaves[key].delete(&slaves)
+	if multiMaster == false {
+		slaves[key].delete(&slaves)
+	}
 	// Call pre-failover script
 	if preScript != "" {
 		logprintf("INFO : Calling pre-failover script")
@@ -75,10 +77,12 @@ func masterFailover(fail bool) {
 		}
 	}
 	// Phase 3: Prepare new master
-	logprint("INFO : Stopping slave thread on new master")
-	err = dbhelper.StopSlave(master.Conn)
-	if err != nil {
-		logprint("WARN : Stopping slave failed on new master")
+	if multiMaster == false {
+		logprint("INFO : Stopping slave thread on new master")
+		err = dbhelper.StopSlave(master.Conn)
+		if err != nil {
+			logprint("WARN : Stopping slave failed on new master")
+		}
 	}
 	// Call post-failover script before unlocking the old master.
 	if postScript != "" {
@@ -89,10 +93,12 @@ func masterFailover(fail bool) {
 		}
 		logprint("INFO : Post-failover script complete", string(out))
 	}
-	logprint("INFO : Resetting slave on new master and set read/write mode on")
-	err = dbhelper.ResetSlave(master.Conn, true)
-	if err != nil {
-		logprint("WARN : Reset slave failed on new master")
+	if multiMaster == false {
+		logprint("INFO : Resetting slave on new master and set read/write mode on")
+		err = dbhelper.ResetSlave(master.Conn, true)
+		if err != nil {
+			logprint("WARN : Reset slave failed on new master")
+		}
 	}
 	err = dbhelper.SetReadOnly(master.Conn, false)
 	if err != nil {
@@ -134,11 +140,17 @@ func masterFailover(fail bool) {
 		}
 		// Add the old master to the slaves list
 		oldMaster.State = stateSlave
-		slaves = append(slaves, oldMaster)
+		if multiMaster == false {
+			slaves = append(slaves, oldMaster)
+		}
 	}
 	// Phase 5: Switch slaves to new master
 	logprint("INFO : Switching other slaves to the new master")
 	for _, sl := range slaves {
+		// Don't switch if slave was the old master or is in a multiple master setup.
+		if sl.URL == oldMaster.URL || sl.State == stateMaster {
+			continue
+		}
 		if fail == false {
 			logprintf("INFO : Waiting for slave %s to sync", sl.URL)
 			dbhelper.MasterPosWait(sl.Conn, oldMaster.BinlogPos)

monitor.go

@@ -215,12 +215,16 @@ func (server *ServerMonitor) electCandidate(l []*ServerMonitor) int {
 		logprintf("DEBUG: Processing %d candidates", ll)
 	}
 	seqList := make([]uint64, ll)
-	i := 0
 	hiseq := 0
-	for _, sl := range l {
+	var max uint64
+	for i, sl := range l {
 		if server.State != stateFailed {
 			if verbose {
-				logprintf("DEBUG: Checking eligibility of slave server %s", sl.URL)
+				logprintf("DEBUG: Checking eligibility of slave server %s [%d]", sl.URL, i)
+			}
+			if multiMaster == true && sl.State == stateMaster {
+				logprintf("WARN : Slave %s has state Master. Skipping", sl.URL)
+				continue
 			}
 			if dbhelper.CheckSlavePrerequisites(sl.Conn, sl.Host) == false {
 				continue
@@ -262,16 +266,15 @@ func (server *ServerMonitor) electCandidate(l []*ServerMonitor) int {
 			return i
 		}
 		seqList[i] = getSeqFromGtid(dbhelper.GetVariableByName(sl.Conn, "GTID_CURRENT_POS"))
-		var max uint64
-		if i == 0 {
-			max = seqList[0]
-		} else if seqList[i] > max {
+		if verbose {
+			logprintf("DEBUG: Got sequence number %d for server [%d]", seqList[i], i)
+		}
+		if seqList[i] > max {
 			max = seqList[i]
 			hiseq = i
 		}
-		i++
 	}
-	if i > 0 {
+	if max > 0 {
 		/* Return key of slave with the highest seqno. */
 		return hiseq
 	}

repmgr.go

@@ -74,6 +74,7 @@ var (
 	failtime    int64
 	checktype   string
 	masterConn  string
+	multiMaster bool
 )
 
 func init() {
@@ -110,6 +111,7 @@ func initRepmgrFlags(cmd *cobra.Command) {
 	cmd.Flags().StringVar(&logfile, "logfile", "", "Write MRM messages to a log file")
 	cmd.Flags().IntVar(&timeout, "connect-timeout", 5, "Database connection timeout in seconds")
 	cmd.Flags().StringVar(&masterConn, "master-connection", "", "Connection name to use for multisource replication")
+	cmd.Flags().BoolVar(&multiMaster, "multimaster", false, "Turn on multi-master detection")
 	viper.BindPFlags(cmd.Flags())
 	preScript = viper.GetString("pre-failover-script")
 	postScript = viper.GetString("post-failover-script")
@@ -122,6 +124,7 @@ func initRepmgrFlags(cmd *cobra.Command) {
 	logfile = viper.GetString("logfile")
 	timeout = viper.GetInt("connect-timeout")
 	masterConn = viper.GetString("master-connection")
+	multiMaster = viper.GetBool("multimaster")
 }
 
 var failoverCmd = &cobra.Command{

topology.go

@@ -17,6 +17,7 @@ type topologyError struct {
 
 func init() {
 	rootCmd.AddCommand(topologyCmd)
+	topologyCmd.Flags().BoolVar(&multiMaster, "multimaster", false, "Turn on multi-master detection")
 }
 
 func (e topologyError) Error() string {
@@ -77,11 +78,26 @@ func topologyInit() error {
 	}
 
 	// Check that all slave servers have the same master.
-	for _, sl := range slaves {
-		if sl.hasSiblings(slaves) == false {
+	if multiMaster == false {
+		for _, sl := range slaves {
+			if sl.hasSiblings(slaves) == false {
+				return topologyError{
+					33,
+					fmt.Sprintf("ERROR: Multiple masters were detected"),
+				}
+			}
+		}
+	} else {
+		srw := 0
+		for _, s := range servers {
+			if s.ReadOnly == "OFF" {
+				srw++
+			}
+		}
+		if srw != 1 {
 			return topologyError{
-				33,
-				fmt.Sprintf("ERROR: Multiple masters were detected"),
+				11,
+				fmt.Sprintf("ERROR: RW server count > 1 in multi-master mode. Please set slaves to RO"),
 			}
 		}
 	}
@@ -91,7 +107,7 @@ func topologyInit() error {
 	// First of all, get a server id from the slaves slice, they should be all the same
 	sid := slaves[0].MasterServerID
 	for k, s := range servers {
-		if s.State == stateUnconn {
+		if multiMaster == false && s.State == stateUnconn {
 			if s.ServerID == sid {
 				master = servers[k]
 				master.State = stateMaster
@@ -101,6 +117,16 @@ func topologyInit() error {
 				break
 			}
 		}
+		if multiMaster == true {
+			if s.ReadOnly == "OFF" {
+				master = servers[k]
+				master.State = stateMaster
+				if verbose {
+					log.Printf("DEBUG: Server %s was autodetected as a master", s.URL)
+				}
+				break
+			}
+		}
 	}
 	// If master is not initialized, find it in the failed hosts list
 	if master == nil {