func clearSlave(dbName string, swm *swarm.Swarm) error {
slaveStatus, err := swm.SlaveShowStatus(dbName)
if err != nil {
return err
}
if slaveStatus.MasterIp == "" {
return nil
}
err = swm.SlaveStop(dbName)
if err != nil {
return err
}
err = swm.SlaveReset(dbName)
if err != nil {
return err
}
return nil
}
func changeMaster(swm *swarm.Swarm, fromName, toIp string, toPort int) error {
slaveStatus, err := swm.SlaveShowStatus(fromName)
if err != nil {
return err
}
log.WithFields(log.Fields{
"DbName": fromName,
"SlaveStatus": slaveStatus,
}).Debug("change master")
if slaveStatus.MasterIp == toIp && slaveStatus.MasterPort == toPort {
if slaveStatus.SlaveIoRunning == "No" && slaveStatus.SlaveSqlRunning == "No" {
// 1.4 start slave
err = swm.SlaveStart(fromName)
if err != nil {
return err
}
}
return nil
}
if slaveStatus.SlaveIoRunning == "Yes" || slaveStatus.SlaveSqlRunning == "Yes" {
// 1.1 stop slave
err = swm.SlaveStop(fromName)
if err != nil {
return err
}
}
if slaveStatus.MasterIp != "" {
// 1.2 reset slave
err = swm.SlaveReset(fromName)
if err != nil {
return err
}
}
// 1.3 get gtid_executed
gtidExecuted, err := swm.GetGtidExecuted(fromName)
if err != nil {
return err
}
if gtidExecuted != "" {
// 1.4 reset master
err = swm.MasterReset(fromName)
if err != nil {
return err
}
// 1.5 set gtid_purged
err = swm.SetGtidPurged(fromName, gtidExecuted)
if err != nil {
return err
}
}
// 1.6 change master
err = swm.ChangeMaster(fromName, toIp, toPort)
if err != nil {
return err
}
// 1.7 start slave
err = swm.SlaveStart(fromName)
if err != nil {
return err
}
return nil
}
// no need RLock
func RecoverDb(dbName string, rwMutex *sync.RWMutex, topology *structs.Topology, swm *swarm.Swarm) (error, []string) {
var err error
var steps []string
defer func() {
if err != nil {
log.WithFields(log.Fields{
"DbName": dbName,
"err": err.Error(),
}).Error("recover fail")
} else {
log.WithFields(log.Fields{
"DbName": dbName,
}).Info("recover success")
}
}()
var recoverDbInfo *structs.DatabaseInfo
for k, v := range topology.DataNodeGroup["default"] {
if k == dbName {
recoverDbInfo = v
break
}
}
if recoverDbInfo == nil {
return errors.New("recover db node not found"), steps
}
// 1. check m slave status
if recoverDbInfo.Type == consts.StandBy || recoverDbInfo.Type == consts.Slave {
mName, _ := getMaster(topology)
var masterSlStatus structs.SlaveStatus
masterSlStatus, err = swm.SlaveShowStatus(mName)
if err != nil {
steps = append(steps, fmt.Sprintf("Step1. Clear M status fail: %s\n", err.Error()))
return err, steps
}
if masterSlStatus.MasterIp != "" {
// 1.1 stop m slave
err = swm.SlaveStop(mName)
if err != nil {
steps = append(steps, fmt.Sprintf("Step1. Clear M status fail: %s\n", err.Error()))
return err, steps
}
// 1.2 reset m slave
err = swm.SlaveReset(mName)
if err != nil {
steps = append(steps, fmt.Sprintf("Step1. Clear M status fail: %s\n", err.Error()))
return err, steps
}
}
}
steps = append(steps, "Step1. Clear M status success\n")
// 2. recover change master
if recoverDbInfo.Type == consts.StandBy {
mName, mDbInfo := getMaster(topology)
err = changeMaster(swm, dbName, mDbInfo.Ip, mDbInfo.Port)
if err != nil {
log.WithFields(log.Fields{
"Sb": dbName + "(" + recoverDbInfo.Ip + ":" + strconv.Itoa(recoverDbInfo.Port) + ")",
"M": mName + "(" + mDbInfo.Ip + ":" + strconv.Itoa(mDbInfo.Port) + ")",
"err:": err.Error(),
}).Error("change master fail, return")
steps = append(steps, fmt.Sprintf("Step2. Sb change master to M fail: %s\n", err.Error()))
return err, steps
}
log.WithFields(log.Fields{
"Sb": dbName + "(" + recoverDbInfo.Ip + ":" + strconv.Itoa(recoverDbInfo.Port) + ")",
"M": mName + "(" + mDbInfo.Ip + ":" + strconv.Itoa(mDbInfo.Port) + ")",
}).Debug("change master success")
steps = append(steps, "Step2. Sb change master to M success\n")
} else if recoverDbInfo.Type == consts.Slave {
sbName, sbDbInfo := getStandBy(topology)
err = changeMaster(swm, dbName, sbDbInfo.Ip, sbDbInfo.Port)
if err != nil {
log.WithFields(log.Fields{
"Sl": dbName + "(" + recoverDbInfo.Ip + ":" + strconv.Itoa(recoverDbInfo.Port) + ")",
"Sb": sbName + "(" + sbDbInfo.Ip + ":" + strconv.Itoa(sbDbInfo.Port) + ")",
"err:": err.Error(),
}).Error("change master fail, return")
steps = append(steps, fmt.Sprintf("Step2. Sl change master to Sb fail: %s\n", err.Error()))
return err, steps
}
log.WithFields(log.Fields{
"Sl": dbName + "(" + recoverDbInfo.Ip + ":" + strconv.Itoa(recoverDbInfo.Port) + ")",
"Sb": sbName + "(" + sbDbInfo.Ip + ":" + strconv.Itoa(sbDbInfo.Port) + ")",
}).Info("change master success")
steps = append(steps, "Step2. Sl change master to Sb success\n")
}
// 3. update topology
rwMutex.Lock()
dbInfo := getDbInfo(dbName, topology)
dbInfo.Status = consts.Normal
updateVersionAndNormalCount(topology, 1)
rwMutex.Unlock()
return nil, steps
}