func main() {
flag.Parse()
if *version == true {
fmt.Println("MariaDB Replication Manager version", repmgrVersion)
}
// if slaves option has been supplied, split into a slice.
if *hosts != "" {
hostList = strings.Split(*hosts, ",")
} else {
log.Fatal("ERROR: No hosts list specified.")
}
// validate users.
if *user == "" {
log.Fatal("ERROR: No master user/pair specified.")
}
dbUser, dbPass = splitPair(*user)
if *rpluser == "" {
log.Fatal("ERROR: No replication user/pair specified.")
}
rplUser, rplPass = splitPair(*rpluser)
// Create a connection to each host.
hostCount := len(hostList)
hhdls = make([]*ServerMonitor, hostCount)
slaveCount := 0
for k, url := range hostList {
var err error
hhdls[k], err = newServerMonitor(url)
if *verbose {
log.Printf("DEBUG: Creating new server: %v", hhdls[k].URL)
}
if err != nil {
if *state == "dead" {
log.Printf("INFO: Server %s is dead. Assuming old master.", hhdls[k].URL)
master = hhdls[k]
continue
}
log.Fatalln("ERROR: Error when establishing initial connection to host", err)
}
defer hhdls[k].Conn.Close()
if *verbose {
log.Printf("DEBUG: Checking if server %s is slave", hhdls[k].URL)
}
ss, err := dbhelper.GetSlaveStatus(hhdls[k].Conn)
if ss.Master_Host != "" {
log.Printf("INFO : Server %s is configured as a slave", hhdls[k].URL)
slave = append(slave, hhdls[k])
slaveCount++
} else {
log.Printf("INFO : Server %s is not a slave. Assuming master status.", hhdls[k].URL)
master = hhdls[k]
}
}
if (hostCount - slaveCount) == 0 {
log.Fatalln("ERROR: Multi-master topologies are not yet supported.")
}
for _, sl := range slave {
if *verbose {
log.Printf("DEBUG: Checking if server %s is a slave of server %s", sl.Host, master.Host)
}
if dbhelper.IsSlaveof(sl.Conn, sl.Host, master.IP) == false {
log.Fatalf("ERROR: Server %s is not a slave of declared master %s", master.URL, master.Host)
}
}
// Check if preferred master is included in Host List
ret := func() bool {
for _, v := range hostList {
if v == *prefMaster {
return true
}
}
return false
}
if ret() == false && *prefMaster != "" {
log.Fatal("ERROR: Preferred master is not included in the hosts option")
}
// Do failover or switchover interactively, else start the interactive monitor.
if *state == "dead" {
master.failover()
} else if *interactive == false {
master.switchover()
} else {
MainLoop:
err := termbox.Init()
if err != nil {
log.Fatalln("Termbox initialization error", err)
}
termboxChan := new_tb_chan()
interval := time.Second
ticker := time.NewTicker(interval * 3)
var command string
for exit == false {
select {
case <-ticker.C:
drawHeader()
master.refresh()
master.drawMaster()
vy = 6
for k, _ := range slave {
slave[k].refresh()
slave[k].drawSlave(&vy)
}
drawFooter(&vy)
termbox.Flush()
case event := <-termboxChan:
switch event.Type {
case termbox.EventKey:
if event.Key == termbox.KeyCtrlS {
command = "switchover"
exit = true
}
if event.Key == termbox.KeyCtrlF {
command = "failover"
exit = true
}
if event.Key == termbox.KeyCtrlQ {
exit = true
}
}
switch event.Ch {
case 's':
termbox.Sync()
}
}
}
termbox.Close()
switch command {
case "switchover":
nmUrl, nsKey := master.switchover()
if nmUrl != "" && nsKey >= 0 {
if *verbose {
log.Printf("DEBUG: Reinstancing new master: %s and new slave: %s [%d]", nmUrl, slave[nsKey].URL, nsKey)
}
master, err = newServerMonitor(nmUrl)
slave[nsKey], err = newServerMonitor(slave[nsKey].URL)
}
log.Println("###### Restarting monitor console in 5 seconds. Press Ctrl-C to exit")
time.Sleep(5 * time.Second)
exit = false
goto MainLoop
case "failover":
nmUrl, _ := master.failover()
if nmUrl != "" {
if *verbose {
log.Printf("DEBUG: Reinstancing new master: %s", nmUrl)
}
master, err = newServerMonitor(nmUrl)
}
log.Println("###### Restarting monitor console in 5 seconds. Press Ctrl-C to exit")
time.Sleep(5 * time.Second)
exit = false
goto MainLoop
}
}
}
func main() {
var errLog = mysql.Logger(log.New(ioutil.Discard, "", 0))
mysql.SetLogger(errLog)
flag.Parse()
if version == true {
fmt.Println("MariaDB Replication Manager version", repmgrVersion)
os.Exit(0)
}
if logfile != "" {
var err error
logPtr, err = os.Create(logfile)
if err != nil {
log.Println("ERROR: Error opening logfile, disabling for the rest of the session.")
logfile = ""
}
}
// if slaves option has been supplied, split into a slice.
if hosts != "" {
hostList = strings.Split(hosts, ",")
} else {
log.Fatal("ERROR: No hosts list specified.")
}
// validate users.
if user == "" {
log.Fatal("ERROR: No master user/pair specified.")
}
dbUser, dbPass = splitPair(user)
if rpluser == "" {
log.Fatal("ERROR: No replication user/pair specified.")
}
rplUser, rplPass = splitPair(rpluser)
// Check that failover and switchover modes are set correctly.
if switchover == "" && failover == "" {
log.Fatal("ERROR: None of the switchover or failover modes are set.")
}
if switchover != "" && failover != "" {
log.Fatal("ERROR: Both switchover and failover modes are set.")
}
if !contains(failOptions, failover) && failover != "" {
log.Fatalf("ERROR: Incorrect failover mode: %s", failover)
}
if !contains(switchOptions, switchover) && switchover != "" {
log.Fatalf("ERROR: Incorrect switchover mode: %s", switchover)
}
// Forced failover implies interactive == false
if failover == "force" && interactive == true {
interactive = false
}
if ignoreSrv != "" {
ignoreList = strings.Split(ignoreSrv, ",")
}
// Create a connection to each host and build list of slaves.
hostCount := len(hostList)
servers = make([]*ServerMonitor, hostCount)
slaveCount := 0
for k, url := range hostList {
var err error
servers[k], err = newServerMonitor(url)
if verbose {
log.Printf("DEBUG: Creating new server: %v", servers[k].URL)
}
if err != nil {
if driverErr, ok := err.(*mysql.MySQLError); ok {
if driverErr.Number == 1045 {
log.Fatalln("ERROR: Database access denied:", err.Error())
}
}
if verbose {
log.Println("ERROR:", err)
}
log.Printf("INFO : Server %s is dead.", servers[k].URL)
servers[k].State = stateFailed
continue
}
defer servers[k].Conn.Close()
if verbose {
log.Printf("DEBUG: Checking if server %s is slave", servers[k].URL)
}
servers[k].refresh()
if servers[k].UsingGtid != "" {
if verbose {
log.Printf("DEBUG: Server %s is configured as a slave", servers[k].URL)
}
servers[k].State = stateSlave
slaves = append(slaves, servers[k])
slaveCount++
} else {
if verbose {
log.Printf("DEBUG: Server %s is not a slave. Setting aside", servers[k].URL)
}
servers[k].State = stateUnconn
}
}
// If no slaves are detected, then bail out
if len(slaves) == 0 {
log.Fatal("ERROR: No slaves were detected.")
}
// Check that all slave servers have the same master.
for _, sl := range slaves {
if sl.hasSiblings(slaves) == false {
log.Fatalln("ERROR: Multi-master topologies are not yet supported.")
}
}
// Check user privileges on live servers
for _, sv := range servers {
if sv.State != stateFailed {
priv, err := dbhelper.GetPrivileges(sv.Conn, dbUser, sv.Host)
if err != nil {
log.Fatalf("ERROR: Error getting privileges for user %s on host %s: %s", dbUser, sv.Host, err)
}
if priv.Repl_client_priv == "N" {
log.Fatalln("ERROR: User must have REPLICATION_CLIENT privilege")
} else if priv.Repl_slave_priv == "N" {
log.Fatalln("ERROR: User must have REPLICATION_SLAVE privilege")
} else if priv.Super_priv == "N" {
log.Fatalln("ERROR: User must have SUPER privilege")
}
}
}
// Depending if we are doing a failover or a switchover, we will find the master in the list of
// dead hosts or unconnected hosts.
if switchover != "" || failover == "monitor" {
// 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 s.ServerID == sid {
master = servers[k]
master.State = stateMaster
if verbose {
log.Printf("DEBUG: Server %s was autodetected as a master", s.URL)
}
break
}
}
}
} else {
// Slave master_host variable must point to dead master
smh := slaves[0].MasterHost
for k, s := range servers {
if s.State == stateFailed {
if s.Host == smh || s.IP == smh {
master = servers[k]
master.State = stateMaster
if verbose {
log.Printf("DEBUG: Server %s was autodetected as a master", s.URL)
}
break
}
}
}
}
// Final check if master has been found
if master == nil {
if switchover != "" || failover == "monitor" {
log.Fatalln("ERROR: Could not autodetect a master!")
} else {
log.Fatalln("ERROR: Could not autodetect a failed master!")
}
}
for _, sl := range slaves {
if verbose {
log.Printf("DEBUG: Checking if server %s is a slave of server %s", sl.Host, master.Host)
}
if dbhelper.IsSlaveof(sl.Conn, sl.Host, master.IP) == false {
log.Printf("WARN : Server %s is not a slave of declared master %s", master.URL, master.Host)
}
}
// Check if preferred master is included in Host List
ret := func() bool {
for _, v := range hostList {
if v == prefMaster {
return true
}
}
return false
}
if ret() == false && prefMaster != "" {
log.Fatal("ERROR: Preferred master is not included in the hosts option")
}
// Do failover or switchover manually, or start the interactive monitor.
if failover == "force" {
masterFailover(true)
} else if switchover != "" && interactive == false {
masterFailover(false)
} else {
err := termbox.Init()
if err != nil {
log.Fatalln("Termbox initialization error", err)
}
_, termlength = termbox.Size()
loglen := termlength - 9 - (hostCount * 3)
tlog = NewTermLog(loglen)
if failover != "" {
tlog.Add("Monitor started in failover mode")
} else {
tlog.Add("Monitor started in switchover mode")
}
termboxChan := newTbChan()
interval := time.Second
ticker := time.NewTicker(interval * 1)
for exit == false {
select {
case <-ticker.C:
display()
case event := <-termboxChan:
switch event.Type {
case termbox.EventKey:
if event.Key == termbox.KeyCtrlS {
if master.State != stateFailed || failCount > 0 {
masterFailover(false)
} else {
logprint("ERROR: Master failed, cannot initiate switchover")
}
}
if event.Key == termbox.KeyCtrlF {
if master.State == stateFailed {
masterFailover(true)
} else {
logprint("ERROR: Master not failed, cannot initiate failover")
}
}
if event.Key == termbox.KeyCtrlD {
for k, v := range servers {
logprint("Servers", k, v)
}
logprint("Master", master)
for k, v := range slaves {
logprint("Slaves", k, v)
}
}
if event.Key == termbox.KeyCtrlR {
logprint("INFO: Setting slaves read-only")
for _, sl := range slaves {
dbhelper.SetReadOnly(sl.Conn, true)
}
}
if event.Key == termbox.KeyCtrlW {
logprint("INFO: Setting slaves read-write")
for _, sl := range slaves {
dbhelper.SetReadOnly(sl.Conn, false)
}
}
if event.Key == termbox.KeyCtrlQ {
exit = true
}
}
switch event.Ch {
case 's':
termbox.Sync()
}
}
if master.State == stateFailed && interactive == false {
rem := (failoverTs + failtime) - time.Now().Unix()
if (failtime == 0) || (failtime > 0 && (rem <= 0 || failoverCtr == 0)) {
masterFailover(true)
if failoverCtr == faillimit {
exitMsg = "INFO : Failover limit reached. Exiting on failover completion."
exit = true
}
} else if failtime > 0 && rem%10 == 0 {
logprintf("WARN : Failover time limit enforced. Next failover available in %d seconds.", rem)
}
}
}
termbox.Close()
if exitMsg != "" {
log.Println(exitMsg)
}
}
}
func main() {
flag.Parse()
if *version == true {
fmt.Println("MariaDB Replication Manager version", repmgrVersion)
}
// if slaves option has been supplied, split into a slice.
if *hosts != "" {
hostList = strings.Split(*hosts, ",")
} else {
log.Fatal("ERROR: No hosts list specified.")
}
// validate users.
if *user == "" {
log.Fatal("ERROR: No master user/pair specified.")
}
dbUser, dbPass = splitPair(*user)
if *rpluser == "" {
log.Fatal("ERROR: No replication user/pair specified.")
}
rplUser, rplPass = splitPair(*rpluser)
// Check that failover and switchover modes are set correctly.
if *switchover == "" && *failover == "" {
log.Fatal("ERROR: None of the switchover or failover modes are set.")
}
if *switchover != "" && *failover != "" {
log.Fatal("ERROR: Both switchover and failover modes are set.")
}
if !contains(failOptions, *failover) && *failover != "" {
log.Fatalf("ERROR: Incorrect failover mode: %s", *failover)
}
if !contains(switchOptions, *switchover) && *switchover != "" {
log.Fatalf("ERROR: Incorrect switchover mode: %s", *switchover)
}
if *ignoreSrv != "" {
ignoreList = strings.Split(*ignoreSrv, ",")
}
// Create a connection to each host and build list of slaves.
hostCount := len(hostList)
servers = make([]*ServerMonitor, hostCount)
slaveCount := 0
for k, url := range hostList {
var err error
servers[k], err = newServerMonitor(url)
if *verbose {
log.Printf("DEBUG: Creating new server: %v", servers[k].URL)
}
if err != nil {
log.Printf("INFO : Server %s is dead.", servers[k].URL)
servers[k].State = STATE_FAILED
continue
}
defer servers[k].Conn.Close()
if *verbose {
log.Printf("DEBUG: Checking if server %s is slave", servers[k].URL)
}
servers[k].refresh()
if servers[k].UsingGtid != "" {
if *verbose {
log.Printf("DEBUG: Server %s is configured as a slave", servers[k].URL)
}
servers[k].State = STATE_SLAVE
slaves = append(slaves, servers[k])
slaveCount++
} else {
if *verbose {
log.Printf("DEBUG: Server %s is not a slave. Setting aside", servers[k].URL)
}
}
}
// Check that all slave servers have the same master.
for _, sl := range slaves {
if sl.hasSiblings(slaves) == false {
log.Fatalln("ERROR: Multi-master topologies are not yet supported.")
}
}
// Depending if we are doing a failover or a switchover, we will find the master in the list of
// dead hosts or unconnected hosts.
if *switchover != "" || *failover == "monitor" {
// 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 == STATE_UNCONN {
if s.ServerId == sid {
master = servers[k]
master.State = STATE_MASTER
if *verbose {
log.Printf("DEBUG: Server %s was autodetected as a master", s.URL)
}
break
}
}
}
} else {
// Slave master_host variable must point to dead master
smh := slaves[0].MasterHost
for k, s := range servers {
if s.State == STATE_FAILED {
if s.Host == smh || s.IP == smh {
master = servers[k]
master.State = STATE_MASTER
if *verbose {
log.Printf("DEBUG: Server %s was autodetected as a master", s.URL)
}
break
}
}
}
}
// Final check if master has been found
if master == nil {
log.Fatalln("ERROR: Could not autodetect a master!")
}
for _, sl := range slaves {
if *verbose {
log.Printf("DEBUG: Checking if server %s is a slave of server %s", sl.Host, master.Host)
}
if dbhelper.IsSlaveof(sl.Conn, sl.Host, master.IP) == false {
log.Printf("WARN : Server %s is not a slave of declared master %s", master.URL, master.Host)
}
}
// Check if preferred master is included in Host List
ret := func() bool {
for _, v := range hostList {
if v == *prefMaster {
return true
}
}
return false
}
if ret() == false && *prefMaster != "" {
log.Fatal("ERROR: Preferred master is not included in the hosts option")
}
// Do failover or switchover manually, or start the interactive monitor.
if *failover == "force" {
master.failover()
} else if *switchover != "" && *interactive == false {
master.switchover()
} else {
MainLoop:
err := termbox.Init()
if err != nil {
log.Fatalln("Termbox initialization error", err)
}
tlog = NewTermLog(20)
if *failover != "" {
tlog.Add("Monitor started in failover mode")
} else {
tlog.Add("Monitor started in switchover mode")
}
termboxChan := new_tb_chan()
interval := time.Second
ticker := time.NewTicker(interval * 3)
var command string
for exit == false {
select {
case <-ticker.C:
display()
case event := <-termboxChan:
switch event.Type {
case termbox.EventKey:
if event.Key == termbox.KeyCtrlS {
nmUrl, nsKey := master.switchover()
if nmUrl != "" && nsKey >= 0 {
if *verbose {
logprintf("DEBUG: Reinstancing new master: %s and new slave: %s [%d]", nmUrl, slaves[nsKey].URL, nsKey)
}
master, err = newServerMonitor(nmUrl)
slaves[nsKey], err = newServerMonitor(slaves[nsKey].URL)
}
}
if event.Key == termbox.KeyCtrlF {
command = "failover"
exit = true
}
if event.Key == termbox.KeyCtrlQ {
exit = true
}
}
switch event.Ch {
case 's':
termbox.Sync()
}
}
if master.State == STATE_FAILED && *interactive == false {
command = "failover"
exit = true
}
}
switch command {
case "failover":
termbox.Close()
nmUrl, nmKey := master.failover()
if nmUrl != "" {
if *verbose {
log.Printf("DEBUG: Reinstancing new master: %s", nmUrl)
}
master, err = newServerMonitor(nmUrl)
// Remove new master from slave slice
slaves = append(slaves[:nmKey], slaves[nmKey+1:]...)
}
log.Println("###### Restarting monitor console in 5 seconds. Press Ctrl-C to exit")
time.Sleep(5 * time.Second)
exit = false
goto MainLoop
}
termbox.Close()
}
}