// GetCandidateSiblingOfIntermediateMaster chooses the best sibling of a dead intermediate master
// to whom the IM's slaves can be moved.
func GetCandidateSiblingOfIntermediateMaster(intermediateMasterInstance *inst.Instance) (*inst.Instance, error) {
siblings, err := inst.ReadSlaveInstances(&intermediateMasterInstance.MasterKey)
if err != nil {
return nil, err
}
if len(siblings) <= 1 {
return nil, log.Errorf("topology_recovery: no siblings found for %+v", intermediateMasterInstance.Key)
}
sort.Sort(sort.Reverse(InstancesByCountSlaves(siblings)))
// In the next series of steps we attempt to return a good replacement.
// None of the below attempts is sure to pick a winning server. Perhaps picked server is not enough up-todate -- but
// this has small likelihood in the general case, and, well, it's an attempt. It's a Plan A, but we have Plan B & C if this fails.
// At first, we try to return an "is_candidate" server in same dc & env
log.Infof("topology_recovery: searching for the best candidate sibling of dead intermediate master")
for _, sibling := range siblings {
sibling := sibling
if isValidAsCandidateSiblingOfIntermediateMaster(intermediateMasterInstance, sibling) &&
sibling.IsCandidate &&
sibling.DataCenter == intermediateMasterInstance.DataCenter &&
sibling.PhysicalEnvironment == intermediateMasterInstance.PhysicalEnvironment {
log.Infof("topology_recovery: found %+v as the ideal candidate", sibling.Key)
return sibling, nil
}
}
// Go for something else in the same DC & ENV
for _, sibling := range siblings {
sibling := sibling
if isValidAsCandidateSiblingOfIntermediateMaster(intermediateMasterInstance, sibling) &&
sibling.DataCenter == intermediateMasterInstance.DataCenter &&
sibling.PhysicalEnvironment == intermediateMasterInstance.PhysicalEnvironment {
log.Infof("topology_recovery: found %+v as a replacement in same dc & environment", sibling.Key)
return sibling, nil
}
}
// Nothing in same DC & env, let's just go for some is_candidate
for _, sibling := range siblings {
sibling := sibling
if isValidAsCandidateSiblingOfIntermediateMaster(intermediateMasterInstance, sibling) && sibling.IsCandidate {
log.Infof("topology_recovery: found %+v as a good candidate", sibling.Key)
return sibling, nil
}
}
// Havent found an "is_candidate". Just whatever is valid.
for _, sibling := range siblings {
sibling := sibling
if isValidAsCandidateSiblingOfIntermediateMaster(intermediateMasterInstance, sibling) {
log.Infof("topology_recovery: found %+v as a replacement", sibling.Key)
return sibling, nil
}
}
return nil, log.Errorf("topology_recovery: cannot find candidate sibling of %+v", intermediateMasterInstance.Key)
}
// StopSlave stops replication on a given instance
func StopSlave(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if !instance.IsSlave() {
return instance, fmt.Errorf("instance is not a slave: %+v", instanceKey)
}
_, err = ExecInstanceNoPrepare(instanceKey, `stop slave`)
if err != nil {
// Patch; current MaxScale behavior for STOP SLAVE is to throw an error if slave already stopped.
if instance.isMaxScale() && err.Error() == "Error 1199: Slave connection is not running" {
err = nil
}
}
if err != nil {
return instance, log.Errore(err)
}
instance, err = ReadTopologyInstance(instanceKey)
log.Infof("Stopped slave on %+v, Self:%+v, Exec:%+v", *instanceKey, instance.SelfBinlogCoordinates, instance.ExecBinlogCoordinates)
return instance, err
}
// ResetSlave resets a slave, breaking the replication
func ResetSlave(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if instance.SlaveRunning() {
return instance, fmt.Errorf("Cannot reset slave on: %+v because slave is running", instanceKey)
}
if *config.RuntimeCLIFlags.Noop {
return instance, fmt.Errorf("noop: aborting reset-slave operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
// MySQL's RESET SLAVE is done correctly; however SHOW SLAVE STATUS still returns old hostnames etc
// and only resets till after next restart. This leads to orchestrator still thinking the instance replicates
// from old host. We therefore forcibly modify the hostname.
// RESET SLAVE ALL command solves this, but only as of 5.6.3
_, err = ExecInstanceNoPrepare(instanceKey, `change master to master_host='_'`)
if err != nil {
return instance, log.Errore(err)
}
_, err = ExecInstanceNoPrepare(instanceKey, `reset slave /*!50603 all */`)
if err != nil {
return instance, log.Errore(err)
}
log.Infof("Reset slave %+v", instanceKey)
instance, err = ReadTopologyInstance(instanceKey)
return instance, err
}
// DetachSlave detaches a slave from replication; forcibly corrupting the binlog coordinates (though in such way
// that is reversible)
func DetachSlave(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if instance.SlaveRunning() {
return instance, fmt.Errorf("Cannot detach slave on: %+v because slave is running", instanceKey)
}
isDetached, _, _ := instance.ExecBinlogCoordinates.DetachedCoordinates()
if isDetached {
return instance, fmt.Errorf("Cannot (need not) detach slave on: %+v because slave is already detached", instanceKey)
}
if *config.RuntimeCLIFlags.Noop {
return instance, fmt.Errorf("noop: aborting detach-slave operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
detachedCoordinates := BinlogCoordinates{LogFile: fmt.Sprintf("//%s:%d", instance.ExecBinlogCoordinates.LogFile, instance.ExecBinlogCoordinates.LogPos), LogPos: instance.ExecBinlogCoordinates.LogPos}
// Encode the current coordinates within the log file name, in such way that replication is broken, but info can still be resurrected
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf(`change master to master_log_file='%s', master_log_pos=%d`, detachedCoordinates.LogFile, detachedCoordinates.LogPos))
if err != nil {
return instance, log.Errore(err)
}
log.Infof("Detach slave %+v", instanceKey)
instance, err = ReadTopologyInstance(instanceKey)
return instance, err
}
// ReattachSlave restores a detached slave back into replication
func ReattachSlave(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if instance.SlaveRunning() {
return instance, fmt.Errorf("Cannot (need not) reattach slave on: %+v because slave is running", instanceKey)
}
isDetached, detachedLogFile, detachedLogPos := instance.ExecBinlogCoordinates.DetachedCoordinates()
if !isDetached {
return instance, fmt.Errorf("Cannot reattach slave on: %+v because slave is not detached", instanceKey)
}
if *config.RuntimeCLIFlags.Noop {
return instance, fmt.Errorf("noop: aborting reattach-slave operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf(`change master to master_log_file='%s', master_log_pos=%s`, detachedLogFile, detachedLogPos))
if err != nil {
return instance, log.Errore(err)
}
log.Infof("Reattach slave %+v", instanceKey)
instance, err = ReadTopologyInstance(instanceKey)
return instance, err
}
// read reads configuration from given file, or silently skips if the file does not exist.
// If the file does exist, then it is expected to be in valid JSON format or the function bails out.
func read(file_name string) (*Configuration, error) {
file, err := os.Open(file_name)
if err == nil {
decoder := json.NewDecoder(file)
err := decoder.Decode(Config)
if err == nil {
log.Infof("Read config: %s", file_name)
} else {
log.Fatal("Cannot read config file:", file_name, err)
}
postReadAdjustments()
}
return Config, err
}
// FlushBinaryLogsTo attempts to 'PURGE BINARY LOGS' until given binary log is reached
func PurgeBinaryLogsTo(instanceKey *InstanceKey, logFile string) (*Instance, error) {
if *config.RuntimeCLIFlags.Noop {
return nil, fmt.Errorf("noop: aborting purge-binary-logs operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
_, err := ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("purge binary logs to '%s'", logFile))
if err != nil {
return nil, log.Errore(err)
}
log.Infof("purge-binary-logs to=%+v on %+v", logFile, *instanceKey)
AuditOperation("purge-binary-logs", instanceKey, "success")
return ReadTopologyInstance(instanceKey)
}
// MasterPosWait issues a MASTER_POS_WAIT() an given instance according to given coordinates.
func MasterPosWait(instanceKey *InstanceKey, binlogCoordinates *BinlogCoordinates) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
_, err = ExecInstance(instanceKey, `select master_pos_wait(?, ?)`, binlogCoordinates.LogFile, binlogCoordinates.LogPos)
if err != nil {
return instance, log.Errore(err)
}
log.Infof("Instance %+v has reached coordinates: %+v", instanceKey, binlogCoordinates)
instance, err = ReadTopologyInstance(instanceKey)
return instance, err
}
// StopSlaveNicely stops a slave such that SQL_thread and IO_thread are aligned (i.e.
// SQL_thread consumes all relay log entries)
// It will actually START the sql_thread even if the slave is completely stopped.
func StopSlaveNicely(instanceKey *InstanceKey, timeout time.Duration) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if !instance.IsSlave() {
return instance, fmt.Errorf("instance is not a slave: %+v", instanceKey)
}
_, err = ExecInstanceNoPrepare(instanceKey, `stop slave io_thread`)
_, err = ExecInstanceNoPrepare(instanceKey, `start slave sql_thread`)
if instance.SQLDelay == 0 {
// Otherwise we don't bother.
startTime := time.Now()
for upToDate := false; !upToDate; {
if timeout > 0 && time.Since(startTime) >= timeout {
// timeout
return nil, log.Errorf("StopSlaveNicely timeout on %+v", *instanceKey)
}
instance, err = ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if instance.SQLThreadUpToDate() {
upToDate = true
} else {
time.Sleep(sqlThreadPollDuration)
}
}
}
_, err = ExecInstanceNoPrepare(instanceKey, `stop slave`)
if err != nil {
// Patch; current MaxScale behavior for STOP SLAVE is to throw an error if slave already stopped.
if instance.isMaxScale() && err.Error() == "Error 1199: Slave connection is not running" {
err = nil
}
}
if err != nil {
return instance, log.Errore(err)
}
instance, err = ReadTopologyInstance(instanceKey)
log.Infof("Stopped slave nicely on %+v, Self:%+v, Exec:%+v", *instanceKey, instance.SelfBinlogCoordinates, instance.ExecBinlogCoordinates)
return instance, err
}
// If a mysql port is available, try to discover against it
func DiscoverAgentInstance(hostname string, port int) error {
agent, err := GetAgent(hostname)
if err != nil {
log.Errorf("Couldn't get agent for %s: %v", hostname, err)
return err
}
instanceKey := agent.GetInstance()
instance, err := inst.ReadTopologyInstance(instanceKey)
if err != nil {
log.Errorf("Failed to read topology for %v", instanceKey)
return err
}
log.Infof("Discovered Agent Instance: %v", instance.Key)
return nil
}
// StartSlaveUntilMasterCoordinates issuesa START SLAVE UNTIL... statement on given instance
func StartSlaveUntilMasterCoordinates(instanceKey *InstanceKey, masterCoordinates *BinlogCoordinates) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if !instance.IsSlave() {
return instance, fmt.Errorf("instance is not a slave: %+v", instanceKey)
}
if instance.SlaveRunning() {
return instance, fmt.Errorf("slave already running: %+v", instanceKey)
}
log.Infof("Will start slave on %+v until coordinates: %+v", instanceKey, masterCoordinates)
// MariaDB has a bug: a CHANGE MASTER TO statement does not work properly with prepared statement... :P
// See https://mariadb.atlassian.net/browse/MDEV-7640
// This is the reason for ExecInstanceNoPrepare
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("start slave until master_log_file='%s', master_log_pos=%d",
masterCoordinates.LogFile, masterCoordinates.LogPos))
if err != nil {
return instance, log.Errore(err)
}
for upToDate := false; !upToDate; {
instance, err = ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
switch {
case instance.ExecBinlogCoordinates.SmallerThan(masterCoordinates):
time.Sleep(sqlThreadPollDuration)
case instance.ExecBinlogCoordinates.Equals(masterCoordinates):
upToDate = true
case masterCoordinates.SmallerThan(&instance.ExecBinlogCoordinates):
return instance, fmt.Errorf("Start SLAVE UNTIL is past coordinates: %+v", instanceKey)
}
}
instance, err = StopSlave(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
return instance, err
}
// FlushBinaryLogs attempts a 'FLUSH BINARY LOGS' statement on the given instance.
func FlushBinaryLogs(instanceKey *InstanceKey, count int) (*Instance, error) {
if *config.RuntimeCLIFlags.Noop {
return nil, fmt.Errorf("noop: aborting flush-binary-logs operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
for i := 0; i < count; i++ {
_, err := ExecInstance(instanceKey, `flush binary logs`)
if err != nil {
return nil, log.Errore(err)
}
}
log.Infof("flush-binary-logs count=%+v on %+v", count, *instanceKey)
AuditOperation("flush-binary-logs", instanceKey, "success")
return ReadTopologyInstance(instanceKey)
}
// executeProcesses executes a list of processes
func executeProcesses(processes []string, description string, topologyRecovery *TopologyRecovery, failOnError bool) error {
var err error
for _, command := range processes {
command := replaceCommandPlaceholders(command, topologyRecovery)
if cmdErr := os.CommandRun(command); cmdErr == nil {
log.Infof("Executed %s command: %s", description, command)
} else {
if err == nil {
// Note first error
err = cmdErr
}
log.Errorf("Failed to execute %s command: %s", description, command)
if failOnError {
return err
}
}
}
return err
}
// SetReadOnly sets or clears the instance's global read_only variable
func SetReadOnly(instanceKey *InstanceKey, readOnly bool) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if *config.RuntimeCLIFlags.Noop {
return instance, fmt.Errorf("noop: aborting set-read-only operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
_, err = ExecInstance(instanceKey, fmt.Sprintf("set global read_only = %t", readOnly))
if err != nil {
return instance, log.Errore(err)
}
instance, err = ReadTopologyInstance(instanceKey)
log.Infof("instance %+v read_only: %t", instanceKey, readOnly)
AuditOperation("read-only", instanceKey, fmt.Sprintf("set as %t", readOnly))
return instance, err
}
// StartSlave starts replication on a given instance
func StartSlave(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if !instance.IsSlave() {
return instance, fmt.Errorf("instance is not a slave: %+v", instanceKey)
}
_, err = ExecInstanceNoPrepare(instanceKey, `start slave`)
if err != nil {
return instance, log.Errore(err)
}
log.Infof("Started slave on %+v", instanceKey)
if config.Config.SlaveStartPostWaitMilliseconds > 0 {
time.Sleep(time.Duration(config.Config.SlaveStartPostWaitMilliseconds) * time.Millisecond)
}
instance, err = ReadTopologyInstance(instanceKey)
return instance, err
}
// ContinuousAgentsPoll starts an asynchronuous infinite process where agents are
// periodically investigated and their status captured, and long since unseen agents are
// purged and forgotten.
func ContinuousAgentsPoll() {
log.Infof("Starting continuous agents poll")
go discoverSeededAgents()
tick := time.Tick(time.Duration(config.Config.DiscoveryPollSeconds) * time.Second)
caretakingTick := time.Tick(time.Hour)
for range tick {
agentsHosts, _ := agent.ReadOutdatedAgentsHosts()
log.Debugf("outdated agents hosts: %+v", agentsHosts)
for _, hostname := range agentsHosts {
go pollAgent(hostname)
}
// See if we should also forget agents (lower frequency)
select {
case <-caretakingTick:
agent.ForgetLongUnseenAgents()
agent.FailStaleSeeds()
default:
}
}
}
// ResetMaster issues a RESET MASTER statement on given instance. Use with extreme care!
func ResetMaster(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if instance.SlaveRunning() {
return instance, fmt.Errorf("Cannot reset master on: %+v because slave is running", instanceKey)
}
if *config.RuntimeCLIFlags.Noop {
return instance, fmt.Errorf("noop: aborting reset-master operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
_, err = ExecInstanceNoPrepare(instanceKey, `reset master`)
if err != nil {
return instance, log.Errore(err)
}
log.Infof("Reset master %+v", instanceKey)
instance, err = ReadTopologyInstance(instanceKey)
return instance, err
}
// KillQuery stops replication on a given instance
func KillQuery(instanceKey *InstanceKey, process int64) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if *config.RuntimeCLIFlags.Noop {
return instance, fmt.Errorf("noop: aborting kill-query operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
_, err = ExecInstance(instanceKey, fmt.Sprintf(`kill query %d`, process))
if err != nil {
return instance, log.Errore(err)
}
instance, err = ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
log.Infof("Killed query on %+v", *instanceKey)
AuditOperation("kill-query", instanceKey, fmt.Sprintf("Killed query %d", process))
return instance, err
}
// replacePromotedSlaveWithCandidate is called after an intermediate master has died and been replaced by some promotedSlave.
// But, is there an even better slave to promote?
// if candidateInstanceKey is given, then it is forced to be promoted over the promotedSlave
// Otherwise, search for the best to promote!
func replacePromotedSlaveWithCandidate(deadInstanceKey *inst.InstanceKey, promotedSlave *inst.Instance, candidateInstanceKey *inst.InstanceKey) (*inst.Instance, error) {
candidateSlaves, _ := inst.ReadClusterCandidateInstances(promotedSlave.ClusterName)
// So we've already promoted a slave.
// However, can we improve on our choice? Are there any slaves marked with "is_candidate"?
// Maybe we actually promoted such a slave. Does that mean we should keep it?
// The current logic is:
// - 1. we prefer to promote a "is_candidate" which is in the same DC & env as the dead intermediate master (or do nothing if the promtoed slave is such one)
// - 2. we prefer to promote a "is_candidate" which is in the same DC & env as the promoted slave (or do nothing if the promtoed slave is such one)
// - 3. keep to current choice
log.Infof("topology_recovery: checking if should replace promoted slave with a better candidate")
if candidateInstanceKey == nil {
if deadInstance, _, err := inst.ReadInstance(deadInstanceKey); err == nil && deadInstance != nil {
for _, candidateSlave := range candidateSlaves {
if promotedSlave.Key.Equals(&candidateSlave.Key) &&
promotedSlave.DataCenter == deadInstance.DataCenter &&
promotedSlave.PhysicalEnvironment == deadInstance.PhysicalEnvironment {
// Seems like we promoted a candidate in the same DC & ENV as dead IM! Ideal! We're happy!
log.Infof("topology_recovery: promoted slave %+v is the ideal candidate", promotedSlave.Key)
return promotedSlave, nil
}
}
}
}
// We didn't pick the ideal candidate; let's see if we can replace with a candidate from same DC and ENV
if candidateInstanceKey == nil {
// Try a candidate slave that is in same DC & env as the dead instance
if deadInstance, _, err := inst.ReadInstance(deadInstanceKey); err == nil && deadInstance != nil {
for _, candidateSlave := range candidateSlaves {
if candidateSlave.DataCenter == deadInstance.DataCenter &&
candidateSlave.PhysicalEnvironment == deadInstance.PhysicalEnvironment &&
candidateSlave.MasterKey.Equals(&promotedSlave.Key) {
// This would make a great candidate
candidateInstanceKey = &candidateSlave.Key
log.Debugf("topology_recovery: no candidate was offered for %+v but orchestrator picks %+v as candidate replacement, based on being in same DC & env as failed instance", promotedSlave.Key, candidateSlave.Key)
}
}
}
}
if candidateInstanceKey == nil {
// We cannot find a candidate in same DC and ENV as dead master
for _, candidateSlave := range candidateSlaves {
if promotedSlave.Key.Equals(&candidateSlave.Key) {
// Seems like we promoted a candidate slave (though not in same DC and ENV as dead master). Good enough.
// No further action required.
log.Infof("topology_recovery: promoted slave %+v is a good candidate", promotedSlave.Key)
return promotedSlave, nil
}
}
}
// Still nothing?
if candidateInstanceKey == nil {
// Try a candidate slave that is in same DC & env as the promoted slave (our promoted slave is not an "is_candidate")
for _, candidateSlave := range candidateSlaves {
if promotedSlave.DataCenter == candidateSlave.DataCenter &&
promotedSlave.PhysicalEnvironment == candidateSlave.PhysicalEnvironment &&
candidateSlave.MasterKey.Equals(&promotedSlave.Key) {
// OK, better than nothing
candidateInstanceKey = &candidateSlave.Key
log.Debugf("topology_recovery: no candidate was offered for %+v but orchestrator picks %+v as candidate replacement, based on being in same DC & env as promoted instance", promotedSlave.Key, candidateSlave.Key)
}
}
}
// So do we have a candidate?
if candidateInstanceKey == nil {
// Found nothing. Stick with promoted slave
return promotedSlave, nil
}
if promotedSlave.Key.Equals(candidateInstanceKey) {
// Sanity. It IS the candidate, nothing to promote...
return promotedSlave, nil
}
// Try and promote suggested candidate, if applicable and possible
log.Debugf("topology_recovery: promoted instance %+v is not the suggested candidate %+v. Will see what can be done", promotedSlave.Key, *candidateInstanceKey)
candidateInstance, _, err := inst.ReadInstance(candidateInstanceKey)
if err != nil {
return promotedSlave, log.Errore(err)
}
if candidateInstance.MasterKey.Equals(&promotedSlave.Key) {
log.Debugf("topology_recovery: suggested candidate %+v is slave of promoted instance %+v. Will try and enslave its master", *candidateInstanceKey, promotedSlave.Key)
candidateInstance, err = inst.EnslaveMaster(&candidateInstance.Key)
if err != nil {
return promotedSlave, log.Errore(err)
}
log.Debugf("topology_recovery: success promoting %+v over %+v", *candidateInstanceKey, promotedSlave.Key)
return candidateInstance, nil
}
log.Debugf("topology_recovery: could not manage to promoted suggested candidate %+v", *candidateInstanceKey)
return promotedSlave, nil
}
// ContinuousDiscovery starts an asynchronuous infinite discovery process where instances are
// periodically investigated and their status captured, and long since unseen instances are
// purged and forgotten.
func ContinuousDiscovery() {
if config.Config.DatabaselessMode__experimental {
log.Fatal("Cannot execute continuous mode in databaseless mode")
}
log.Infof("Starting continuous discovery")
inst.LoadHostnameResolveCache()
go handleDiscoveryRequests()
discoveryTick := time.Tick(time.Duration(config.Config.DiscoveryPollSeconds) * time.Second)
caretakingTick := time.Tick(time.Minute)
recoveryTick := time.Tick(time.Duration(config.Config.RecoveryPollSeconds) * time.Second)
var snapshotTopologiesTick <-chan time.Time
if config.Config.SnapshotTopologiesIntervalHours > 0 {
snapshotTopologiesTick = time.Tick(time.Duration(config.Config.SnapshotTopologiesIntervalHours) * time.Hour)
}
go ometrics.InitGraphiteMetrics()
go acceptSignals()
if *config.RuntimeCLIFlags.GrabElection {
process.GrabElection()
}
for {
select {
case <-discoveryTick:
go func() {
wasAlreadyElected := isElectedNode
if isElectedNode, _ = process.AttemptElection(); isElectedNode {
instanceKeys, _ := inst.ReadOutdatedInstanceKeys()
log.Debugf("outdated keys: %+v", instanceKeys)
for _, instanceKey := range instanceKeys {
instanceKey := instanceKey
go func() { discoveryInstanceKeys <- instanceKey }()
}
if !wasAlreadyElected {
// Just turned to be leader!
go process.RegisterNode("", "", false)
}
} else {
log.Debugf("Not elected as active node; polling")
}
}()
case <-caretakingTick:
// Various periodic internal maintenance tasks
go func() {
if isElectedNode {
go inst.ForgetLongUnseenInstances()
go inst.ForgetUnseenInstancesDifferentlyResolved()
go inst.ForgetExpiredHostnameResolves()
go inst.DeleteInvalidHostnameResolves()
go inst.ReviewUnseenInstances()
go inst.InjectUnseenMasters()
go inst.ResolveUnknownMasterHostnameResolves()
go inst.UpdateClusterAliases()
go inst.ExpireMaintenance()
go inst.ExpireDowntime()
go inst.ExpireCandidateInstances()
go inst.ExpireHostnameUnresolve()
go inst.ExpireClusterDomainName()
go inst.ExpireAudit()
go inst.ExpireMasterPositionEquivalence()
go inst.FlushNontrivialResolveCacheToDatabase()
go process.ExpireNodesHistory()
}
if !isElectedNode {
// Take this opportunity to refresh yourself
go inst.LoadHostnameResolveCache()
}
go inst.ReadClusterAliases()
}()
case <-recoveryTick:
go func() {
if isElectedNode {
go ClearActiveFailureDetections()
go ClearActiveRecoveries()
go ExpireBlockedRecoveries()
go AcknowledgeCrashedRecoveries()
go inst.ExpireInstanceAnalysisChangelog()
go CheckAndRecover(nil, nil, false)
}
}()
case <-snapshotTopologiesTick:
go func() {
go inst.SnapshotTopologies()
}()
}
}
}
// ChangeMasterTo changes the given instance's master according to given input.
func ChangeMasterTo(instanceKey *InstanceKey, masterKey *InstanceKey, masterBinlogCoordinates *BinlogCoordinates, skipUnresolve bool, gtidHint OperationGTIDHint) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if instance.SlaveRunning() {
return instance, fmt.Errorf("ChangeMasterTo: Cannot change master on: %+v because slave is running", *instanceKey)
}
log.Debugf("ChangeMasterTo: will attempt changing master on %+v to %+v, %+v", *instanceKey, *masterKey, *masterBinlogCoordinates)
changeToMasterKey := masterKey
if !skipUnresolve {
unresolvedMasterKey, nameUnresolved, err := UnresolveHostname(masterKey)
if err != nil {
log.Debugf("ChangeMasterTo: aborting operation on %+v due to resolving error on %+v: %+v", *instanceKey, *masterKey, err)
return instance, err
}
if nameUnresolved {
log.Debugf("ChangeMasterTo: Unresolved %+v into %+v", *masterKey, unresolvedMasterKey)
}
changeToMasterKey = &unresolvedMasterKey
}
if *config.RuntimeCLIFlags.Noop {
return instance, fmt.Errorf("noop: aborting CHANGE MASTER TO operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
originalMasterKey := instance.MasterKey
originalExecBinlogCoordinates := instance.ExecBinlogCoordinates
changedViaGTID := false
if instance.UsingMariaDBGTID && gtidHint != GTIDHintDeny {
// MariaDB has a bug: a CHANGE MASTER TO statement does not work properly with prepared statement... :P
// See https://mariadb.atlassian.net/browse/MDEV-7640
// This is the reason for ExecInstanceNoPrepare
// Keep on using GTID
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("change master to master_host='%s', master_port=%d",
changeToMasterKey.Hostname, changeToMasterKey.Port))
changedViaGTID = true
} else if instance.UsingMariaDBGTID && gtidHint == GTIDHintDeny {
// Make sure to not use GTID
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("change master to master_host='%s', master_port=%d, master_log_file='%s', master_log_pos=%d, master_use_gtid=no",
changeToMasterKey.Hostname, changeToMasterKey.Port, masterBinlogCoordinates.LogFile, masterBinlogCoordinates.LogPos))
} else if instance.IsMariaDB() && gtidHint == GTIDHintForce {
// Is MariaDB; not using GTID, turn into GTID
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("change master to master_host='%s', master_port=%d, master_use_gtid=slave_pos",
changeToMasterKey.Hostname, changeToMasterKey.Port))
changedViaGTID = true
} else if instance.UsingOracleGTID && gtidHint != GTIDHintDeny {
// Is Oracle; already uses GTID; keep using it.
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("change master to master_host='%s', master_port=%d",
changeToMasterKey.Hostname, changeToMasterKey.Port))
changedViaGTID = true
} else if instance.UsingOracleGTID && gtidHint == GTIDHintDeny {
// Is Oracle; already uses GTID
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("change master to master_host='%s', master_port=%d, master_log_file='%s', master_log_pos=%d, master_auto_position=0",
changeToMasterKey.Hostname, changeToMasterKey.Port, masterBinlogCoordinates.LogFile, masterBinlogCoordinates.LogPos))
} else if instance.SupportsOracleGTID && gtidHint == GTIDHintForce {
// Is Oracle; not using GTID right now; turn into GTID
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("change master to master_host='%s', master_port=%d, master_auto_position=1",
changeToMasterKey.Hostname, changeToMasterKey.Port))
changedViaGTID = true
} else {
// Normal binlog file:pos
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("change master to master_host='%s', master_port=%d, master_log_file='%s', master_log_pos=%d",
changeToMasterKey.Hostname, changeToMasterKey.Port, masterBinlogCoordinates.LogFile, masterBinlogCoordinates.LogPos))
}
if err != nil {
return instance, log.Errore(err)
}
WriteMasterPositionEquivalence(&originalMasterKey, &originalExecBinlogCoordinates, changeToMasterKey, masterBinlogCoordinates)
log.Infof("ChangeMasterTo: Changed master on %+v to: %+v, %+v. GTID: %+v", *instanceKey, masterKey, masterBinlogCoordinates, changedViaGTID)
instance, err = ReadTopologyInstance(instanceKey)
return instance, err
}
// Cli initiates a command line interface, executing requested command.
func Cli(command string, strict bool, instance string, destination string, owner string, reason string, duration string, pattern string, clusterAlias string, pool string, hostnameFlag string) {
if instance != "" && !strings.Contains(instance, ":") {
instance = fmt.Sprintf("%s:%d", instance, config.Config.DefaultInstancePort)
}
instanceKey, err := inst.ParseInstanceKey(instance)
if err != nil {
instanceKey = nil
}
rawInstanceKey, err := inst.NewRawInstanceKey(instance)
if err != nil {
rawInstanceKey = nil
}
if destination != "" && !strings.Contains(destination, ":") {
destination = fmt.Sprintf("%s:%d", destination, config.Config.DefaultInstancePort)
}
destinationKey, err := inst.ParseInstanceKey(destination)
if err != nil {
destinationKey = nil
}
if hostname, err := os.Hostname(); err == nil {
thisInstanceKey = &inst.InstanceKey{Hostname: hostname, Port: int(config.Config.DefaultInstancePort)}
}
postponedFunctionsContainer := inst.NewPostponedFunctionsContainer()
if len(owner) == 0 {
// get os username as owner
usr, err := user.Current()
if err != nil {
log.Fatale(err)
}
owner = usr.Username
}
inst.SetMaintenanceOwner(owner)
skipDatabaseCommands := false
switch command {
case "reset-internal-db-deployment":
skipDatabaseCommands = true
case "help":
skipDatabaseCommands = true
}
if !skipDatabaseCommands {
process.ContinuousRegistration(string(process.OrchestratorExecutionCliMode), command)
}
// begin commands
switch command {
// smart mode
case registerCliCommand("relocate", "Smart relocation", `Relocate a slave beneath another instance`), registerCliCommand("relocate-below", "Smart relocation", `Synonym to 'relocate', will be deprecated`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
if destinationKey == nil {
log.Fatal("Cannot deduce destination:", destination)
}
_, err := inst.RelocateBelow(instanceKey, destinationKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%s<%s", instanceKey.DisplayString(), destinationKey.DisplayString()))
}
case registerCliCommand("relocate-slaves", "Smart relocation", `Relocates all or part of the slaves of a given instance under another instance`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
if destinationKey == nil {
log.Fatal("Cannot deduce destination:", destination)
}
slaves, _, err, errs := inst.RelocateSlaves(instanceKey, destinationKey, pattern)
if err != nil {
log.Fatale(err)
} else {
for _, e := range errs {
log.Errore(e)
}
for _, slave := range slaves {
fmt.Println(slave.Key.DisplayString())
}
}
}
case registerCliCommand("regroup-slaves", "Smart relocation", `Given an instance, pick one of its slave and make it local master of its siblings`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
lostSlaves, equalSlaves, aheadSlaves, promotedSlave, err := inst.RegroupSlaves(instanceKey, false, func(candidateSlave *inst.Instance) { fmt.Println(candidateSlave.Key.DisplayString()) }, postponedFunctionsContainer)
postponedFunctionsContainer.InvokePostponed()
if promotedSlave == nil {
log.Fatalf("Could not regroup slaves of %+v; error: %+v", *instanceKey, err)
}
fmt.Println(fmt.Sprintf("%s lost: %d, trivial: %d, pseudo-gtid: %d",
promotedSlave.Key.DisplayString(), len(lostSlaves), len(equalSlaves), len(aheadSlaves)))
if err != nil {
log.Fatale(err)
}
}
// General replication commands
// move, binlog file:pos
case registerCliCommand("move-up", "Classic file:pos relocation", `Move a slave one level up the topology`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
instance, err := inst.MoveUp(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%s<%s", instanceKey.DisplayString(), instance.MasterKey.DisplayString()))
}
case registerCliCommand("move-up-slaves", "Classic file:pos relocation", `Moves slaves of the given instance one level up the topology`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
movedSlaves, _, err, errs := inst.MoveUpSlaves(instanceKey, pattern)
if err != nil {
log.Fatale(err)
} else {
for _, e := range errs {
log.Errore(e)
}
for _, slave := range movedSlaves {
fmt.Println(slave.Key.DisplayString())
}
}
}
case registerCliCommand("move-below", "Classic file:pos relocation", `Moves a slave beneath its sibling. Both slaves must be actively replicating from same master.`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
if destinationKey == nil {
log.Fatal("Cannot deduce destination/sibling:", destination)
}
_, err := inst.MoveBelow(instanceKey, destinationKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%s<%s", instanceKey.DisplayString(), destinationKey.DisplayString()))
}
case registerCliCommand("move-equivalent", "Classic file:pos relocation", `Moves a slave beneath another server, based on previously recorded "equivalence coordinates"`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
if destinationKey == nil {
log.Fatal("Cannot deduce destination:", destination)
}
_, err := inst.MoveEquivalent(instanceKey, destinationKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%s<%s", instanceKey.DisplayString(), destinationKey.DisplayString()))
}
case registerCliCommand("repoint", "Classic file:pos relocation", `Make the given instance replicate from another instance without changing the binglog coordinates. Use with care`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
// destinationKey can be null, in which case the instance repoints to its existing master
instance, err := inst.Repoint(instanceKey, destinationKey, inst.GTIDHintNeutral)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%s<%s", instanceKey.DisplayString(), instance.MasterKey.DisplayString()))
}
case registerCliCommand("repoint-slaves", "Classic file:pos relocation", `Repoint all slaves of given instance to replicate back from the instance. Use with care`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
repointedSlaves, err, errs := inst.RepointSlavesTo(instanceKey, pattern, destinationKey)
if err != nil {
log.Fatale(err)
} else {
for _, e := range errs {
log.Errore(e)
}
for _, slave := range repointedSlaves {
fmt.Println(fmt.Sprintf("%s<%s", slave.Key.DisplayString(), instanceKey.DisplayString()))
}
}
}
case registerCliCommand("enslave-siblings", "Classic file:pos relocation", `Turn all siblings of a slave into its sub-slaves.`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
_, _, err := inst.EnslaveSiblings(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("enslave-master", "Classic file:pos relocation", `Turn an instance into a master of its own master; essentially switch the two.`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
_, err := inst.EnslaveMaster(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("make-co-master", "Classic file:pos relocation", `Create a master-master replication. Given instance is a slave which replicates directly from a master.`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.MakeCoMaster(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("get-candidate-slave", "Classic file:pos relocation", `Information command suggesting the most up-to-date slave of a given instance that is good for promotion`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
instance, _, _, _, err := inst.GetCandidateSlave(instanceKey, false)
if err != nil {
log.Fatale(err)
} else {
fmt.Println(instance.Key.DisplayString())
}
}
case registerCliCommand("regroup-slaves-bls", "Binlog server relocation", `Regroup Binlog Server slaves of a given instance`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
_, promotedBinlogServer, err := inst.RegroupSlavesBinlogServers(instanceKey, false)
if promotedBinlogServer == nil {
log.Fatalf("Could not regroup binlog server slaves of %+v; error: %+v", *instanceKey, err)
}
fmt.Println(promotedBinlogServer.Key.DisplayString())
if err != nil {
log.Fatale(err)
}
}
// move, GTID
case registerCliCommand("move-gtid", "GTID relocation", `Move a slave beneath another instance.`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
if destinationKey == nil {
log.Fatal("Cannot deduce destination:", destination)
}
_, err := inst.MoveBelowGTID(instanceKey, destinationKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%s<%s", instanceKey.DisplayString(), destinationKey.DisplayString()))
}
case registerCliCommand("move-slaves-gtid", "GTID relocation", `Moves all slaves of a given instance under another (destination) instance using GTID`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
if destinationKey == nil {
log.Fatal("Cannot deduce destination:", destination)
}
movedSlaves, _, err, errs := inst.MoveSlavesGTID(instanceKey, destinationKey, pattern)
if err != nil {
log.Fatale(err)
} else {
for _, e := range errs {
log.Errore(e)
}
for _, slave := range movedSlaves {
fmt.Println(slave.Key.DisplayString())
}
}
}
case registerCliCommand("regroup-slaves-gtid", "GTID relocation", `Given an instance, pick one of its slave and make it local master of its siblings, using GTID.`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
lostSlaves, movedSlaves, promotedSlave, err := inst.RegroupSlavesGTID(instanceKey, false, func(candidateSlave *inst.Instance) { fmt.Println(candidateSlave.Key.DisplayString()) })
if promotedSlave == nil {
log.Fatalf("Could not regroup slaves of %+v; error: %+v", *instanceKey, err)
}
fmt.Println(fmt.Sprintf("%s lost: %d, moved: %d",
promotedSlave.Key.DisplayString(), len(lostSlaves), len(movedSlaves)))
if err != nil {
log.Fatale(err)
}
}
// Pseudo-GTID
case registerCliCommand("match", "Pseudo-GTID relocation", `Matches a slave beneath another (destination) instance using Pseudo-GTID`),
registerCliCommand("match-below", "Pseudo-GTID relocation", `Synonym to 'match', will be deprecated`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
if destinationKey == nil {
log.Fatal("Cannot deduce destination:", destination)
}
_, _, err := inst.MatchBelow(instanceKey, destinationKey, true)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%s<%s", instanceKey.DisplayString(), destinationKey.DisplayString()))
}
case registerCliCommand("match-up", "Pseudo-GTID relocation", `Transport the slave one level up the hierarchy, making it child of its grandparent, using Pseudo-GTID`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
instance, _, err := inst.MatchUp(instanceKey, true)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%s<%s", instanceKey.DisplayString(), instance.MasterKey.DisplayString()))
}
case registerCliCommand("rematch", "Pseudo-GTID relocation", `Reconnect a slave onto its master, via PSeudo-GTID.`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
instance, _, err := inst.RematchSlave(instanceKey, true)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%s<%s", instanceKey.DisplayString(), instance.MasterKey.DisplayString()))
}
case registerCliCommand("match-slaves", "Pseudo-GTID relocation", `Matches all slaves of a given instance under another (destination) instance using Pseudo-GTID`),
registerCliCommand("multi-match-slaves", "Pseudo-GTID relocation", `Synonym to 'match-slaves', will be deprecated`):
{
// Move all slaves of "instance" beneath "destination"
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
if destinationKey == nil {
log.Fatal("Cannot deduce destination:", destination)
}
matchedSlaves, _, err, errs := inst.MultiMatchSlaves(instanceKey, destinationKey, pattern)
if err != nil {
log.Fatale(err)
} else {
for _, e := range errs {
log.Errore(e)
}
for _, slave := range matchedSlaves {
fmt.Println(slave.Key.DisplayString())
}
}
}
case registerCliCommand("match-up-slaves", "Pseudo-GTID relocation", `Matches slaves of the given instance one level up the topology, making them siblings of given instance, using Pseudo-GTID`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
matchedSlaves, _, err, errs := inst.MatchUpSlaves(instanceKey, pattern)
if err != nil {
log.Fatale(err)
} else {
for _, e := range errs {
log.Errore(e)
}
for _, slave := range matchedSlaves {
fmt.Println(slave.Key.DisplayString())
}
}
}
case registerCliCommand("regroup-slaves-pgtid", "Pseudo-GTID relocation", `Given an instance, pick one of its slave and make it local master of its siblings, using Pseudo-GTID.`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
lostSlaves, equalSlaves, aheadSlaves, promotedSlave, err := inst.RegroupSlavesPseudoGTID(instanceKey, false, func(candidateSlave *inst.Instance) { fmt.Println(candidateSlave.Key.DisplayString()) }, postponedFunctionsContainer)
postponedFunctionsContainer.InvokePostponed()
if promotedSlave == nil {
log.Fatalf("Could not regroup slaves of %+v; error: %+v", *instanceKey, err)
}
fmt.Println(fmt.Sprintf("%s lost: %d, trivial: %d, pseudo-gtid: %d",
promotedSlave.Key.DisplayString(), len(lostSlaves), len(equalSlaves), len(aheadSlaves)))
if err != nil {
log.Fatale(err)
}
}
// General replication commands
case registerCliCommand("enable-gtid", "Replication, general", `If possible, turn on GTID replication`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.EnableGTID(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("disable-gtid", "Replication, general", `Turn off GTID replication, back to file:pos replication`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.DisableGTID(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("reset-master-gtid-remove-own-uuid", "Replication, general", `Reset master on instance, remove GTID entries generated by instance`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.ResetMasterGTIDOperation(instanceKey, true, "")
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("skip-query", "Replication, general", `Skip a single statement on a slave; either when running with GTID or without`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.SkipQuery(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("stop-slave", "Replication, general", `Issue a STOP SLAVE on an instance`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.StopSlave(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("start-slave", "Replication, general", `Issue a START SLAVE on an instance`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.StartSlave(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("restart-slave", "Replication, general", `STOP and START SLAVE on an instance`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.RestartSlave(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("reset-slave", "Replication, general", `Issues a RESET SLAVE command; use with care`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.ResetSlaveOperation(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("detach-slave", "Replication, general", `Stops replication and modifies binlog position into an impossible, yet reversible, value.`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.DetachSlaveOperation(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("reattach-slave", "Replication, general", `Undo a detach-slave operation`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.ReattachSlaveOperation(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("detach-slave-master-host", "Replication, general", `Stops replication and modifies Master_Host into an impossible, yet reversible, value.`):
{
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
_, err := inst.DetachSlaveMasterHost(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("reattach-slave-master-host", "Replication, general", `Undo a detach-slave-master-host operation`):
{
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
_, err := inst.ReattachSlaveMasterHost(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("master-pos-wait", "Replication, general", `Wait until slave reaches given replication coordinates (--binlog=file:pos)`):
{
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatalf("Unresolved instance")
}
instance, err := inst.ReadTopologyInstance(instanceKey)
if err != nil {
log.Fatale(err)
}
if instance == nil {
log.Fatalf("Instance not found: %+v", *instanceKey)
}
var binlogCoordinates *inst.BinlogCoordinates
if binlogCoordinates, err = inst.ParseBinlogCoordinates(*config.RuntimeCLIFlags.BinlogFile); err != nil {
log.Fatalf("Expecing --binlog argument as file:pos")
}
_, err = inst.MasterPosWait(instanceKey, binlogCoordinates)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
// Pool
case registerCliCommand("set-read-only", "Instance", `Turn an instance read-only, via SET GLOBAL read_only := 1`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.SetReadOnly(instanceKey, true)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("set-writeable", "Instance", `Turn an instance writeable, via SET GLOBAL read_only := 0`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
_, err := inst.SetReadOnly(instanceKey, false)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
// Binary log operations
case registerCliCommand("flush-binary-logs", "Binary logs", `Flush binary logs on an instance`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
var err error
if *config.RuntimeCLIFlags.BinlogFile == "" {
_, err = inst.FlushBinaryLogs(instanceKey, 1)
} else {
_, err = inst.FlushBinaryLogsTo(instanceKey, *config.RuntimeCLIFlags.BinlogFile)
}
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("purge-binary-logs", "Binary logs", `Purge binary logs of an instance`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
var err error
if *config.RuntimeCLIFlags.BinlogFile == "" {
log.Fatal("expecting --binlog value")
}
_, err = inst.PurgeBinaryLogsTo(instanceKey, *config.RuntimeCLIFlags.BinlogFile)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("last-pseudo-gtid", "Binary logs", `Find latest Pseudo-GTID entry in instance's binary logs`):
{
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatalf("Unresolved instance")
}
instance, err := inst.ReadTopologyInstance(instanceKey)
if err != nil {
log.Fatale(err)
}
if instance == nil {
log.Fatalf("Instance not found: %+v", *instanceKey)
}
coordinates, text, err := inst.FindLastPseudoGTIDEntry(instance, instance.RelaylogCoordinates, nil, strict, nil)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%+v:%s", *coordinates, text))
}
case registerCliCommand("find-binlog-entry", "Binary logs", `Get binlog file:pos of entry given by --pattern (exact full match, not a regular expression) in a given instance`):
{
if pattern == "" {
log.Fatal("No pattern given")
}
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatalf("Unresolved instance")
}
instance, err := inst.ReadTopologyInstance(instanceKey)
if err != nil {
log.Fatale(err)
}
if instance == nil {
log.Fatalf("Instance not found: %+v", *instanceKey)
}
coordinates, err := inst.SearchEntryInInstanceBinlogs(instance, pattern, false)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%+v", *coordinates))
}
case registerCliCommand("correlate-binlog-pos", "Binary logs", `Given an instance (-i) and binlog coordinates (--binlog=file:pos), find the correlated coordinates in another instance (-d)`):
{
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatalf("Unresolved instance")
}
instance, err := inst.ReadTopologyInstance(instanceKey)
if err != nil {
log.Fatale(err)
}
if instance == nil {
log.Fatalf("Instance not found: %+v", *instanceKey)
}
if !instance.LogBinEnabled {
log.Fatalf("Instance does not have binary logs: %+v", *instanceKey)
}
if destinationKey == nil {
log.Fatal("Cannot deduce target instance:", destination)
}
otherInstance, err := inst.ReadTopologyInstance(destinationKey)
if err != nil {
log.Fatale(err)
}
if otherInstance == nil {
log.Fatalf("Instance not found: %+v", *destinationKey)
}
var binlogCoordinates *inst.BinlogCoordinates
if *config.RuntimeCLIFlags.BinlogFile == "" {
binlogCoordinates = &instance.SelfBinlogCoordinates
} else {
if binlogCoordinates, err = inst.ParseBinlogCoordinates(*config.RuntimeCLIFlags.BinlogFile); err != nil {
log.Fatalf("Expecing --binlog argument as file:pos")
}
}
coordinates, _, err := inst.CorrelateBinlogCoordinates(instance, binlogCoordinates, otherInstance)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%+v", *coordinates))
}
// Pool
case registerCliCommand("submit-pool-instances", "Pools", `Submit a pool name with a list of instances in that pool`):
{
if pool == "" {
log.Fatal("Please submit --pool")
}
err := inst.ApplyPoolInstances(pool, instance)
if err != nil {
log.Fatale(err)
}
}
case registerCliCommand("cluster-pool-instances", "Pools", `List all pools and their associated instances`):
{
clusterPoolInstances, err := inst.ReadAllClusterPoolInstances()
if err != nil {
log.Fatale(err)
}
for _, clusterPoolInstance := range clusterPoolInstances {
fmt.Println(fmt.Sprintf("%s\t%s\t%s\t%s:%d", clusterPoolInstance.ClusterName, clusterPoolInstance.ClusterAlias, clusterPoolInstance.Pool, clusterPoolInstance.Hostname, clusterPoolInstance.Port))
}
}
// Information
case registerCliCommand("find", "Information", `Find instances whose hostname matches given regex pattern`):
{
if pattern == "" {
log.Fatal("No pattern given")
}
instances, err := inst.FindInstances(pattern)
if err != nil {
log.Fatale(err)
} else {
for _, instance := range instances {
fmt.Println(instance.Key.DisplayString())
}
}
}
case registerCliCommand("clusters", "Information", `List all clusters known to orchestrator`):
{
clusters, err := inst.ReadClusters()
if err != nil {
log.Fatale(err)
} else {
fmt.Println(strings.Join(clusters, "\n"))
}
}
case registerCliCommand("topology", "Information", `Show an ascii-graph of a replication topology, given a member of that topology`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
output, err := inst.ASCIITopology(instanceKey, pattern)
if err != nil {
log.Fatale(err)
}
fmt.Println(output)
}
case registerCliCommand("which-instance", "Information", `Output the fully-qualified hostname:port representation of the given instance, or error if unknown`):
{
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatalf("Unable to get master: unresolved instance")
}
instance, _, err := inst.ReadInstance(instanceKey)
if err != nil {
log.Fatale(err)
}
if instance == nil {
log.Fatalf("Instance not found: %+v", *instanceKey)
}
fmt.Println(instance.Key.DisplayString())
}
case registerCliCommand("which-cluster", "Information", `Output the name of the cluster an instance belongs to, or error if unknown to orchestrator`):
{
clusterName := getClusterName(clusterAlias, instanceKey)
fmt.Println(clusterName)
}
case registerCliCommand("which-cluster-instances", "Information", `Output the list of instances participating in same cluster as given instance`):
{
clusterName := getClusterName(clusterAlias, instanceKey)
instances, err := inst.ReadClusterInstances(clusterName)
if err != nil {
log.Fatale(err)
}
for _, clusterInstance := range instances {
fmt.Println(clusterInstance.Key.DisplayString())
}
}
case registerCliCommand("which-cluster-osc-slaves", "Information", `Output a list of slaves in same cluster as given instance, that could serve as a pt-online-schema-change operation control slaves`):
{
clusterName := getClusterName(clusterAlias, instanceKey)
instances, err := inst.GetClusterOSCSlaves(clusterName)
if err != nil {
log.Fatale(err)
}
for _, clusterInstance := range instances {
fmt.Println(clusterInstance.Key.DisplayString())
}
}
case registerCliCommand("which-master", "Information", `Output the fully-qualified hostname:port representation of a given instance's master`):
{
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatalf("Unable to get master: unresolved instance")
}
instance, _, err := inst.ReadInstance(instanceKey)
if err != nil {
log.Fatale(err)
}
if instance == nil {
log.Fatalf("Instance not found: %+v", *instanceKey)
}
fmt.Println(instance.MasterKey.DisplayString())
}
case registerCliCommand("which-slaves", "Information", `Output the fully-qualified hostname:port list of slaves of a given instance`):
{
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatalf("Unable to get slaves: unresolved instance")
}
slaves, err := inst.ReadSlaveInstances(instanceKey)
if err != nil {
log.Fatale(err)
}
for _, slave := range slaves {
fmt.Println(slave.Key.DisplayString())
}
}
case registerCliCommand("instance-status", "Information", `Output short status on a given instance`):
{
if instanceKey == nil {
instanceKey = assignThisInstanceKey()
}
if instanceKey == nil {
log.Fatalf("Unable to get status: unresolved instance")
}
instance, _, err := inst.ReadInstance(instanceKey)
if err != nil {
log.Fatale(err)
}
if instance == nil {
log.Fatalf("Instance not found: %+v", *instanceKey)
}
fmt.Println(instance.HumanReadableDescription())
}
case registerCliCommand("get-cluster-heuristic-lag", "Information", `For a given cluster (indicated by an instance or alias), output a heuristic "representative" lag of that cluster`):
{
clusterName := getClusterName(clusterAlias, instanceKey)
lag, err := inst.GetClusterHeuristicLag(clusterName)
if err != nil {
log.Fatale(err)
}
fmt.Println(lag)
}
// Instance management
case registerCliCommand("discover", "Instance management", `Lookup an instance, investigate it`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
instance, err := inst.ReadTopologyInstance(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instance.Key.DisplayString())
}
case registerCliCommand("forget", "Instance management", `Forget about an instance's existence`):
{
if rawInstanceKey == nil {
rawInstanceKey = assignThisInstanceKey()
}
if rawInstanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
err := inst.ForgetInstance(rawInstanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(rawInstanceKey.DisplayString())
}
case registerCliCommand("begin-maintenance", "Instance management", `Request a maintenance lock on an instance`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
if reason == "" {
log.Fatal("--reason option required")
}
var durationSeconds int = 0
if duration != "" {
durationSeconds, err = util.SimpleTimeToSeconds(duration)
if err != nil {
log.Fatale(err)
}
if durationSeconds < 0 {
log.Fatalf("Duration value must be non-negative. Given value: %d", durationSeconds)
}
}
maintenanceKey, err := inst.BeginBoundedMaintenance(instanceKey, inst.GetMaintenanceOwner(), reason, uint(durationSeconds))
if err == nil {
log.Infof("Maintenance key: %+v", maintenanceKey)
log.Infof("Maintenance duration: %d seconds", durationSeconds)
}
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("end-maintenance", "Instance management", `Remove maintenance lock from an instance`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
err := inst.EndMaintenanceByInstanceKey(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("begin-downtime", "Instance management", `Mark an instance as downtimed`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
if reason == "" {
log.Fatal("--reason option required")
}
var durationSeconds int = 0
if duration != "" {
durationSeconds, err = util.SimpleTimeToSeconds(duration)
if err != nil {
log.Fatale(err)
}
if durationSeconds < 0 {
log.Fatalf("Duration value must be non-negative. Given value: %d", durationSeconds)
}
}
err := inst.BeginDowntime(instanceKey, inst.GetMaintenanceOwner(), reason, uint(durationSeconds))
if err == nil {
log.Infof("Downtime duration: %d seconds", durationSeconds)
} else {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("end-downtime", "Instance management", `Indicate an instance is no longer downtimed`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
err := inst.EndDowntime(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
// Recovery & analysis
case registerCliCommand("recover", "Recovery", `Do auto-recovery given a dead instance`), registerCliCommand("recover-lite", "Recovery", `Do auto-recovery given a dead instance. Orchestrator chooses the best course of actionwithout executing external processes`):
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
recoveryAttempted, promotedInstanceKey, err := logic.CheckAndRecover(instanceKey, destinationKey, (command == "recover-lite"))
if err != nil {
log.Fatale(err)
}
if recoveryAttempted {
if promotedInstanceKey == nil {
log.Fatalf("Recovery attempted yet no slave promoted")
}
fmt.Println(promotedInstanceKey.DisplayString())
}
}
case registerCliCommand("replication-analysis", "Recovery", `Request an analysis of potential crash incidents in all known topologies`):
{
analysis, err := inst.GetReplicationAnalysis("", false, false)
if err != nil {
log.Fatale(err)
}
for _, entry := range analysis {
fmt.Println(fmt.Sprintf("%s (cluster %s): %s", entry.AnalyzedInstanceKey.DisplayString(), entry.ClusterDetails.ClusterName, entry.Analysis))
}
}
case registerCliCommand("ack-cluster-recoveries", "Recovery", `Acknowledge recoveries for a given cluster; this unblocks pending future recoveries`):
{
if reason == "" {
log.Fatal("--reason option required (comment your ack)")
}
clusterName := getClusterName(clusterAlias, instanceKey)
countRecoveries, err := logic.AcknowledgeClusterRecoveries(clusterName, inst.GetMaintenanceOwner(), reason)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%d recoveries acknowldged", countRecoveries))
}
case registerCliCommand("ack-instance-recoveries", "Recovery", `Acknowledge recoveries for a given instance; this unblocks pending future recoveries`):
{
if reason == "" {
log.Fatal("--reason option required (comment your ack)")
}
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
countRecoveries, err := logic.AcknowledgeInstanceRecoveries(instanceKey, inst.GetMaintenanceOwner(), reason)
if err != nil {
log.Fatale(err)
}
fmt.Println(fmt.Sprintf("%d recoveries acknowldged", countRecoveries))
}
// Instance meta
case registerCliCommand("register-candidate", "Instance, meta", `Indicate that a specific instance is a preferred candidate for master promotion`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
err := inst.RegisterCandidateInstance(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("register-hostname-unresolve", "Instance, meta", `Assigns the given instance a virtual (aka "unresolved") name`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
err := inst.RegisterHostnameUnresolve(instanceKey, hostnameFlag)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
case registerCliCommand("deregister-hostname-unresolve", "Instance, meta", `Explicitly deregister/dosassociate a hostname with an "unresolved" name`):
{
instanceKey = deduceInstanceKeyIfNeeded(instance, instanceKey)
err := inst.DeregisterHostnameUnresolve(instanceKey)
if err != nil {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
// meta
case registerCliCommand("snapshot-topologies", "Meta", `Take a snapshot of existing topologies.`):
{
err := inst.SnapshotTopologies()
if err != nil {
log.Fatale(err)
}
}
case registerCliCommand("continuous", "Meta", `Enter continuous mode, and actively poll for instances, diagnose problems, do maintenance`):
{
logic.ContinuousDiscovery()
}
case registerCliCommand("resolve", "Meta", `Resolve given hostname`):
{
if rawInstanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
if conn, err := net.Dial("tcp", rawInstanceKey.DisplayString()); err == nil {
log.Debugf("tcp test is good; got connection %+v", conn)
conn.Close()
} else {
log.Fatale(err)
}
if cname, err := inst.GetCNAME(rawInstanceKey.Hostname); err == nil {
log.Debugf("GetCNAME() %+v, %+v", cname, err)
rawInstanceKey.Hostname = cname
fmt.Println(rawInstanceKey.DisplayString())
} else {
log.Fatale(err)
}
}
case registerCliCommand("reset-hostname-resolve-cache", "Meta", `Clear the hostname resolve cache`):
{
err := inst.ResetHostnameResolveCache()
if err != nil {
log.Fatale(err)
}
fmt.Println("hostname resolve cache cleared")
}
case registerCliCommand("reset-internal-db-deployment", "Meta, internal", `Clear internal db deployment history, use if somehow corrupted internal deployment history`):
{
config.Config.SkipOrchestratorDatabaseUpdate = true
db.ResetInternalDeployment()
fmt.Println("Internal db deployment history reset. Next orchestrator execution will rebuild internal db structure (no data will be lost)")
}
// Help
case "help":
{
fmt.Fprintf(os.Stderr, availableCommandsUsage())
}
default:
log.Fatalf("Unknown command: \"%s\". %s", command, availableCommandsUsage())
}
}
