// MakeCoMaster will attempt to make an instance co-master with its master, by making its master a slave of its own.
// This only works out if the master is not replicating; the master does not have a known master (it may have an unknown master).
func MakeCoMaster(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, err
}
master, err := GetInstanceMaster(instance)
if err != nil {
return instance, err
}
rinstance, _, _ := ReadInstance(&master.Key)
if canMove, merr := rinstance.CanMoveAsCoMaster(); !canMove {
return instance, merr
}
rinstance, _, _ = ReadInstance(instanceKey)
if canMove, merr := rinstance.CanMove(); !canMove {
return instance, merr
}
if instanceKey.Equals(&master.MasterKey) {
return instance, errors.New(fmt.Sprintf("instance %+v is already co master of %+v", instanceKey, master.Key))
}
if _, found, _ := ReadInstance(&master.MasterKey); found {
return instance, errors.New(fmt.Sprintf("master %+v already has known master: %+v", master.Key, master.MasterKey))
}
if canReplicate, err := master.CanReplicateFrom(instance); !canReplicate {
return instance, err
}
log.Infof("Will make %+v co-master of %+v", instanceKey, master.Key)
if maintenanceToken, merr := BeginMaintenance(instanceKey, "orchestrator", fmt.Sprintf("make co-master of %+v", master.Key)); merr != nil {
err = errors.New(fmt.Sprintf("Cannot begin maintenance on %+v", *instanceKey))
goto Cleanup
} else {
defer EndMaintenance(maintenanceToken)
}
if maintenanceToken, merr := BeginMaintenance(&master.Key, "orchestrator", fmt.Sprintf("%+v turns into co-master of this", *instanceKey)); merr != nil {
err = errors.New(fmt.Sprintf("Cannot begin maintenance on %+v", master.Key))
goto Cleanup
} else {
defer EndMaintenance(maintenanceToken)
}
// the coMaster used to be merely a slave. Just point master into *some* position
// within coMaster...
master, err = ChangeMasterTo(&master.Key, instanceKey, &instance.SelfBinlogCoordinates)
if err != nil {
goto Cleanup
}
Cleanup:
master, _ = StartSlave(&master.Key)
if err != nil {
return instance, log.Errore(err)
}
// and we're done (pending deferred functions)
AuditOperation("make-co-master", instanceKey, fmt.Sprintf("%+v made co-master of %+v", *instanceKey, master.Key))
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)
}
}
return Config, err
}
// 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, fmt.Sprintf("select master_pos_wait('%s', %d)",
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
}
// 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, errors.New(fmt.Sprintf("instance is not a slave: %+v", instanceKey))
}
_, err = ExecInstance(instanceKey, `stop slave`)
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
}
// 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, errors.New(fmt.Sprintf("instance is not a slave: %+v", instanceKey))
}
if instance.SlaveRunning() {
return instance, errors.New(fmt.Sprintf("slave already running: %+v", instanceKey))
}
log.Infof("Will start slave on %+v until coordinates: %+v", instanceKey, masterCoordinates)
_, err = ExecInstance(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 up_to_date := false; !up_to_date; {
instance, err = ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
switch {
case instance.ExecBinlogCoordinates.SmallerThan(masterCoordinates):
time.Sleep(200 * time.Millisecond)
case instance.ExecBinlogCoordinates.Equals(masterCoordinates):
up_to_date = true
case masterCoordinates.SmallerThan(&instance.ExecBinlogCoordinates):
return instance, errors.New(fmt.Sprintf("Start SLAVE UNTIL is past coordinates: %+v", instanceKey))
}
}
instance, err = StopSlave(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
return instance, err
}
// DetachSlave detaches a slave from its master. Instead of performing destructive RESET SLAVE,
// this function merely resets the MASTER_PORT, which effectively disconnects from master and changes its key altogether.
func DetachSlave(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if instance.SlaveRunning() {
return instance, errors.New(fmt.Sprintf("Cannot detach slave on: %+v because slave is running", instanceKey))
}
_, err = ExecInstance(instanceKey, fmt.Sprintf("change master to master_port=%d", InvalidPort))
if err != nil {
return instance, log.Errore(err)
}
log.Infof("Detached slave %+v", instanceKey)
instance, err = ReadTopologyInstance(instanceKey)
return instance, err
}
// ChangeMasterTo changes the given instance's master according to given input.
func ChangeMasterTo(instanceKey *InstanceKey, masterKey *InstanceKey, masterBinlogCoordinates *BinlogCoordinates) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if instance.SlaveRunning() {
return instance, errors.New(fmt.Sprintf("Cannot change master on: %+v because slave is running", instanceKey))
}
_, err = ExecInstance(instanceKey, fmt.Sprintf("change master to master_host='%s', master_port=%d, master_log_file='%s', master_log_pos=%d",
masterKey.Hostname, masterKey.Port, masterBinlogCoordinates.LogFile, masterBinlogCoordinates.LogPos))
if err != nil {
return instance, log.Errore(err)
}
log.Infof("Changed master on %+v to: %+v, %+v", instanceKey, masterKey, masterBinlogCoordinates)
instance, err = ReadTopologyInstance(instanceKey)
return instance, err
}
// DetachSlaveFromMaster will detach an instance from being a slave, and break its replication.
// This only works if the instance is indeed a slave of a known instance.
func DetachSlaveFromMaster(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, err
}
master, err := GetInstanceMaster(instance)
if err != nil {
return instance, err
}
log.Infof("Will detach %+v from its master %+v", instanceKey, master.Key)
if maintenanceToken, merr := BeginMaintenance(instanceKey, "orchestrator", fmt.Sprintf("detach from master %+v", master.Key)); merr != nil {
err = errors.New(fmt.Sprintf("Cannot begin maintenance on %+v", *instanceKey))
goto Cleanup
} else {
defer EndMaintenance(maintenanceToken)
}
instance, err = StopSlave(instanceKey)
if err != nil {
goto Cleanup
}
instance, err = DetachSlave(instanceKey)
if err != nil {
goto Cleanup
}
Cleanup:
instance, _ = StartSlave(instanceKey)
_, _ = RefreshInstanceSlaveHosts(&master.Key)
master, _ = ReadTopologyInstance(&master.Key)
if err != nil {
return instance, log.Errore(err)
}
// and we're done (pending deferred functions)
AuditOperation("detach slave", instanceKey, fmt.Sprintf("%+v detached from master %+v", *instanceKey, master.Key))
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, errors.New(fmt.Sprintf("instance is not a slave: %+v", instanceKey))
}
_, err = ExecInstance(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
}
// MoveUp will attempt moving instance indicated by instanceKey up the topology hierarchy.
// It will perform all safety and sanity checks and will tamper with this instance's replication
// as well as its master.
func MoveUp(instanceKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, err
}
if !instance.IsSlave() {
return instance, errors.New(fmt.Sprintf("instance is not a slave: %+v", instanceKey))
}
rinstance, _, _ := ReadInstance(&instance.Key)
if canMove, merr := rinstance.CanMove(); !canMove {
return instance, merr
}
master, err := GetInstanceMaster(instance)
if err != nil {
return instance, log.Errorf("Cannot GetInstanceMaster() for %+v. error=%+v", instance, err)
}
if !master.IsSlave() {
return instance, errors.New(fmt.Sprintf("master is not a slave itself: %+v", master.Key))
}
if canReplicate, err := instance.CanReplicateFrom(master); canReplicate == false {
return instance, err
}
log.Infof("Will move %+v up the topology", *instanceKey)
if maintenanceToken, merr := BeginMaintenance(instanceKey, "orchestrator", "move up"); merr != nil {
err = errors.New(fmt.Sprintf("Cannot begin maintenance on %+v", *instanceKey))
goto Cleanup
} else {
defer EndMaintenance(maintenanceToken)
}
if maintenanceToken, merr := BeginMaintenance(&master.Key, "orchestrator", fmt.Sprintf("child %+v moves up", *instanceKey)); merr != nil {
err = errors.New(fmt.Sprintf("Cannot begin maintenance on %+v", master.Key))
goto Cleanup
} else {
defer EndMaintenance(maintenanceToken)
}
master, err = StopSlave(&master.Key)
if err != nil {
goto Cleanup
}
instance, err = StopSlave(instanceKey)
if err != nil {
goto Cleanup
}
instance, err = StartSlaveUntilMasterCoordinates(instanceKey, &master.SelfBinlogCoordinates)
if err != nil {
goto Cleanup
}
instance, err = ChangeMasterTo(instanceKey, &master.MasterKey, &master.ExecBinlogCoordinates)
if err != nil {
goto Cleanup
}
Cleanup:
instance, _ = StartSlave(instanceKey)
master, _ = StartSlave(&master.Key)
if err != nil {
return instance, log.Errore(err)
}
// and we're done (pending deferred functions)
AuditOperation("move-up", instanceKey, fmt.Sprintf("moved up %+v. Previous master: %+v", *instanceKey, master.Key))
return instance, err
}
// MoveBelow will attempt moving instance indicated by instanceKey below its supposed sibling indicated by sinblingKey.
// It will perform all safety and sanity checks and will tamper with this instance's replication
// as well as its sibling.
func MoveBelow(instanceKey, siblingKey *InstanceKey) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, err
}
sibling, err := ReadTopologyInstance(siblingKey)
if err != nil {
return instance, err
}
rinstance, _, _ := ReadInstance(&instance.Key)
if canMove, merr := rinstance.CanMove(); !canMove {
return instance, merr
}
rinstance, _, _ = ReadInstance(&sibling.Key)
if canMove, merr := rinstance.CanMove(); !canMove {
return instance, merr
}
if !InstancesAreSiblings(instance, sibling) {
return instance, errors.New(fmt.Sprintf("instances are not siblings: %+v, %+v", *instanceKey, *siblingKey))
}
if canReplicate, err := instance.CanReplicateFrom(sibling); !canReplicate {
return instance, err
}
log.Infof("Will move %+v below its sibling %+v", instanceKey, siblingKey)
if maintenanceToken, merr := BeginMaintenance(instanceKey, "orchestrator", fmt.Sprintf("move below %+v", *siblingKey)); merr != nil {
err = errors.New(fmt.Sprintf("Cannot begin maintenance on %+v", *instanceKey))
goto Cleanup
} else {
defer EndMaintenance(maintenanceToken)
}
if maintenanceToken, merr := BeginMaintenance(siblingKey, "orchestrator", fmt.Sprintf("%+v moves below this", *instanceKey)); merr != nil {
err = errors.New(fmt.Sprintf("Cannot begin maintenance on %+v", *siblingKey))
goto Cleanup
} else {
defer EndMaintenance(maintenanceToken)
}
instance, err = StopSlave(instanceKey)
if err != nil {
goto Cleanup
}
sibling, err = StopSlave(siblingKey)
if err != nil {
goto Cleanup
}
if instance.ExecBinlogCoordinates.SmallerThan(&sibling.ExecBinlogCoordinates) {
instance, err = StartSlaveUntilMasterCoordinates(instanceKey, &sibling.ExecBinlogCoordinates)
if err != nil {
goto Cleanup
}
} else if sibling.ExecBinlogCoordinates.SmallerThan(&instance.ExecBinlogCoordinates) {
sibling, err = StartSlaveUntilMasterCoordinates(siblingKey, &instance.ExecBinlogCoordinates)
if err != nil {
goto Cleanup
}
}
// At this point both siblings have executed exact same statements and are identical
instance, err = ChangeMasterTo(instanceKey, &sibling.Key, &sibling.SelfBinlogCoordinates)
if err != nil {
goto Cleanup
}
Cleanup:
instance, _ = StartSlave(instanceKey)
sibling, _ = StartSlave(siblingKey)
if err != nil {
return instance, log.Errore(err)
}
// and we're done (pending deferred functions)
AuditOperation("move-below", instanceKey, fmt.Sprintf("moved %+v below %+v", *instanceKey, *siblingKey))
return instance, err
}
// Cli initiates a command line interface, executing requested command.
func Cli(command string, instance string, sibling string, owner string, reason string) {
instanceKey, err := inst.ParseInstanceKey(instance)
if err != nil {
instanceKey = nil
}
siblingKey, err := inst.ParseInstanceKey(sibling)
if err != nil {
siblingKey = nil
}
if len(owner) == 0 {
// get os username as owner
usr, err := user.Current()
if err != nil {
log.Fatale(err)
}
owner = usr.Username
}
if len(command) == 0 {
log.Fatal("expected command (-c) (discover|forget|continuous|move-up|move-below|begin-maintenance|end-maintenance|clusters|topology|resolve)")
}
switch command {
case "move-up":
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
_, err := inst.MoveUp(instanceKey)
if err != nil {
log.Errore(err)
}
}
case "move-below":
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
if siblingKey == nil {
log.Fatal("Cannot deduce sibling:", sibling)
}
_, err := inst.MoveBelow(instanceKey, siblingKey)
if err != nil {
log.Errore(err)
}
}
case "discover":
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
orchestrator.StartDiscovery(*instanceKey)
}
case "forget":
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
inst.ForgetInstance(instanceKey)
}
case "begin-maintenance":
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
if owner == "" {
log.Fatal("--owner option required")
}
if reason == "" {
log.Fatal("--reason option required")
}
maintenanceKey, err := inst.BeginMaintenance(instanceKey, owner, reason)
if err == nil {
log.Infof("Maintenance key: %+v", maintenanceKey)
}
if err != nil {
log.Errore(err)
}
}
case "end-maintenance":
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
err := inst.EndMaintenanceByInstanceKey(instanceKey)
if err != nil {
log.Errore(err)
}
}
case "clusters":
{
clusters, err := inst.ReadClusters()
if err != nil {
log.Errore(err)
} else {
fmt.Println(strings.Join(clusters, "\n"))
}
}
case "topology":
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
output, err := inst.AsciiTopology(instance)
if err != nil {
log.Errore(err)
} else {
fmt.Println(output)
}
}
case "continuous":
{
orchestrator.ContinuousDiscovery()
}
case "resolve":
{
if instanceKey == nil {
log.Fatal("Cannot deduce instance:", instance)
}
if conn, err := net.Dial("tcp", instanceKey.DisplayString()); err == nil {
conn.Close()
} else {
log.Fatale(err)
}
fmt.Println(instanceKey.DisplayString())
}
default:
log.Fatal("Unknown command:", command)
}
}
// DiscoverInstance will attempt discovering an instance (unless it is already up to date) and will
// list down its master and slaves (if any) for further discovery.
func DiscoverInstance(instanceKey inst.InstanceKey) {
instanceKey.Formalize()
if !instanceKey.IsValid() {
return
}
instance, found, err := inst.ReadInstance(&instanceKey)
if found && instance.IsUpToDate && instance.IsLastCheckValid {
// we've already discovered this one. Skip!
goto Cleanup
}
// First we've ever heard of this instance. Continue investigation:
instance, err = inst.ReadTopologyInstance(&instanceKey)
// panic can occur (IO stuff). Therefore it may happen
// that instance is nil. Check it.
if err != nil || instance == nil {
goto Cleanup
}
fmt.Printf("host: %+v, master: %+v\n", instance.Key, instance.MasterKey)
// Investigate slaves:
for _, slaveKey := range instance.SlaveHosts.GetInstanceKeys() {
discoveryInstanceKeys <- slaveKey
}
// Investigate master:
discoveryInstanceKeys <- instance.MasterKey
Cleanup:
}
// Start discovery begins a one time asynchronuous discovery process for the given
// instance and all of its topology connected instances.
// That is, the instance will be investigated for master and slaves, and the routines will follow on
// each and every such found master/slave.
// In essense, assuming all slaves in a replication topology are running, and given a single instance
// in such topology, this function will detect the entire topology.
func StartDiscovery(instanceKey inst.InstanceKey) {
log.Infof("Starting discovery at %+v", instanceKey)
pendingTokens := make(chan bool, maxConcurrency)
completedTokens := make(chan bool, maxConcurrency)
AccountedDiscoverInstance(instanceKey, pendingTokens, completedTokens)
go handleDiscoveryRequests(pendingTokens, completedTokens)
// Block until all are complete
for {
select {
case <-pendingTokens:
<-completedTokens
default:
inst.AuditOperation("start-discovery", &instanceKey, "")
return
}
}
}
// 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() {
log.Infof("Starting continuous discovery")
go handleDiscoveryRequests(nil, nil)
tick := time.Tick(time.Duration(config.Config.DiscoveryPollSeconds) * time.Second)
forgetUnseenTick := time.Tick(time.Hour)
for _ = range tick {
instanceKeys, _ := inst.ReadOutdatedInstanceKeys()
log.Debugf("outdated keys: %+v", instanceKeys)
for _, instanceKey := range instanceKeys {
discoveryInstanceKeys <- instanceKey
}
// See if we should also forget instances (lower frequency)
select {
case <-forgetUnseenTick:
inst.ForgetLongUnseenInstances()
default:
}
}
}
// ReadInstance reads an instance from the orchestrator backend database
func ReadInstance(instanceKey *InstanceKey) (*Instance, bool, error) {
db, err := db.OpenOrchestrator()
if err != nil {
return nil, false, log.Errore(err)
}
instance := NewInstance()
instance.Key = *instanceKey
var slaveHostsJson string
var secondsSinceLastChecked uint
err = db.QueryRow(`
select
server_id,
version,
binlog_format,
log_bin,
log_slave_updates,
binary_log_file,
binary_log_pos,
master_host,
master_port,
slave_sql_running,
slave_io_running,
master_log_file,
read_master_log_pos,
relay_master_log_file,
exec_master_log_pos,
seconds_behind_master,
slave_lag_seconds,
slave_hosts,
cluster_name,
timestampdiff(second, last_checked, now()) as seconds_since_last_checked,
(last_checked <= last_seen) is true as is_last_check_valid,
timestampdiff(second, last_seen, now()) as seconds_since_last_seen
from database_instance
where hostname=? and port=?`,
instanceKey.Hostname, instanceKey.Port).Scan(
&instance.ServerID,
&instance.Version,
&instance.Binlog_format,
&instance.LogBinEnabled,
&instance.LogSlaveUpdatesEnabled,
&instance.SelfBinlogCoordinates.LogFile,
&instance.SelfBinlogCoordinates.LogPos,
&instance.MasterKey.Hostname,
&instance.MasterKey.Port,
&instance.Slave_SQL_Running,
&instance.Slave_IO_Running,
&instance.ReadBinlogCoordinates.LogFile,
&instance.ReadBinlogCoordinates.LogPos,
&instance.ExecBinlogCoordinates.LogFile,
&instance.ExecBinlogCoordinates.LogPos,
&instance.SecondsBehindMaster,
&instance.SlaveLagSeconds,
&slaveHostsJson,
&instance.ClusterName,
&secondsSinceLastChecked,
&instance.IsLastCheckValid,
&instance.SecondsSinceLastSeen,
)
if err == sql.ErrNoRows {
log.Infof("No entry for %+v", instanceKey)
return instance, false, err
}
if err != nil {
log.Error("error on", instanceKey, err)
return instance, false, err
}
instance.IsUpToDate = (secondsSinceLastChecked <= config.Config.InstancePollSeconds)
instance.IsRecentlyChecked = (secondsSinceLastChecked <= config.Config.InstancePollSeconds*5)
instance.ReadSlaveHostsFromJson(slaveHostsJson)
return instance, true, err
}