// SwapTablesQuickAndBumpy issues a two-step swap table operation:
// - rename original table to _old
// - rename ghost table to original
// There is a point in time in between where the table does not exist.
func (this *Applier) SwapTablesQuickAndBumpy() error {
query := fmt.Sprintf(`alter /* gh-ost */ table %s.%s rename %s`,
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.OriginalTableName),
sql.EscapeName(this.migrationContext.GetOldTableName()),
)
log.Infof("Renaming original table")
this.migrationContext.RenameTablesStartTime = time.Now()
if _, err := sqlutils.ExecNoPrepare(this.singletonDB, query); err != nil {
return err
}
query = fmt.Sprintf(`alter /* gh-ost */ table %s.%s rename %s`,
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.GetGhostTableName()),
sql.EscapeName(this.migrationContext.OriginalTableName),
)
log.Infof("Renaming ghost table")
if _, err := sqlutils.ExecNoPrepare(this.db, query); err != nil {
return err
}
this.migrationContext.RenameTablesEndTime = time.Now()
log.Infof("Tables renamed")
return nil
}
// CreateChangelogTable creates the changelog table on the applier host
func (this *Applier) CreateChangelogTable() error {
if err := this.DropChangelogTable(); err != nil {
return err
}
query := fmt.Sprintf(`create /* gh-ost */ table %s.%s (
id bigint auto_increment,
last_update timestamp not null DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
hint varchar(64) charset ascii not null,
value varchar(255) charset ascii not null,
primary key(id),
unique key hint_uidx(hint)
) auto_increment=256
`,
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.GetChangelogTableName()),
)
log.Infof("Creating changelog table %s.%s",
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.GetChangelogTableName()),
)
if _, err := sqlutils.ExecNoPrepare(this.db, query); err != nil {
return err
}
log.Infof("Changelog table created")
return nil
}
// GetCandidateSiblingOfIntermediateMaster chooses the best sibling of a dead intermediate master
// to whom the IM's slaves can be moved.
func GetCandidateSiblingOfIntermediateMaster(intermediateMasterKey *inst.InstanceKey) (*inst.Instance, error) {
intermediateMasterInstance, _, err := inst.ReadInstance(intermediateMasterKey)
if err != nil {
return nil, err
}
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", *intermediateMasterKey)
}
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", *intermediateMasterKey)
}
// initiateInspector connects, validates and inspects the "inspector" server.
// The "inspector" server is typically a replica; it is where we issue some
// queries such as:
// - table row count
// - schema validation
// - heartbeat
// When `--allow-on-master` is supplied, the inspector is actually the master.
func (this *Migrator) initiateInspector() (err error) {
this.inspector = NewInspector()
if err := this.inspector.InitDBConnections(); err != nil {
return err
}
if err := this.inspector.ValidateOriginalTable(); err != nil {
return err
}
if err := this.inspector.InspectOriginalTable(); err != nil {
return err
}
// So far so good, table is accessible and valid.
// Let's get master connection config
if this.migrationContext.ApplierConnectionConfig, err = this.inspector.getMasterConnectionConfig(); err != nil {
return err
}
if this.migrationContext.TestOnReplica || this.migrationContext.MigrateOnReplica {
if this.migrationContext.InspectorIsAlsoApplier() {
return fmt.Errorf("Instructed to --test-on-replica or --migrate-on-replica, but the server we connect to doesn't seem to be a replica")
}
log.Infof("--test-on-replica or --migrate-on-replica given. Will not execute on master %+v but rather on replica %+v itself",
*this.migrationContext.ApplierConnectionConfig.ImpliedKey, *this.migrationContext.InspectorConnectionConfig.ImpliedKey,
)
this.migrationContext.ApplierConnectionConfig = this.migrationContext.InspectorConnectionConfig.Duplicate()
if this.migrationContext.GetThrottleControlReplicaKeys().Len() == 0 {
this.migrationContext.AddThrottleControlReplicaKey(this.migrationContext.InspectorConnectionConfig.Key)
}
} else if this.migrationContext.InspectorIsAlsoApplier() && !this.migrationContext.AllowedRunningOnMaster {
return fmt.Errorf("It seems like this migration attempt to run directly on master. Preferably it would be executed on a replica (and this reduces load from the master). To proceed please provide --allow-on-master")
}
log.Infof("Master found to be %+v", *this.migrationContext.ApplierConnectionConfig.ImpliedKey)
return nil
}
// finalCleanup takes actions at very end of migration, dropping tables etc.
func (this *Migrator) finalCleanup() error {
atomic.StoreInt64(&this.cleanupImminentFlag, 1)
if this.migrationContext.Noop {
if createTableStatement, err := this.inspector.showCreateTable(this.migrationContext.GetGhostTableName()); err == nil {
log.Infof("New table structure follows")
fmt.Println(createTableStatement)
} else {
log.Errore(err)
}
}
if err := this.retryOperation(this.applier.DropChangelogTable); err != nil {
return err
}
if this.migrationContext.OkToDropTable && !this.migrationContext.TestOnReplica {
if err := this.retryOperation(this.applier.DropOldTable); err != nil {
return err
}
} else {
if !this.migrationContext.Noop {
log.Infof("Am not dropping old table because I want this operation to be as live as possible. If you insist I should do it, please add `--ok-to-drop-table` next time. But I prefer you do not. To drop the old table, issue:")
log.Infof("-- drop table %s.%s", sql.EscapeName(this.migrationContext.DatabaseName), sql.EscapeName(this.migrationContext.GetOldTableName()))
}
}
if this.migrationContext.Noop {
if err := this.retryOperation(this.applier.DropGhostTable); err != nil {
return err
}
}
return nil
}
// DropAtomicCutOverSentryTableIfExists checks if the "old" table name
// happens to be a cut-over magic table; if so, it drops it.
func (this *Applier) CreateAtomicCutOverSentryTable() error {
if err := this.DropAtomicCutOverSentryTableIfExists(); err != nil {
return err
}
tableName := this.migrationContext.GetOldTableName()
query := fmt.Sprintf(`create /* gh-ost */ table %s.%s (
id int auto_increment primary key
) comment='%s'
`,
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(tableName),
atomicCutOverMagicHint,
)
log.Infof("Creating magic cut-over table %s.%s",
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(tableName),
)
if _, err := sqlutils.ExecNoPrepare(this.db, query); err != nil {
return err
}
log.Infof("Magic cut-over table created")
return nil
}
// InspectOriginalAndGhostTables compares original and ghost tables to see whether the migration
// makes sense and is valid. It extracts the list of shared columns and the chosen migration unique key
func (this *Inspector) InspectOriginalAndGhostTables() (err error) {
this.migrationContext.GhostTableColumns, this.migrationContext.GhostTableUniqueKeys, err = this.InspectTableColumnsAndUniqueKeys(this.migrationContext.GetGhostTableName())
if err != nil {
return err
}
sharedUniqueKeys, err := this.getSharedUniqueKeys(this.migrationContext.OriginalTableUniqueKeys, this.migrationContext.GhostTableUniqueKeys)
if err != nil {
return err
}
if len(sharedUniqueKeys) == 0 {
return fmt.Errorf("No shared unique key can be found after ALTER! Bailing out")
}
this.migrationContext.UniqueKey = sharedUniqueKeys[0]
log.Infof("Chosen shared unique key is %s", this.migrationContext.UniqueKey.Name)
if this.migrationContext.UniqueKey.HasNullable {
if this.migrationContext.NullableUniqueKeyAllowed {
log.Warningf("Chosen key (%s) has nullable columns. You have supplied with --allow-nullable-unique-key and so this migration proceeds. As long as there aren't NULL values in this key's column, migration should be fine. NULL values will corrupt migration's data", this.migrationContext.UniqueKey)
} else {
return fmt.Errorf("Chosen key (%s) has nullable columns. Bailing out. To force this operation to continue, supply --allow-nullable-unique-key flag. Only do so if you are certain there are no actual NULL values in this key. As long as there aren't, migration should be fine. NULL values in columns of this key will corrupt migration's data", this.migrationContext.UniqueKey)
}
}
if !this.migrationContext.UniqueKey.IsPrimary() {
if this.migrationContext.OriginalBinlogRowImage != "FULL" {
return fmt.Errorf("binlog_row_image is '%s' and chosen key is %s, which is not the primary key. This operation cannot proceed. You may `set global binlog_row_image='full'` and try again", this.migrationContext.OriginalBinlogRowImage, this.migrationContext.UniqueKey)
}
}
this.migrationContext.SharedColumns, this.migrationContext.MappedSharedColumns = this.getSharedColumns(this.migrationContext.OriginalTableColumns, this.migrationContext.GhostTableColumns, this.migrationContext.ColumnRenameMap)
log.Infof("Shared columns are %s", this.migrationContext.SharedColumns)
// By fact that a non-empty unique key exists we also know the shared columns are non-empty
return nil
}
// StopSlaveIOThread is applicable with --test-on-replica; it stops the IO thread, duh.
// We need to keep the SQL thread active so as to complete processing received events,
// and have them written to the binary log, so that we can then read them via streamer.
func (this *Applier) StopSlaveIOThread() error {
query := `stop /* gh-ost */ slave io_thread`
log.Infof("Stopping replication")
if _, err := sqlutils.ExecNoPrepare(this.db, query); err != nil {
return err
}
log.Infof("Replication stopped")
return nil
}
// StartSlaveSQLThread is applicable with --test-on-replica
func (this *Applier) StartSlaveSQLThread() error {
query := `start /* gh-ost */ slave sql_thread`
log.Infof("Verifying SQL thread is running")
if _, err := sqlutils.ExecNoPrepare(this.db, query); err != nil {
return err
}
log.Infof("SQL thread started")
return nil
}
// UnlockTables makes tea. No wait, it unlocks tables.
func (this *Applier) UnlockTables() error {
query := `unlock /* gh-ost */ tables`
log.Infof("Unlocking tables")
if _, err := sqlutils.ExecNoPrepare(this.singletonDB, query); err != nil {
return err
}
log.Infof("Tables unlocked")
return nil
}
// DropAtomicCutOverSentryTableIfExists checks if the "old" table name
// happens to be a cut-over magic table; if so, it drops it.
func (this *Applier) DropAtomicCutOverSentryTableIfExists() error {
log.Infof("Looking for magic cut-over table")
tableName := this.migrationContext.GetOldTableName()
rowMap := this.showTableStatus(tableName)
if rowMap == nil {
// Table does not exist
return nil
}
if rowMap["Comment"].String != atomicCutOverMagicHint {
return fmt.Errorf("Expected magic comment on %s, did not find it", tableName)
}
log.Infof("Dropping magic cut-over table")
return this.dropTable(tableName)
}
// dropTable drops a given table on the applied host
func (this *Applier) dropTable(tableName string) error {
query := fmt.Sprintf(`drop /* gh-ost */ table if exists %s.%s`,
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(tableName),
)
log.Infof("Droppping table %s.%s",
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(tableName),
)
if _, err := sqlutils.ExecNoPrepare(this.db, query); err != nil {
return err
}
log.Infof("Table dropped")
return nil
}
// Http starts serving HTTP (api/web) requests
func Http() {
martini.Env = martini.Prod
m := martini.Classic()
if config.Config.HTTPAuthUser != "" {
m.Use(auth.Basic(config.Config.HTTPAuthUser, config.Config.HTTPAuthPassword))
}
m.Use(gzip.All())
// Render html templates from templates directory
m.Use(render.Renderer(render.Options{
Directory: "resources",
Layout: "templates/layout",
HTMLContentType: "text/html",
}))
m.Use(martini.Static("resources/public"))
go agent.ContinuousOperation()
log.Infof("Starting HTTP on port %d", config.Config.HTTPPort)
http.API.RegisterRequests(m)
// Serve
if config.Config.UseSSL {
log.Info("Serving via SSL")
err := nethttp.ListenAndServeTLS(fmt.Sprintf(":%d", config.Config.HTTPPort), config.Config.SSLCertFile, config.Config.SSLPrivateKeyFile, m)
if err != nil {
log.Fatale(err)
}
} else {
nethttp.ListenAndServe(fmt.Sprintf(":%d", config.Config.HTTPPort), m)
}
}
// ConnectBinlogStreamer
func (this *GoMySQLReader) ConnectBinlogStreamer(coordinates mysql.BinlogCoordinates) (err error) {
if coordinates.IsEmpty() {
return log.Errorf("Emptry coordinates at ConnectBinlogStreamer()")
}
log.Infof("Registering replica at %+v:%+v", this.connectionConfig.Key.Hostname, uint16(this.connectionConfig.Key.Port))
if err := this.binlogSyncer.RegisterSlave(this.connectionConfig.Key.Hostname, uint16(this.connectionConfig.Key.Port), this.connectionConfig.User, this.connectionConfig.Password); err != nil {
return err
}
this.currentCoordinates = coordinates
log.Infof("Connecting binlog streamer at %+v", this.currentCoordinates)
// Start sync with sepcified binlog file and position
this.binlogStreamer, err = this.binlogSyncer.StartSync(gomysql.Position{this.currentCoordinates.LogFile, uint32(this.currentCoordinates.LogPos)})
return err
}
// CountTableRows counts exact number of rows on the original table
func (this *Inspector) CountTableRows() error {
atomic.StoreInt64(&this.migrationContext.CountingRowsFlag, 1)
defer atomic.StoreInt64(&this.migrationContext.CountingRowsFlag, 0)
log.Infof("As instructed, I'm issuing a SELECT COUNT(*) on the table. This may take a while")
query := fmt.Sprintf(`select /* gh-ost */ count(*) as rows from %s.%s`, sql.EscapeName(this.migrationContext.DatabaseName), sql.EscapeName(this.migrationContext.OriginalTableName))
if err := this.db.QueryRow(query).Scan(&this.migrationContext.RowsEstimate); err != nil {
return err
}
this.migrationContext.UsedRowsEstimateMethod = base.CountRowsEstimate
log.Infof("Exact number of rows via COUNT: %d", this.migrationContext.RowsEstimate)
return nil
}
// validateTable makes sure the table we need to operate on actually exists
func (this *Inspector) validateTable() error {
query := fmt.Sprintf(`show /* gh-ost */ table status from %s like '%s'`, sql.EscapeName(this.migrationContext.DatabaseName), this.migrationContext.OriginalTableName)
tableFound := false
err := sqlutils.QueryRowsMap(this.db, query, func(rowMap sqlutils.RowMap) error {
this.migrationContext.TableEngine = rowMap.GetString("Engine")
this.migrationContext.RowsEstimate = rowMap.GetInt64("Rows")
this.migrationContext.UsedRowsEstimateMethod = base.TableStatusRowsEstimate
if rowMap.GetString("Comment") == "VIEW" {
return fmt.Errorf("%s.%s is a VIEW, not a real table. Bailing out", sql.EscapeName(this.migrationContext.DatabaseName), sql.EscapeName(this.migrationContext.OriginalTableName))
}
tableFound = true
return nil
})
if err != nil {
return err
}
if !tableFound {
return log.Errorf("Cannot find table %s.%s!", sql.EscapeName(this.migrationContext.DatabaseName), sql.EscapeName(this.migrationContext.OriginalTableName))
}
log.Infof("Table found. Engine=%s", this.migrationContext.TableEngine)
log.Debugf("Estimated number of rows via STATUS: %d", this.migrationContext.RowsEstimate)
return nil
}
// 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
}
// 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
}
// LockOriginalTable places a write lock on the original table
func (this *Applier) LockOriginalTable() error {
query := fmt.Sprintf(`lock /* gh-ost */ tables %s.%s write`,
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.OriginalTableName),
)
log.Infof("Locking %s.%s",
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.OriginalTableName),
)
this.migrationContext.LockTablesStartTime = time.Now()
if _, err := sqlutils.ExecNoPrepare(this.singletonDB, query); err != nil {
return err
}
log.Infof("Table locked")
return nil
}
// ChangeMasterCredentials issues a CHANGE MASTER TO... MASTER_USER=, MASTER_PASSWORD=...
func ChangeMasterCredentials(instanceKey *InstanceKey, masterUser string, masterPassword string) (*Instance, error) {
instance, err := ReadTopologyInstance(instanceKey)
if err != nil {
return instance, log.Errore(err)
}
if masterUser == "" {
return instance, log.Errorf("Empty user in ChangeMasterCredentials() for %+v", *instanceKey)
}
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 credentials on %+v", *instanceKey)
if *config.RuntimeCLIFlags.Noop {
return instance, fmt.Errorf("noop: aborting CHANGE MASTER TO operation on %+v; signalling error but nothing went wrong.", *instanceKey)
}
_, err = ExecInstanceNoPrepare(instanceKey, fmt.Sprintf("change master to master_user='******', master_password='******'",
masterUser, masterPassword))
if err != nil {
return instance, log.Errore(err)
}
log.Infof("ChangeMasterTo: Changed master credentials on %+v", *instanceKey)
instance, err = ReadTopologyInstance(instanceKey)
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
}
// 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
}
// EnableSemiSync sets the rpl_semi_sync_(master|slave)_enabled variables
// on a given instance.
func EnableSemiSync(instanceKey *InstanceKey, master, slave bool) error {
log.Infof("instance %+v rpl_semi_sync_master_enabled: %t, rpl_semi_sync_slave_enabled: %t", instanceKey, master, slave)
_, err := ExecInstanceNoPrepare(instanceKey,
`set global rpl_semi_sync_master_enabled = ?, global rpl_semi_sync_slave_enabled = ?`,
master, slave)
return err
}
// CreateGhostTable creates the ghost table on the applier host
func (this *Applier) CreateGhostTable() error {
query := fmt.Sprintf(`create /* gh-ost */ table %s.%s like %s.%s`,
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.GetGhostTableName()),
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.OriginalTableName),
)
log.Infof("Creating ghost table %s.%s",
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.GetGhostTableName()),
)
if _, err := sqlutils.ExecNoPrepare(this.db, query); err != nil {
return err
}
log.Infof("Ghost table created")
return nil
}
// AlterGhost applies `alter` statement on ghost table
func (this *Applier) AlterGhost() error {
query := fmt.Sprintf(`alter /* gh-ost */ table %s.%s %s`,
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.GetGhostTableName()),
this.migrationContext.AlterStatement,
)
log.Infof("Altering ghost table %s.%s",
sql.EscapeName(this.migrationContext.DatabaseName),
sql.EscapeName(this.migrationContext.GetGhostTableName()),
)
log.Debugf("ALTER statement: %s", query)
if _, err := sqlutils.ExecNoPrepare(this.db, query); err != nil {
return err
}
log.Infof("Ghost table altered")
return nil
}
// validateGrants verifies the user by which we're executing has necessary grants
// to do its thang.
func (this *Inspector) validateGrants() error {
query := `show /* gh-ost */ grants for current_user()`
foundAll := false
foundSuper := false
foundReplicationSlave := false
foundDBAll := false
err := sqlutils.QueryRowsMap(this.db, query, func(rowMap sqlutils.RowMap) error {
for _, grantData := range rowMap {
grant := grantData.String
if strings.Contains(grant, `GRANT ALL PRIVILEGES ON *.*`) {
foundAll = true
}
if strings.Contains(grant, `SUPER`) && strings.Contains(grant, ` ON *.*`) {
foundSuper = true
}
if strings.Contains(grant, `REPLICATION SLAVE`) && strings.Contains(grant, ` ON *.*`) {
foundReplicationSlave = true
}
if strings.Contains(grant, fmt.Sprintf("GRANT ALL PRIVILEGES ON `%s`.*", this.migrationContext.DatabaseName)) {
foundDBAll = true
}
if base.StringContainsAll(grant, `ALTER`, `CREATE`, `DELETE`, `DROP`, `INDEX`, `INSERT`, `LOCK TABLES`, `SELECT`, `TRIGGER`, `UPDATE`, ` ON *.*`) {
foundDBAll = true
}
if base.StringContainsAll(grant, `ALTER`, `CREATE`, `DELETE`, `DROP`, `INDEX`, `INSERT`, `LOCK TABLES`, `SELECT`, `TRIGGER`, `UPDATE`, fmt.Sprintf(" ON `%s`.*", this.migrationContext.DatabaseName)) {
foundDBAll = true
}
}
return nil
})
if err != nil {
return err
}
if foundAll {
log.Infof("User has ALL privileges")
return nil
}
if foundSuper && foundReplicationSlave && foundDBAll {
log.Infof("User has SUPER, REPLICATION SLAVE privileges, and has ALL privileges on `%s`", this.migrationContext.DatabaseName)
return nil
}
return log.Errorf("User has insufficient privileges for migration.")
}
// Inject the "AllEventsUpToLockProcessed" state hint, wait for it to appear in the binary logs,
// make sure the queue is drained.
func (this *Migrator) waitForEventsUpToLock() (err error) {
this.migrationContext.MarkPointOfInterest()
waitForEventsUpToLockStartTime := time.Now()
log.Infof("Writing changelog state: %+v", AllEventsUpToLockProcessed)
if _, err := this.applier.WriteChangelogState(string(AllEventsUpToLockProcessed)); err != nil {
return err
}
log.Infof("Waiting for events up to lock")
atomic.StoreInt64(&this.allEventsUpToLockProcessedInjectedFlag, 1)
<-this.allEventsUpToLockProcessed
waitForEventsUpToLockDuration := time.Now().Sub(waitForEventsUpToLockStartTime)
log.Infof("Done waiting for events up to lock; duration=%+v", waitForEventsUpToLockDuration)
this.printStatus(ForcePrintStatusAndHint)
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)
if instance.SemiSyncEnforced {
// Send ACK only from promotable instances.
sendACK := instance.PromotionRule != MustNotPromoteRule
// Always disable master setting, in case we're converting a former master.
if err := EnableSemiSync(instanceKey, false, sendACK); err != nil {
return instance, log.Errore(err)
}
}
// 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
}
// Http starts serving HTTP (api/web) requests
func Http() {
martini.Env = martini.Prod
m := martini.Classic()
if config.Config.HTTPAuthUser != "" {
m.Use(auth.Basic(config.Config.HTTPAuthUser, config.Config.HTTPAuthPassword))
}
m.Use(gzip.All())
// Render html templates from templates directory
m.Use(render.Renderer(render.Options{
Directory: "resources",
Layout: "templates/layout",
HTMLContentType: "text/html",
}))
m.Use(martini.Static("resources/public"))
if config.Config.UseMutualTLS {
m.Use(ssl.VerifyOUs(config.Config.SSLValidOUs))
}
go agent.ContinuousOperation()
log.Infof("Starting HTTP on port %d", config.Config.HTTPPort)
http.API.RegisterRequests(m)
listenAddress := fmt.Sprintf(":%d", config.Config.HTTPPort)
// Serve
if config.Config.UseSSL {
if len(config.Config.SSLCertFile) == 0 {
log.Fatale(errors.New("UseSSL is true but SSLCertFile is unspecified"))
}
if len(config.Config.SSLPrivateKeyFile) == 0 {
log.Fatale(errors.New("UseSSL is true but SSLPrivateKeyFile is unspecified"))
}
log.Info("Starting HTTPS listener")
tlsConfig, err := ssl.NewTLSConfig(config.Config.SSLCAFile, config.Config.UseMutualTLS)
if err != nil {
log.Fatale(err)
}
if err = ssl.AppendKeyPair(tlsConfig, config.Config.SSLCertFile, config.Config.SSLPrivateKeyFile); err != nil {
log.Fatale(err)
}
if err = ssl.ListenAndServeTLS(listenAddress, m, tlsConfig); err != nil {
log.Fatale(err)
}
} else {
log.Info("Starting HTTP listener")
if err := nethttp.ListenAndServe(listenAddress, m); err != nil {
log.Fatale(err)
}
}
log.Info("Web server started")
}
// ExpectUsedLock expects the special hint voluntary lock to exist on given session
func (this *Applier) ExpectUsedLock(sessionId int64) error {
var result int64
query := `select is_used_lock(?)`
lockName := this.GetSessionLockName(sessionId)
log.Infof("Checking session lock: %s", lockName)
if err := this.db.QueryRow(query, lockName).Scan(&result); err != nil || result != sessionId {
return fmt.Errorf("Session lock %s expected to be found but wasn't", lockName)
}
return nil
}