// 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)
}

// 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)
	}
}

// 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
}

// 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='%s', master_password='%s'",
		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
}

// 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")
}