예제 #1
0
func run() error {
	version := fmt.Sprintf("percona-agent %s rev %s", agent.VERSION, agent.REVISION)
	if flagVersion {
		fmt.Println(version)
		return nil
	}
	golog.Printf("Running %s pid %d\n", version, os.Getpid())

	if err := pct.Basedir.Init(flagBasedir); err != nil {
		return err
	}

	// Start-lock file is used to let agent1 self-update, create start-lock,
	// start updated agent2, exit cleanly, then agent2 starts.  agent1 may
	// not use a PID file, so this special file is required.
	if err := pct.WaitStartLock(); err != nil {
		return err
	}
	// NOTE: This must run last, and defer if LIFO, so it must be declared first.
	defer os.Remove(pct.Basedir.File("start-lock"))

	/**
	 * Agent config (require API key and agent UUID)
	 */

	if !pct.FileExists(pct.Basedir.ConfigFile("agent")) {
		return fmt.Errorf("Agent config file %s does not exist", pct.Basedir.ConfigFile("agent"))
	}

	bytes, err := agent.LoadConfig()
	if err != nil {
		return fmt.Errorf("Invalid agent config: %s\n", err)
	}
	agentConfig := &agent.Config{}
	if err := json.Unmarshal(bytes, agentConfig); err != nil {
		return fmt.Errorf("Error parsing "+pct.Basedir.ConfigFile("agent")+": ", err)
	}

	golog.Println("ApiHostname: " + agentConfig.ApiHostname)
	golog.Println("AgentUuid: " + agentConfig.AgentUuid)

	/**
	 * Ping and exit, maybe.
	 */

	if flagPing {
		t0 := time.Now()
		code, err := pct.Ping(agentConfig.ApiHostname, agentConfig.ApiKey)
		d := time.Now().Sub(t0)
		if err != nil || code != 200 {
			return fmt.Errorf("Ping FAIL (%d %d %s)", d, code, err)
		} else {
			golog.Printf("Ping OK (%s)", d)
			return nil
		}
	}

	/**
	 * PID file
	 */

	if flagPidFile != "" {
		pidFile := pct.NewPidFile()
		if err := pidFile.Set(flagPidFile); err != nil {
			golog.Fatalln(err)
		}
		defer pidFile.Remove()
	}

	/**
	 * REST API
	 */

	api, err := ConnectAPI(agentConfig)
	if err != nil {
		golog.Fatal(err)
	}

	/**
	 * Log relay
	 */

	logChan := make(chan *proto.LogEntry, log.BUFFER_SIZE*3)

	// Log websocket client, possibly disabled later.
	logClient, err := client.NewWebsocketClient(pct.NewLogger(logChan, "log-ws"), api, "log")
	if err != nil {
		golog.Fatalln(err)
	}
	logManager := log.NewManager(
		logClient,
		logChan,
	)
	if err := logManager.Start(); err != nil {
		return fmt.Errorf("Error starting logmanager: %s\n", err)
	}

	/**
	 * Instance manager
	 */

	itManager := instance.NewManager(
		pct.NewLogger(logChan, "instance-manager"),
		pct.Basedir.Dir("config"),
		api,
	)
	if err := itManager.Start(); err != nil {
		return fmt.Errorf("Error starting instance manager: %s\n", err)
	}

	/**
	 * Data spooler and sender
	 */

	hostname, _ := os.Hostname()

	dataClient, err := client.NewWebsocketClient(pct.NewLogger(logChan, "data-ws"), api, "data")
	if err != nil {
		golog.Fatalln(err)
	}
	dataManager := data.NewManager(
		pct.NewLogger(logChan, "data"),
		pct.Basedir.Dir("data"),
		hostname,
		dataClient,
	)
	if err := dataManager.Start(); err != nil {
		return fmt.Errorf("Error starting data manager: %s\n", err)
	}

	/**
	 * Collecct/report ticker (master clock)
	 */

	nowFunc := func() int64 { return time.Now().UTC().UnixNano() }
	clock := ticker.NewClock(&ticker.RealTickerFactory{}, nowFunc)

	/**
	 * Metric and system config monitors
	 */

	mmManager := mm.NewManager(
		pct.NewLogger(logChan, "mm"),
		mmMonitor.NewFactory(logChan, itManager.Repo()),
		clock,
		dataManager.Spooler(),
		itManager.Repo(),
	)
	if err := mmManager.Start(); err != nil {
		return fmt.Errorf("Error starting mm manager: %s\n", err)
	}

	sysconfigManager := sysconfig.NewManager(
		pct.NewLogger(logChan, "sysconfig"),
		sysconfigMonitor.NewFactory(logChan, itManager.Repo()),
		clock,
		dataManager.Spooler(),
		itManager.Repo(),
	)
	if err := sysconfigManager.Start(); err != nil {
		return fmt.Errorf("Error starting sysconfig manager: %s\n", err)
	}

	/**
	 * Query Analytics
	 */

	qanManager := qan.NewManager(
		pct.NewLogger(logChan, "qan"),
		&mysql.RealConnectionFactory{},
		clock,
		qan.NewFileIntervalIterFactory(logChan),
		qan.NewSlowLogWorkerFactory(logChan),
		dataManager.Spooler(),
		itManager.Repo(),
	)
	if err := qanManager.Start(); err != nil {
		return fmt.Errorf("Error starting qan manager: %s\n", err)
	}

	/**
	 * Signal handler
	 */

	// Generally the agent has a crash-only design, but QAN is so far the only service
	// which reconfigures MySQL: it enables the slow log, sets long_query_time, etc.
	// It's not terrible to leave slow log on, but it's nicer to turn it off.
	sigChan := make(chan os.Signal, 1)
	stopChan := make(chan error, 2)
	signal.Notify(sigChan, os.Interrupt, syscall.SIGTERM)
	go func() {
		sig := <-sigChan
		golog.Printf("Caught %s signal, shutting down...\n", sig)
		stopChan <- qanManager.Stop()
	}()

	/**
	 * Agent
	 */

	cmdClient, err := client.NewWebsocketClient(pct.NewLogger(logChan, "agent-ws"), api, "cmd")
	if err != nil {
		golog.Fatal(err)
	}

	// The official list of services known to the agent.  Adding a new service
	// requires a manager, starting the manager as above, and adding the manager
	// to this map.
	services := map[string]pct.ServiceManager{
		"log":       logManager,
		"data":      dataManager,
		"qan":       qanManager,
		"mm":        mmManager,
		"instance":  itManager,
		"sysconfig": sysconfigManager,
	}

	agent := agent.NewAgent(
		agentConfig,
		pct.NewLogger(logChan, "agent"),
		api,
		cmdClient,
		services,
	)

	/**
	 * Run agent, wait for it to stop or signal.
	 */

	go func() {
		stopChan <- agent.Run()
	}()
	stopErr := <-stopChan // agent or signal
	golog.Println("Agent stopped, shutting down...")
	qanManager.Stop()           // see Signal handler ^
	time.Sleep(2 * time.Second) // wait for final replies and log entries
	return stopErr
}
예제 #2
0
func run() error {
	version := fmt.Sprintf("percona-agent %s%s rev %s", agent.VERSION, agent.REL, agent.REVISION)
	if flagVersion {
		fmt.Println(version)
		return nil
	}
	golog.Printf("Running %s pid %d\n", version, os.Getpid())

	if err := pct.Basedir.Init(flagBasedir); err != nil {
		return err
	}

	// Start-lock file is used to let agent1 self-update, create start-lock,
	// start updated agent2, exit cleanly, then agent2 starts.  agent1 may
	// not use a PID file, so this special file is required.
	if err := pct.WaitStartLock(); err != nil {
		return err
	}
	// NOTE: This must run last, and defer if LIFO, so it must be declared first.
	defer os.Remove(pct.Basedir.File("start-lock"))

	/**
	 * Agent config (require API key and agent UUID)
	 */

	if !pct.FileExists(pct.Basedir.ConfigFile("agent")) {
		return fmt.Errorf("Agent config file %s does not exist", pct.Basedir.ConfigFile("agent"))
	}

	bytes, err := agent.LoadConfig()
	if err != nil {
		return fmt.Errorf("Invalid agent config: %s\n", err)
	}
	agentConfig := &agent.Config{}
	if err := json.Unmarshal(bytes, agentConfig); err != nil {
		return fmt.Errorf("Error parsing "+pct.Basedir.ConfigFile("agent")+": ", err)
	}

	golog.Println("ApiHostname: " + agentConfig.ApiHostname)
	golog.Println("AgentUuid: " + agentConfig.AgentUuid)

	/**
	 * Ping and exit, maybe.
	 */

	// Set for all connections to API.  X-Percona-API-Key is set automatically
	// using the pct.APIConnector.
	headers := map[string]string{
		"X-Percona-Agent-Version": agent.VERSION,
	}

	if flagPing {
		t0 := time.Now()
		code, err := pct.Ping(agentConfig.ApiHostname, agentConfig.ApiKey, headers)
		d := time.Now().Sub(t0)
		if err != nil || code != 200 {
			return fmt.Errorf("Ping FAIL (%d %d %s)", d, code, err)
		} else {
			golog.Printf("Ping OK (%s)", d)
			return nil
		}
	}

	/**
	 * PID file
	 */

	pidFilePath := agentConfig.PidFile
	if flagPidFile != "" {
		pidFilePath = flagPidFile
	}
	if pidFilePath != "" {
		pidFile := pct.NewPidFile()
		if err := pidFile.Set(pidFilePath); err != nil {
			golog.Fatalln(err)
		}
		defer pidFile.Remove()
	}

	/**
	 * REST API
	 */

	retry := -1 // unlimited
	if flagStatus {
		retry = 1
	}
	api, err := ConnectAPI(agentConfig, retry)
	if err != nil {
		golog.Fatal(err)
	}

	// Get agent status via API and exit.
	if flagStatus {
		code, bytes, err := api.Get(agentConfig.ApiKey, api.AgentLink("self")+"/status")
		if err != nil {
			return err
		}
		if code == 404 {
			return fmt.Errorf("Agent not found")
		}
		status := make(map[string]string)
		if err := json.Unmarshal(bytes, &status); err != nil {
			return err
		}
		golog.Println(status)
		return nil
	}

	/**
	 * Connection factory
	 */
	connFactory := &mysql.RealConnectionFactory{}

	/**
	 * Log relay
	 */

	logChan := make(chan *proto.LogEntry, log.BUFFER_SIZE*3)

	// Log websocket client, possibly disabled later.
	logClient, err := client.NewWebsocketClient(pct.NewLogger(logChan, "log-ws"), api, "log", headers)
	if err != nil {
		golog.Fatalln(err)
	}
	logManager := log.NewManager(
		logClient,
		logChan,
	)
	if err := logManager.Start(); err != nil {
		return fmt.Errorf("Error starting logmanager: %s\n", err)
	}

	/**
	 * MRMS (MySQL Restart Monitoring Service)
	 */
	mrm := mrmsMonitor.NewMonitor(
		pct.NewLogger(logChan, "mrms-monitor"),
		connFactory,
	)
	mrmsManager := mrms.NewManager(
		pct.NewLogger(logChan, "mrms-manager"),
		mrm,
	)
	if err := mrmsManager.Start(); err != nil {
		return fmt.Errorf("Error starting mrms manager: %s\n", err)
	}

	/**
	 * Instance manager
	 */
	itManager := instance.NewManager(
		pct.NewLogger(logChan, "instance-manager"),
		pct.Basedir.Dir("config"),
		api,
		mrm,
	)
	if err := itManager.Start(); err != nil {
		return fmt.Errorf("Error starting instance manager: %s\n", err)
	}

	/**
	 * Data spooler and sender
	 */

	hostname, _ := os.Hostname()

	dataClient, err := client.NewWebsocketClient(pct.NewLogger(logChan, "data-ws"), api, "data", headers)
	if err != nil {
		golog.Fatalln(err)
	}
	dataManager := data.NewManager(
		pct.NewLogger(logChan, "data"),
		pct.Basedir.Dir("data"),
		pct.Basedir.Dir("trash"),
		hostname,
		dataClient,
	)
	if err := dataManager.Start(); err != nil {
		return fmt.Errorf("Error starting data manager: %s\n", err)
	}

	/**
	 * Collecct/report ticker (master clock)
	 */

	nowFunc := func() int64 { return time.Now().UTC().UnixNano() }
	clock := ticker.NewClock(&ticker.RealTickerFactory{}, nowFunc)

	/**
	 * Metric and system config monitors
	 */

	mmManager := mm.NewManager(
		pct.NewLogger(logChan, "mm"),
		mmMonitor.NewFactory(logChan, itManager.Repo(), mrm),
		clock,
		dataManager.Spooler(),
		itManager.Repo(),
		mrm,
	)
	if err := mmManager.Start(); err != nil {
		return fmt.Errorf("Error starting mm manager: %s\n", err)
	}

	sysconfigManager := sysconfig.NewManager(
		pct.NewLogger(logChan, "sysconfig"),
		sysconfigMonitor.NewFactory(logChan, itManager.Repo()),
		clock,
		dataManager.Spooler(),
		itManager.Repo(),
	)
	if err := sysconfigManager.Start(); err != nil {
		return fmt.Errorf("Error starting sysconfig manager: %s\n", err)
	}

	/**
	 * Query service (real-time EXPLAIN, SHOW CREATE TABLE, etc.)
	 */

	queryManager := query.NewManager(
		pct.NewLogger(logChan, "query"),
		itManager.Repo(),
		&mysql.RealConnectionFactory{},
	)
	if err := queryManager.Start(); err != nil {
		return fmt.Errorf("Error starting query manager: %s\n", err)
	}

	/**
	 * Query Analytics
	 */

	qanManager := qan.NewManager(
		pct.NewLogger(logChan, "qan"),
		clock,
		itManager.Repo(),
		mrm,
		connFactory,
		qanFactory.NewRealAnalyzerFactory(
			logChan,
			qanFactory.NewRealIntervalIterFactory(logChan),
			slowlog.NewRealWorkerFactory(logChan),
			perfschema.NewRealWorkerFactory(logChan),
			dataManager.Spooler(),
			clock,
		),
	)

	if err := qanManager.Start(); err != nil {
		return fmt.Errorf("Error starting qan manager: %s\n", err)
	}

	/**
	 * Sysinfo
	 */
	sysinfoManager := sysinfo.NewManager(
		pct.NewLogger(logChan, "sysinfo"),
	)

	// MySQL Sysinfo
	mysqlSysinfoService := mysqlSysinfo.NewMySQL(
		pct.NewLogger(logChan, "sysinfo-mysql"),
		itManager.Repo(),
	)
	if err := sysinfoManager.RegisterService("MySQLSummary", mysqlSysinfoService); err != nil {
		return fmt.Errorf("Error registering Mysql Sysinfo service: %s\n", err)
	}

	// System Sysinfo
	systemSysinfoService := systemSysinfo.NewSystem(
		pct.NewLogger(logChan, "sysinfo-system"),
	)
	if err := sysinfoManager.RegisterService("SystemSummary", systemSysinfoService); err != nil {
		return fmt.Errorf("Error registering System Sysinfo service: %s\n", err)
	}

	// Start Sysinfo manager
	if err := sysinfoManager.Start(); err != nil {
		return fmt.Errorf("Error starting Sysinfo manager: %s\n", err)
	}

	/**
	 * Signal handler
	 */

	// Generally the agent has a crash-only design, but QAN is so far the only service
	// which reconfigures MySQL: it enables the slow log, sets long_query_time, etc.
	// It's not terrible to leave slow log on, but it's nicer to turn it off.
	sigChan := make(chan os.Signal, 1)
	stopChan := make(chan error, 2)
	signal.Notify(sigChan, os.Interrupt, syscall.SIGTERM)
	go func() {
		sig := <-sigChan
		golog.Printf("Caught %s signal, shutting down...\n", sig)
		stopChan <- qanManager.Stop()
	}()

	/**
	 * Agent
	 */

	cmdClient, err := client.NewWebsocketClient(pct.NewLogger(logChan, "agent-ws"), api, "cmd", headers)
	if err != nil {
		golog.Fatal(err)
	}

	// The official list of services known to the agent.  Adding a new service
	// requires a manager, starting the manager as above, and adding the manager
	// to this map.
	services := map[string]pct.ServiceManager{
		"log":       logManager,
		"data":      dataManager,
		"qan":       qanManager,
		"mm":        mmManager,
		"instance":  itManager,
		"mrms":      mrmsManager,
		"sysconfig": sysconfigManager,
		"query":     queryManager,
		"sysinfo":   sysinfoManager,
	}

	// Set the global pct/cmd.Factory, used for the Restart cmd.
	pctCmd.Factory = &pctCmd.RealCmdFactory{}

	agentLogger := pct.NewLogger(logChan, "agent")

	agent := agent.NewAgent(
		agentConfig,
		agentLogger,
		api,
		cmdClient,
		services,
	)

	/**
	 * Run agent, wait for it to stop, signal, or crash.
	 */

	var stopErr error
	go func() {
		defer func() {
			if err := recover(); err != nil {
				errMsg := fmt.Sprintf("Agent crashed: %s", err)
				golog.Println(errMsg)
				agentLogger.Error(errMsg)
				stopChan <- fmt.Errorf("%s", errMsg)
			}
		}()
		stopChan <- agent.Run()
	}()

	// Wait for agent to stop, or for signals.
	agentRunning := true
	statusSigChan := make(chan os.Signal, 1)
	signal.Notify(statusSigChan, syscall.SIGUSR1) // kill -USER1 PID
	reconnectSigChan := make(chan os.Signal, 1)
	signal.Notify(reconnectSigChan, syscall.SIGHUP) // kill -HUP PID
	for agentRunning {
		select {
		case stopErr = <-stopChan: // agent or signal
			golog.Println("Agent stopped, shutting down...")
			agentLogger.Info("Agent stopped")
			agentRunning = false
		case <-statusSigChan:
			status := agent.AllStatus()
			golog.Printf("Status: %+v\n", status)
		case <-reconnectSigChan:
			u, _ := user.Current()
			cmd := &proto.Cmd{
				Ts:        time.Now().UTC(),
				User:      u.Username + " (SIGHUP)",
				AgentUuid: agentConfig.AgentUuid,
				Service:   "agent",
				Cmd:       "Reconnect",
			}
			agent.Handle(cmd)
		}
	}

	qanManager.Stop()           // see Signal handler ^
	time.Sleep(2 * time.Second) // wait for final replies and log entries
	return stopErr
}