func (i *Installer) Run() error {
/**
* Check for pt-agent, upgrade if found.
*/
var ptagentDSN *mysql.DSN
ptagentUpgrade := false
ptagentConf := "/root/.pt-agent.conf"
if pct.FileExists(ptagentConf) {
fmt.Println("Found pt-agent, upgrading and removing because it is no longer supported...")
ptagentUpgrade = true
// Stop pt-agent
if err := StopPTAgent(); err != nil {
fmt.Printf("Error stopping pt-agent: %s\n\n", err)
fmt.Println("WARNING: pt-agent must be stopped before installing percona-agent. " +
"Please verify that pt-agent is not running and has been removed from cron. " +
"Enter 'Y' to confirm and continue installing percona-agent.")
ok, err := i.term.PromptBool("pt-agent has stopped?", "N")
if err != nil {
return err
}
if !ok {
return fmt.Errorf("Failed to stop pt-agent")
}
}
// Get its settings (API key, UUID, etc.).
agent, dsn, err := GetPTAgentSettings(ptagentConf)
if err != nil {
return fmt.Errorf("Error upgrading pt-agent: %s", err)
}
if agent.ApiKey != "" {
i.agentConfig.ApiKey = agent.ApiKey
}
if agent.AgentUuid != "" {
i.agentConfig.AgentUuid = agent.AgentUuid
fmt.Printf("Upgrading pt-agent %s...\n", agent.AgentUuid)
}
ptagentDSN = dsn
}
/**
* Get the API key.
*/
fmt.Printf("API host: %s\n", i.agentConfig.ApiHostname)
for i.agentConfig.ApiKey == "" {
apiKey, err := i.term.PromptString("API key", "")
if err != nil {
return err
}
if apiKey == "" {
fmt.Println("API key is required, please try again.")
continue
}
i.agentConfig.ApiKey = apiKey
break
}
/**
* Verify the API key by pinging the API.
*/
VERIFY_API_KEY:
for {
startTime := time.Now()
fmt.Printf("Verifying API key %s...\n", i.agentConfig.ApiKey)
code, err := pct.Ping(i.agentConfig.ApiHostname, i.agentConfig.ApiKey)
elapsedTime := time.Since(startTime)
elapsedTimeInSeconds := elapsedTime / time.Second
timeout := false
if urlErr, ok := err.(*url.Error); ok {
if netOpErr, ok := urlErr.Err.(*net.OpError); ok && netOpErr.Timeout() {
timeout = true
}
}
if i.flags["debug"] {
log.Printf("code=%d\n", code)
log.Printf("err=%s\n", err)
}
ok := false
if timeout {
fmt.Printf(
"Error: API connection timeout (%ds): %s\n"+
"Before you try again, please check your connection and DNS configuration.\n",
elapsedTimeInSeconds,
err,
)
} else if err != nil {
fmt.Printf("Error: %s\n", err)
} else if code >= 500 {
fmt.Printf("Sorry, there's an API problem (status code %d). "+
"Please try to install again. If the problem continues, contact Percona.\n",
code)
} else if code == 401 {
return fmt.Errorf("Access denied. Check the API key and try again.")
} else if code >= 300 {
fmt.Printf("Sorry, there's an installer problem (status code %d). "+
"Please try to install again. If the problem continues, contact Percona.\n",
code)
} else if code != 200 {
fmt.Printf("Sorry, there's an installer problem (status code %d). "+
"Please try to install again. If the problem continues, contact Percona.\n",
code)
} else {
ok = true
}
if !ok {
again, err := i.term.PromptBool("Try again?", "Y")
if err != nil {
return err
}
if !again {
return fmt.Errorf("Failed to verify API key")
}
continue VERIFY_API_KEY
}
fmt.Printf("API key %s is OK\n", i.agentConfig.ApiKey)
if elapsedTimeInSeconds >= 0 {
fmt.Printf(
"WARNING: We have detected that request to api took %d second(-s) while usually it shouldn't take more than 1s.\n"+
"This might be due to connection problems or slow DNS resolution.\n"+
"Before you continue please check your connection and DNS configuration as this might impact performance of percona-agent.\n"+
"If you are using CentOS or Fedora 19+ in a vagrant box then you might be interested in this bug report:\n"+
"https://github.com/mitchellh/vagrant/issues/1172\n",
elapsedTimeInSeconds,
)
proceed, err := i.term.PromptBool("Continue anyway?", "Y")
if err != nil {
return err
}
if !proceed {
return fmt.Errorf("Failed because of slow connection")
}
}
break
}
var si *proto.ServerInstance
var mi *proto.MySQLInstance
/**
* Create new service instances.
*/
var err error
if i.flags["create-server-instance"] {
si, err = i.createServerInstance()
if err != nil {
return err
}
fmt.Printf("Created server instance: hostname=%s id=%d\n", si.Hostname, si.Id)
} else {
fmt.Println("Not creating server instance (-create-server-instance=false)")
}
if i.flags["create-mysql-instance"] {
// Create MySQL user for agent, or using existing one, then verify MySQL connection.
agentDSN, err := i.doMySQL(ptagentDSN)
if err != nil {
return err
}
// Create MySQL instance in API.
mi, err = i.createMySQLInstance(agentDSN)
if err != nil {
return err
}
fmt.Printf("Created MySQL instance: dsn=%s hostname=%s id=%d\n", mi.DSN, mi.Hostname, si.Id)
} else {
fmt.Println("Not creating MySQL instance (-create-mysql-instance=false)")
}
if err := i.writeInstances(si, mi); err != nil {
return fmt.Errorf("Created agent but failed to write service instances: %s", err)
}
/**
* Get default configs for all services.
*/
configs := []proto.AgentConfig{}
if i.flags["start-services"] {
// Server metrics monitor
config, err := i.getMmServerConfig(si)
if err != nil {
fmt.Println(err)
fmt.Println("WARNING: cannot start server metrics monitor")
} else {
configs = append(configs, *config)
}
if i.flags["start-mysql-services"] {
// MySQL metrics tracker
config, err = i.getMmMySQLConfig(mi)
if err != nil {
fmt.Println(err)
fmt.Println("WARNING: cannot start MySQL metrics monitor")
} else {
configs = append(configs, *config)
}
// MySQL config tracker
config, err = i.getSysconfigMySQLConfig(mi)
if err != nil {
fmt.Println(err)
fmt.Println("WARNING: cannot start MySQL configuration monitor")
} else {
configs = append(configs, *config)
}
// QAN
// MySQL is local if the server hostname == MySQL hostname without port number.
if i.hostname == portNumberRe.ReplaceAllLiteralString(mi.Hostname, "") {
if i.flags["debug"] {
log.Printf("MySQL is local")
}
config, err := i.getQanConfig(mi)
if err != nil {
fmt.Println(err)
fmt.Println("WARNING: cannot start Query Analytics")
} else {
configs = append(configs, *config)
}
}
} else {
fmt.Println("Not starting MySQL services (-start-mysql-services=false)")
}
} else {
fmt.Println("Not starting default services (-start-services=false)")
}
/**
* Create agent with initial service configs.
*/
if ptagentUpgrade {
agent, err := i.updateAgent(i.agentConfig.AgentUuid)
if err != nil {
return err
}
fmt.Println("pt-agent upgraded to percona-agent")
if err := i.writeConfigs(agent, configs); err != nil {
return fmt.Errorf("Upgraded pt-agent but failed to write percona-agent configs: %s", err)
}
} else if i.flags["create-agent"] {
agent, err := i.createAgent(configs)
if err != nil {
return err
}
fmt.Printf("Created agent: uuid=%s\n", agent.Uuid)
if err := i.writeConfigs(agent, configs); err != nil {
return fmt.Errorf("Created agent but failed to write configs: %s", err)
}
} else {
fmt.Println("Not creating agent (-create-agent=false)")
}
/**
* Remove pt-agent if upgrading.
*/
if ptagentUpgrade {
RemovePTAgent(ptagentConf)
fmt.Println("pt-agent removed")
}
return nil // success
}
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
}
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
}