func Main() int {
if err := parse(os.Args[1:]); err != nil {
return 2
}
versionInfoTmpl.Execute(os.Stdout, BuildInfo)
if cfg.printVersion {
return 0
}
memStorage := local.NewMemorySeriesStorage(&cfg.storage)
var (
sampleAppender storage.SampleAppender
remoteStorageQueues []*remote.StorageQueueManager
)
if cfg.opentsdbURL == "" && cfg.influxdbURL == "" {
log.Warnf("No remote storage URLs provided; not sending any samples to long-term storage")
sampleAppender = memStorage
} else {
fanout := storage.Fanout{memStorage}
addRemoteStorage := func(c remote.StorageClient) {
qm := remote.NewStorageQueueManager(c, 100*1024)
fanout = append(fanout, qm)
remoteStorageQueues = append(remoteStorageQueues, qm)
}
if cfg.opentsdbURL != "" {
addRemoteStorage(opentsdb.NewClient(cfg.opentsdbURL, cfg.remoteStorageTimeout))
}
if cfg.influxdbURL != "" {
addRemoteStorage(influxdb.NewClient(cfg.influxdbURL, cfg.remoteStorageTimeout, cfg.influxdbDatabase, cfg.influxdbRetentionPolicy))
}
sampleAppender = fanout
}
var (
notificationHandler = notification.NewNotificationHandler(&cfg.notification)
targetManager = retrieval.NewTargetManager(sampleAppender)
queryEngine = promql.NewEngine(memStorage, &cfg.queryEngine)
)
ruleManager := rules.NewManager(&rules.ManagerOptions{
SampleAppender: sampleAppender,
NotificationHandler: notificationHandler,
QueryEngine: queryEngine,
PrometheusURL: cfg.prometheusURL,
PathPrefix: cfg.web.PathPrefix,
})
flags := map[string]string{}
cfg.fs.VisitAll(func(f *flag.Flag) {
flags[f.Name] = f.Value.String()
})
status := &web.PrometheusStatus{
BuildInfo: BuildInfo,
TargetPools: targetManager.Pools,
Rules: ruleManager.Rules,
Flags: flags,
Birth: time.Now(),
}
webHandler := web.New(memStorage, queryEngine, ruleManager, status, &cfg.web)
if !reloadConfig(cfg.configFile, status, targetManager, ruleManager) {
os.Exit(1)
}
// Wait for reload or termination signals. Start the handler for SIGHUP as
// early as possible, but ignore it until we are ready to handle reloading
// our config.
hup := make(chan os.Signal)
hupReady := make(chan bool)
signal.Notify(hup, syscall.SIGHUP)
go func() {
<-hupReady
for range hup {
reloadConfig(cfg.configFile, status, targetManager, ruleManager)
}
}()
// Start all components.
if err := memStorage.Start(); err != nil {
log.Errorln("Error opening memory series storage:", err)
return 1
}
defer func() {
if err := memStorage.Stop(); err != nil {
log.Errorln("Error stopping storage:", err)
}
}()
// The storage has to be fully initialized before registering.
registry.MustRegister(memStorage)
registry.MustRegister(notificationHandler)
for _, q := range remoteStorageQueues {
registry.MustRegister(q)
go q.Run()
defer q.Stop()
}
go ruleManager.Run()
defer ruleManager.Stop()
go notificationHandler.Run()
defer notificationHandler.Stop()
go targetManager.Run()
defer targetManager.Stop()
defer queryEngine.Stop()
go webHandler.Run()
// Wait for reload or termination signals.
close(hupReady) // Unblock SIGHUP handler.
term := make(chan os.Signal)
signal.Notify(term, os.Interrupt, syscall.SIGTERM)
select {
case <-term:
log.Warn("Received SIGTERM, exiting gracefully...")
case <-webHandler.Quit():
log.Warn("Received termination request via web service, exiting gracefully...")
}
close(hup)
log.Info("See you next time!")
return 0
}
// NewPrometheus creates a new prometheus object based on flag values.
// Call Serve() to start serving and Close() for clean shutdown.
func NewPrometheus() *prometheus {
conf, err := config.LoadFromFile(*configFile)
if err != nil {
glog.Fatalf("Error loading configuration from %s: %v", *configFile, err)
}
unwrittenSamples := make(chan clientmodel.Samples, *samplesQueueCapacity)
ingester := &retrieval.MergeLabelsIngester{
Labels: conf.GlobalLabels(),
CollisionPrefix: clientmodel.ExporterLabelPrefix,
Ingester: retrieval.ChannelIngester(unwrittenSamples),
}
targetManager := retrieval.NewTargetManager(ingester)
targetManager.AddTargetsFromConfig(conf)
notificationHandler := notification.NewNotificationHandler(*alertmanagerURL, *notificationQueueCapacity)
o := &local.MemorySeriesStorageOptions{
MemoryChunks: *numMemoryChunks,
PersistenceStoragePath: *metricsStoragePath,
PersistenceRetentionPeriod: *storageRetentionPeriod,
CheckpointInterval: *checkpointInterval,
CheckpointDirtySeriesLimit: *checkpointDirtySeriesLimit,
Dirty: *storageDirty,
}
memStorage, err := local.NewMemorySeriesStorage(o)
if err != nil {
glog.Fatal("Error opening memory series storage: ", err)
}
ruleManager := manager.NewRuleManager(&manager.RuleManagerOptions{
Results: unwrittenSamples,
NotificationHandler: notificationHandler,
EvaluationInterval: conf.EvaluationInterval(),
Storage: memStorage,
PrometheusURL: web.MustBuildServerURL(),
})
if err := ruleManager.AddRulesFromConfig(conf); err != nil {
glog.Fatal("Error loading rule files: ", err)
}
var remoteTSDBQueue *remote.TSDBQueueManager
if *remoteTSDBUrl == "" {
glog.Warningf("No TSDB URL provided; not sending any samples to long-term storage")
} else {
openTSDB := opentsdb.NewClient(*remoteTSDBUrl, *remoteTSDBTimeout)
remoteTSDBQueue = remote.NewTSDBQueueManager(openTSDB, 512)
}
flags := map[string]string{}
flag.VisitAll(func(f *flag.Flag) {
flags[f.Name] = f.Value.String()
})
prometheusStatus := &web.PrometheusStatusHandler{
BuildInfo: BuildInfo,
Config: conf.String(),
RuleManager: ruleManager,
TargetPools: targetManager.Pools(),
Flags: flags,
Birth: time.Now(),
}
alertsHandler := &web.AlertsHandler{
RuleManager: ruleManager,
}
consolesHandler := &web.ConsolesHandler{
Storage: memStorage,
}
metricsService := &api.MetricsService{
Config: &conf,
TargetManager: targetManager,
Storage: memStorage,
}
webService := &web.WebService{
StatusHandler: prometheusStatus,
MetricsHandler: metricsService,
ConsolesHandler: consolesHandler,
AlertsHandler: alertsHandler,
}
p := &prometheus{
unwrittenSamples: unwrittenSamples,
ruleManager: ruleManager,
targetManager: targetManager,
notificationHandler: notificationHandler,
storage: memStorage,
remoteTSDBQueue: remoteTSDBQueue,
webService: webService,
}
webService.QuitDelegate = p.Close
return p
}
// Main manages the startup and shutdown lifecycle of the entire Prometheus server.
func Main() int {
if err := parse(os.Args[1:]); err != nil {
return 2
}
printVersion()
if cfg.printVersion {
return 0
}
var reloadables []Reloadable
var (
memStorage = local.NewMemorySeriesStorage(&cfg.storage)
remoteStorage = remote.New(&cfg.remote)
sampleAppender = storage.Fanout{memStorage}
)
if remoteStorage != nil {
sampleAppender = append(sampleAppender, remoteStorage)
reloadables = append(reloadables, remoteStorage)
}
var (
notificationHandler = notification.NewNotificationHandler(&cfg.notification)
targetManager = retrieval.NewTargetManager(sampleAppender)
queryEngine = promql.NewEngine(memStorage, &cfg.queryEngine)
)
ruleManager := rules.NewManager(&rules.ManagerOptions{
SampleAppender: sampleAppender,
NotificationHandler: notificationHandler,
QueryEngine: queryEngine,
ExternalURL: cfg.web.ExternalURL,
})
flags := map[string]string{}
cfg.fs.VisitAll(func(f *flag.Flag) {
flags[f.Name] = f.Value.String()
})
status := &web.PrometheusStatus{
TargetPools: targetManager.Pools,
Rules: ruleManager.Rules,
Flags: flags,
Birth: time.Now(),
}
webHandler := web.New(memStorage, queryEngine, ruleManager, status, &cfg.web)
reloadables = append(reloadables, status, targetManager, ruleManager, webHandler, notificationHandler)
if !reloadConfig(cfg.configFile, reloadables...) {
return 1
}
// Wait for reload or termination signals. Start the handler for SIGHUP as
// early as possible, but ignore it until we are ready to handle reloading
// our config.
hup := make(chan os.Signal)
hupReady := make(chan bool)
signal.Notify(hup, syscall.SIGHUP)
go func() {
<-hupReady
for {
select {
case <-hup:
case <-webHandler.Reload():
}
reloadConfig(cfg.configFile, reloadables...)
}
}()
// Start all components.
if err := memStorage.Start(); err != nil {
log.Errorln("Error opening memory series storage:", err)
return 1
}
defer func() {
if err := memStorage.Stop(); err != nil {
log.Errorln("Error stopping storage:", err)
}
}()
if remoteStorage != nil {
prometheus.MustRegister(remoteStorage)
go remoteStorage.Run()
defer remoteStorage.Stop()
}
// The storage has to be fully initialized before registering.
prometheus.MustRegister(memStorage)
prometheus.MustRegister(notificationHandler)
prometheus.MustRegister(configSuccess)
prometheus.MustRegister(configSuccessTime)
go ruleManager.Run()
defer ruleManager.Stop()
go notificationHandler.Run()
defer notificationHandler.Stop()
go targetManager.Run()
defer targetManager.Stop()
defer queryEngine.Stop()
go webHandler.Run()
// Wait for reload or termination signals.
close(hupReady) // Unblock SIGHUP handler.
term := make(chan os.Signal)
signal.Notify(term, os.Interrupt, syscall.SIGTERM)
select {
case <-term:
log.Warn("Received SIGTERM, exiting gracefully...")
case <-webHandler.Quit():
log.Warn("Received termination request via web service, exiting gracefully...")
case err := <-webHandler.ListenError():
log.Errorln("Error starting web server, exiting gracefully:", err)
}
log.Info("See you next time!")
return 0
}
func main() {
// TODO(all): Future additions to main should be, where applicable, glumped
// into the prometheus struct above---at least where the scoping of the entire
// server is concerned.
flag.Parse()
versionInfoTmpl.Execute(os.Stdout, BuildInfo)
if *printVersion {
os.Exit(0)
}
conf, err := config.LoadFromFile(*configFile)
if err != nil {
glog.Fatalf("Error loading configuration from %s: %v", *configFile, err)
}
ts, err := tiered.NewTieredStorage(uint(*diskAppendQueueCapacity), 100, *arenaFlushInterval, *arenaTTL, *metricsStoragePath)
if err != nil {
glog.Fatal("Error opening storage: ", err)
}
var remoteTSDBQueue *remote.TSDBQueueManager = nil
if *remoteTSDBUrl == "" {
glog.Warningf("No TSDB URL provided; not sending any samples to long-term storage")
} else {
openTSDB := opentsdb.NewClient(*remoteTSDBUrl, *remoteTSDBTimeout)
remoteTSDBQueue = remote.NewTSDBQueueManager(openTSDB, 512)
go remoteTSDBQueue.Run()
}
unwrittenSamples := make(chan *extraction.Result, *samplesQueueCapacity)
ingester := &retrieval.MergeLabelsIngester{
Labels: conf.GlobalLabels(),
CollisionPrefix: clientmodel.ExporterLabelPrefix,
Ingester: retrieval.ChannelIngester(unwrittenSamples),
}
compactionTimer := time.NewTicker(*compactInterval)
deletionTimer := time.NewTicker(*deleteInterval)
// Queue depth will need to be exposed
targetManager := retrieval.NewTargetManager(ingester, *concurrentRetrievalAllowance)
targetManager.AddTargetsFromConfig(conf)
notifications := make(chan notification.NotificationReqs, *notificationQueueCapacity)
// Queue depth will need to be exposed
ruleManager := rules.NewRuleManager(&rules.RuleManagerOptions{
Results: unwrittenSamples,
Notifications: notifications,
EvaluationInterval: conf.EvaluationInterval(),
Storage: ts,
PrometheusUrl: web.MustBuildServerUrl(),
})
if err := ruleManager.AddRulesFromConfig(conf); err != nil {
glog.Fatal("Error loading rule files: ", err)
}
go ruleManager.Run()
notificationHandler := notification.NewNotificationHandler(*alertmanagerUrl, notifications)
go notificationHandler.Run()
flags := map[string]string{}
flag.VisitAll(func(f *flag.Flag) {
flags[f.Name] = f.Value.String()
})
prometheusStatus := &web.PrometheusStatusHandler{
BuildInfo: BuildInfo,
Config: conf.String(),
RuleManager: ruleManager,
TargetPools: targetManager.Pools(),
Flags: flags,
Birth: time.Now(),
}
alertsHandler := &web.AlertsHandler{
RuleManager: ruleManager,
}
databasesHandler := &web.DatabasesHandler{
Provider: ts.DiskStorage,
RefreshInterval: 5 * time.Minute,
}
metricsService := &api.MetricsService{
Config: &conf,
TargetManager: targetManager,
Storage: ts,
}
prometheus := &prometheus{
compactionTimer: compactionTimer,
deletionTimer: deletionTimer,
curationState: prometheusStatus,
curationSema: make(chan struct{}, 1),
unwrittenSamples: unwrittenSamples,
stopBackgroundOperations: make(chan struct{}),
ruleManager: ruleManager,
targetManager: targetManager,
notifications: notifications,
storage: ts,
remoteTSDBQueue: remoteTSDBQueue,
}
defer prometheus.Close()
webService := &web.WebService{
StatusHandler: prometheusStatus,
MetricsHandler: metricsService,
DatabasesHandler: databasesHandler,
AlertsHandler: alertsHandler,
QuitDelegate: prometheus.Close,
}
prometheus.curationSema <- struct{}{}
storageStarted := make(chan bool)
go ts.Serve(storageStarted)
<-storageStarted
go prometheus.interruptHandler()
go func() {
for _ = range prometheus.compactionTimer.C {
glog.Info("Starting compaction...")
err := prometheus.compact(*compactAgeInclusiveness, *compactGroupSize)
if err != nil {
glog.Error("could not compact: ", err)
}
glog.Info("Done")
}
}()
go func() {
for _ = range prometheus.deletionTimer.C {
glog.Info("Starting deletion of stale values...")
err := prometheus.delete(*deleteAge, deletionBatchSize)
if err != nil {
glog.Error("could not delete: ", err)
}
glog.Info("Done")
}
}()
go func() {
err := webService.ServeForever()
if err != nil {
glog.Fatal(err)
}
}()
// TODO(all): Migrate this into prometheus.serve().
for block := range unwrittenSamples {
if block.Err == nil && len(block.Samples) > 0 {
ts.AppendSamples(block.Samples)
if remoteTSDBQueue != nil {
remoteTSDBQueue.Queue(block.Samples)
}
}
}
}