Ejemplo n.º 1
0
// Stop implements Storage.
func (s *memorySeriesStorage) Stop() error {
	log.Info("Stopping local storage...")

	log.Info("Stopping maintenance loop...")
	close(s.loopStopping)
	<-s.loopStopped

	log.Info("Stopping series quarantining...")
	close(s.quarantineStopping)
	<-s.quarantineStopped

	log.Info("Stopping chunk eviction...")
	close(s.evictStopping)
	<-s.evictStopped

	// One final checkpoint of the series map and the head chunks.
	if err := s.persistence.checkpointSeriesMapAndHeads(s.fpToSeries, s.fpLocker); err != nil {
		return err
	}
	if err := s.mapper.checkpoint(); err != nil {
		return err
	}

	if err := s.persistence.close(); err != nil {
		return err
	}
	log.Info("Local storage stopped.")
	return nil
}
Ejemplo n.º 2
0
func (p *persistence) rebuildLabelIndexes(
	fpToSeries map[model.Fingerprint]*memorySeries,
) error {
	count := 0
	log.Info("Rebuilding label indexes.")
	log.Info("Indexing metrics in memory.")
	for fp, s := range fpToSeries {
		p.indexMetric(fp, s.metric)
		count++
		if count%10000 == 0 {
			log.Infof("%d metrics queued for indexing.", count)
		}
	}
	log.Info("Indexing archived metrics.")
	var fp codable.Fingerprint
	var m codable.Metric
	if err := p.archivedFingerprintToMetrics.ForEach(func(kv index.KeyValueAccessor) error {
		if err := kv.Key(&fp); err != nil {
			return err
		}
		if err := kv.Value(&m); err != nil {
			return err
		}
		p.indexMetric(model.Fingerprint(fp), model.Metric(m))
		count++
		if count%10000 == 0 {
			log.Infof("%d metrics queued for indexing.", count)
		}
		return nil
	}); err != nil {
		return err
	}
	log.Info("All requests for rebuilding the label indexes queued. (Actual processing may lag behind.)")
	return nil
}
Ejemplo n.º 3
0
// Stop the rule manager's rule evaluation cycles.
func (m *Manager) Stop() {
	log.Info("Stopping rule manager...")

	for _, eg := range m.groups {
		eg.stop()
	}

	log.Info("Rule manager stopped.")
}
Ejemplo n.º 4
0
// Run starts background processing to handle target updates.
func (tm *TargetManager) Run() {
	log.Info("Starting target manager...")

	tm.done = make(chan struct{})

	sources := map[string]struct{}{}
	updates := []<-chan targetGroupUpdate{}

	for scfg, provs := range tm.providers {
		for _, prov := range provs {
			// Get an initial set of available sources so we don't remove
			// target groups from the last run that are still available.
			for _, src := range prov.Sources() {
				sources[src] = struct{}{}
			}

			tgc := make(chan config.TargetGroup)
			// Run the target provider after cleanup of the stale targets is done.
			defer func(prov TargetProvider, tgc chan<- config.TargetGroup, done <-chan struct{}) {
				go prov.Run(tgc, done)
			}(prov, tgc, tm.done)

			tgupc := make(chan targetGroupUpdate)
			updates = append(updates, tgupc)

			go func(scfg *config.ScrapeConfig, done <-chan struct{}) {
				defer close(tgupc)
				for {
					select {
					case tg := <-tgc:
						tgupc <- targetGroupUpdate{tg: tg, scfg: scfg}
					case <-done:
						return
					}
				}
			}(scfg, tm.done)
		}
	}

	// Merge all channels of incoming target group updates into a single
	// one and keep applying the updates.
	go tm.handleUpdates(merge(tm.done, updates...), tm.done)

	tm.mtx.Lock()
	defer tm.mtx.Unlock()

	// Remove old target groups that are no longer in the set of sources.
	tm.removeTargets(func(src string) bool {
		if _, ok := sources[src]; ok {
			return false
		}
		return true
	})

	tm.running = true
	log.Info("Target manager started.")
}
Ejemplo n.º 5
0
// Stop the rule manager's rule evaluation cycles.
func (m *Manager) Stop() {
	m.mtx.Lock()
	defer m.mtx.Unlock()

	log.Info("Stopping rule manager...")

	for _, eg := range m.groups {
		eg.stop()
	}

	log.Info("Rule manager stopped.")
}
Ejemplo n.º 6
0
// stop background processing of the target manager. If removeTargets is true,
// existing targets will be stopped and removed.
func (tm *TargetManager) stop(removeTargets bool) {
	log.Info("Stopping target manager...")
	defer log.Info("Target manager stopped.")

	close(tm.done)

	tm.mtx.Lock()
	defer tm.mtx.Unlock()

	if removeTargets {
		tm.removeTargets(nil)
	}

	tm.running = false
}
Ejemplo n.º 7
0
func main() {
	config := createConfig()

	log.Info("Create provisioner...")
	provisioner, err := provisioner.NewProvisioner(config.Token)
	if err != nil {
		log.Fatal(err)
	}
	go backgroundRunner(provisioner)

	log.Info("Create server...")
	server := web.CreateServer(provisioner)

	log.Infof("Listen on %s", config.Port)
	log.Fatal(http.ListenAndServe(":"+config.Port, server))
}
Ejemplo n.º 8
0
func interruptHandler(l net.Listener) {
	notifier := make(chan os.Signal)
	signal.Notify(notifier, os.Interrupt, syscall.SIGTERM)
	<-notifier
	log.Info("Received SIGINT/SIGTERM; exiting gracefully...")
	l.Close()
}
Ejemplo n.º 9
0
// Stop stops sending samples to the remote storage and waits for pending
// sends to complete.
func (t *StorageQueueManager) Stop() {
	log.Infof("Stopping remote storage...")
	for _, shard := range t.shards {
		close(shard)
	}
	t.wg.Wait()
	log.Info("Remote storage stopped.")
}
Ejemplo n.º 10
0
func (s *memorySeriesStorage) loop() {
	checkpointTimer := time.NewTimer(s.checkpointInterval)

	dirtySeriesCount := 0

	defer func() {
		checkpointTimer.Stop()
		log.Info("Maintenance loop stopped.")
		close(s.loopStopped)
	}()

	memoryFingerprints := s.cycleThroughMemoryFingerprints()
	archivedFingerprints := s.cycleThroughArchivedFingerprints()

loop:
	for {
		select {
		case <-s.loopStopping:
			break loop
		case <-checkpointTimer.C:
			err := s.persistence.checkpointSeriesMapAndHeads(s.fpToSeries, s.fpLocker)
			if err != nil {
				log.Errorln("Error while checkpointing:", err)
			} else {
				dirtySeriesCount = 0
			}
			// If a checkpoint takes longer than checkpointInterval, unluckily timed
			// combination with the Reset(0) call below can lead to a case where a
			// time is lurking in C leading to repeated checkpointing without break.
			select {
			case <-checkpointTimer.C: // Get rid of the lurking time.
			default:
			}
			checkpointTimer.Reset(s.checkpointInterval)
		case fp := <-memoryFingerprints:
			if s.maintainMemorySeries(fp, model.Now().Add(-s.dropAfter)) {
				dirtySeriesCount++
				// Check if we have enough "dirty" series so that we need an early checkpoint.
				// However, if we are already behind persisting chunks, creating a checkpoint
				// would be counterproductive, as it would slow down chunk persisting even more,
				// while in a situation like that, where we are clearly lacking speed of disk
				// maintenance, the best we can do for crash recovery is to persist chunks as
				// quickly as possible. So only checkpoint if the urgency score is < 1.
				if dirtySeriesCount >= s.checkpointDirtySeriesLimit &&
					s.calculatePersistenceUrgencyScore() < 1 {
					checkpointTimer.Reset(0)
				}
			}
		case fp := <-archivedFingerprints:
			s.maintainArchivedSeries(fp, model.Now().Add(-s.dropAfter))
		}
	}
	// Wait until both channels are closed.
	for range memoryFingerprints {
	}
	for range archivedFingerprints {
	}
}
Ejemplo n.º 11
0
// Stop stops sending samples to the remote storage and waits for pending
// sends to complete.
func (t *StorageQueueManager) Stop() {
	log.Infof("Stopping remote storage...")
	close(t.queue)
	<-t.drained
	for i := 0; i < maxConcurrentSends; i++ {
		t.sendSemaphore <- true
	}
	log.Info("Remote storage stopped.")
}
Ejemplo n.º 12
0
// Start implements Storage.
func (s *memorySeriesStorage) Start() (err error) {
	var syncStrategy syncStrategy
	switch s.options.SyncStrategy {
	case Never:
		syncStrategy = func() bool { return false }
	case Always:
		syncStrategy = func() bool { return true }
	case Adaptive:
		syncStrategy = func() bool { return s.calculatePersistenceUrgencyScore() < 1 }
	default:
		panic("unknown sync strategy")
	}

	var p *persistence
	p, err = newPersistence(
		s.options.PersistenceStoragePath,
		s.options.Dirty, s.options.PedanticChecks,
		syncStrategy,
		s.options.MinShrinkRatio,
	)
	if err != nil {
		return err
	}
	s.persistence = p
	// Persistence must start running before loadSeriesMapAndHeads() is called.
	go s.persistence.run()

	defer func() {
		if err != nil {
			if e := p.close(); e != nil {
				log.Errorln("Error closing persistence:", e)
			}
		}
	}()

	log.Info("Loading series map and head chunks...")
	s.fpToSeries, s.numChunksToPersist, err = p.loadSeriesMapAndHeads()
	if err != nil {
		return err
	}
	log.Infof("%d series loaded.", s.fpToSeries.length())
	s.numSeries.Set(float64(s.fpToSeries.length()))

	s.mapper, err = newFPMapper(s.fpToSeries, p)
	if err != nil {
		return err
	}

	go s.handleEvictList()
	go s.handleQuarantine()
	go s.logThrottling()
	go s.loop()

	return nil
}
Ejemplo n.º 13
0
// Run starts background processing to handle target updates.
func (tm *TargetManager) Run() {
	log.Info("Starting target manager...")

	tm.mtx.Lock()

	tm.ctx, tm.cancel = context.WithCancel(context.Background())
	tm.reload()

	tm.mtx.Unlock()

	tm.wg.Wait()
}
Ejemplo n.º 14
0
func main() {
	var (
		bindURI       = flag.String("bind.stats-url", "http://localhost:8053/", "HTTP XML API address of an Bind server.")
		bindTimeout   = flag.Duration("bind.timeout", 10*time.Second, "Timeout for trying to get stats from Bind.")
		bindPidFile   = flag.String("bind.pid-file", "", "Path to Bind's pid file to export process information.")
		bindVersion   = flag.String("bind.stats-version", "auto", "BIND statistics version. Can be detected automatically. Available: [xml.v2, xml.v3, auto]")
		showVersion   = flag.Bool("version", false, "Print version information.")
		listenAddress = flag.String("web.listen-address", ":9119", "Address to listen on for web interface and telemetry.")
		metricsPath   = flag.String("web.telemetry-path", "/metrics", "Path under which to expose metrics.")

		groups = statisticGroups{bind.ServerStats, bind.ViewStats}
	)
	flag.Var(&groups, "bind.stats-groups", "Comma-separated list of statistics to collect. Available: [server, view, tasks]")
	flag.Parse()

	if *showVersion {
		fmt.Fprintln(os.Stdout, version.Print("bind_exporter"))
		os.Exit(0)
	}
	log.Infoln("Starting bind_exporter", version.Info())
	log.Infoln("Build context", version.BuildContext())
	log.Infoln("Configured to collect statistics", groups.String())

	prometheus.MustRegister(NewExporter(*bindVersion, *bindURI, *bindTimeout, groups))
	if *bindPidFile != "" {
		procExporter := prometheus.NewProcessCollectorPIDFn(
			func() (int, error) {
				content, err := ioutil.ReadFile(*bindPidFile)
				if err != nil {
					return 0, fmt.Errorf("Can't read pid file: %s", err)
				}
				value, err := strconv.Atoi(strings.TrimSpace(string(content)))
				if err != nil {
					return 0, fmt.Errorf("Can't parse pid file: %s", err)
				}
				return value, nil
			}, namespace)
		prometheus.MustRegister(procExporter)
	}

	log.Info("Starting Server: ", *listenAddress)
	http.Handle(*metricsPath, prometheus.Handler())
	http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
		w.Write([]byte(`<html>
             <head><title>Bind Exporter</title></head>
             <body>
             <h1>Bind Exporter</h1>
             <p><a href='` + *metricsPath + `'>Metrics</a></p>
             </body>
             </html>`))
	})
	log.Fatal(http.ListenAndServe(*listenAddress, nil))
}
Ejemplo n.º 15
0
func (s *memorySeriesStorage) loop() {
	checkpointTimer := time.NewTimer(s.checkpointInterval)

	dirtySeriesCount := 0

	defer func() {
		checkpointTimer.Stop()
		log.Info("Maintenance loop stopped.")
		close(s.loopStopped)
	}()

	memoryFingerprints := s.cycleThroughMemoryFingerprints()
	archivedFingerprints := s.cycleThroughArchivedFingerprints()

loop:
	for {
		select {
		case <-s.loopStopping:
			break loop
		case <-checkpointTimer.C:
			err := s.persistence.checkpointSeriesMapAndHeads(s.fpToSeries, s.fpLocker)
			if err != nil {
				log.Errorln("Error while checkpointing:", err)
			} else {
				dirtySeriesCount = 0
			}
			checkpointTimer.Reset(s.checkpointInterval)
		case fp := <-memoryFingerprints:
			if s.maintainMemorySeries(fp, model.Now().Add(-s.dropAfter)) {
				dirtySeriesCount++
				// Check if we have enough "dirty" series so that we need an early checkpoint.
				// However, if we are already behind persisting chunks, creating a checkpoint
				// would be counterproductive, as it would slow down chunk persisting even more,
				// while in a situation like that, where we are clearly lacking speed of disk
				// maintenance, the best we can do for crash recovery is to persist chunks as
				// quickly as possible. So only checkpoint if the storage is not in "graceful
				// degradation mode".
				if dirtySeriesCount >= s.checkpointDirtySeriesLimit && !s.isDegraded() {
					checkpointTimer.Reset(0)
				}
			}
		case fp := <-archivedFingerprints:
			s.maintainArchivedSeries(fp, model.Now().Add(-s.dropAfter))
		}
	}
	// Wait until both channels are closed.
	for range memoryFingerprints {
	}
	for range archivedFingerprints {
	}
}
Ejemplo n.º 16
0
func testChunk(t *testing.T, encoding chunkEncoding) {
	samples := make(model.Samples, 500000)
	for i := range samples {
		samples[i] = &model.Sample{
			Timestamp: model.Time(i),
			Value:     model.SampleValue(float64(i) * 0.2),
		}
	}
	s, closer := NewTestStorage(t, encoding)
	defer closer.Close()

	for _, sample := range samples {
		s.Append(sample)
	}
	s.WaitForIndexing()

	for m := range s.fpToSeries.iter() {
		s.fpLocker.Lock(m.fp)
		defer s.fpLocker.Unlock(m.fp) // TODO remove, see below
		var values []model.SamplePair
		for _, cd := range m.series.chunkDescs {
			if cd.isEvicted() {
				continue
			}
			it := cd.c.newIterator()
			for it.scan() {
				values = append(values, it.value())
			}
			if it.err() != nil {
				t.Error(it.err())
			}
		}

		for i, v := range values {
			if samples[i].Timestamp != v.Timestamp {
				t.Errorf("%d. Got %v; want %v", i, v.Timestamp, samples[i].Timestamp)
			}
			if samples[i].Value != v.Value {
				t.Errorf("%d. Got %v; want %v", i, v.Value, samples[i].Value)
			}
		}
		//s.fpLocker.Unlock(m.fp)
	}
	log.Info("test done, closing")
}
Ejemplo n.º 17
0
func (s *memorySeriesStorage) handleQuarantine() {
	for {
		select {
		case req := <-s.quarantineRequests:
			s.purgeSeries(req.fp, req.metric, req.reason)
			log.
				With("fingerprint", req.fp).
				With("metric", req.metric).
				With("reason", req.reason).
				Warn("Series quarantined.")
		case <-s.quarantineStopping:
			log.Info("Series quarantining stopped.")
			close(s.quarantineStopped)
			return
		}
	}

}
Ejemplo n.º 18
0
func (s *memorySeriesStorage) handleEvictList() {
	ticker := time.NewTicker(maxEvictInterval)
	count := 0

	for {
		// To batch up evictions a bit, this tries evictions at least
		// once per evict interval, but earlier if the number of evict
		// requests with evict==true that have happened since the last
		// evict run is more than maxMemoryChunks/1000.
		select {
		case req := <-s.evictRequests:
			if req.evict {
				req.cd.evictListElement = s.evictList.PushBack(req.cd)
				count++
				if count > s.maxMemoryChunks/1000 {
					s.maybeEvict()
					count = 0
				}
			} else {
				if req.cd.evictListElement != nil {
					s.evictList.Remove(req.cd.evictListElement)
					req.cd.evictListElement = nil
				}
			}
		case <-ticker.C:
			if s.evictList.Len() > 0 {
				s.maybeEvict()
			}
		case <-s.evictStopping:
			// Drain evictRequests forever in a goroutine to not let
			// requesters hang.
			go func() {
				for {
					<-s.evictRequests
				}
			}()
			ticker.Stop()
			log.Info("Chunk eviction stopped.")
			close(s.evictStopped)
			return
		}
	}
}
Ejemplo n.º 19
0
// NewDiskMetricStore returns a DiskMetricStore ready to use. To cleanly shut it
// down and free resources, the Shutdown() method has to be called.  If
// persistenceFile is the empty string, no persisting to disk will
// happen. Otherwise, a file of that name is used for persisting metrics to
// disk. If the file already exists, metrics are read from it as part of the
// start-up. Persisting is happening upon shutdown and after every write action,
// but the latter will only happen persistenceDuration after the previous
// persisting.
func NewDiskMetricStore(
	persistenceFile string,
	persistenceInterval time.Duration,
) *DiskMetricStore {
	dms := &DiskMetricStore{
		writeQueue:      make(chan WriteRequest, writeQueueCapacity),
		drain:           make(chan struct{}),
		done:            make(chan error),
		metricGroups:    GroupingKeyToMetricGroup{},
		persistenceFile: persistenceFile,
	}
	if err := dms.restore(); err != nil {
		log.Errorln("Could not load persisted metrics:", err)
		log.Info("Retrying assuming legacy format for persisted metrics...")
		if err := dms.legacyRestore(); err != nil {
			log.Errorln("Could not load persisted metrics in legacy format: ", err)
		}
	}

	go dms.loop(persistenceInterval)
	return dms
}
Ejemplo n.º 20
0
// Run the rule manager's periodic rule evaluation.
func (m *Manager) Run() {
	defer log.Info("Rule manager stopped.")

	m.Lock()
	lastInterval := m.interval
	m.Unlock()

	ticker := time.NewTicker(lastInterval)
	defer ticker.Stop()

	for {
		// The outer select clause makes sure that m.done is looked at
		// first. Otherwise, if m.runIteration takes longer than
		// m.interval, there is only a 50% chance that m.done will be
		// looked at before the next m.runIteration call happens.
		select {
		case <-m.done:
			return
		default:
			select {
			case <-ticker.C:
				start := time.Now()
				m.runIteration()
				iterationDuration.Observe(float64(time.Since(start) / time.Millisecond))

				m.Lock()
				if lastInterval != m.interval {
					ticker.Stop()
					ticker = time.NewTicker(m.interval)
					lastInterval = m.interval
				}
				m.Unlock()
			case <-m.done:
				return
			}
		}
	}
}
Ejemplo n.º 21
0
func main() {
	flag.Parse()
	uri, err := url.Parse(*marathonUri)
	if err != nil {
		log.Fatal(err)
	}

	retryTimeout := time.Duration(10 * time.Second)
	for {
		err := marathonConnect(uri)
		if err == nil {
			break
		}

		log.Debugf("Problem connecting to Marathon: %v", err)
		log.Infof("Couldn't connect to Marathon! Trying again in %v", retryTimeout)
		time.Sleep(retryTimeout)
	}

	exporter := NewExporter(&scraper{uri})
	prometheus.MustRegister(exporter)

	http.Handle(*metricsPath, prometheus.Handler())
	http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
		w.Write([]byte(`<html>
             <head><title>Marathon Exporter</title></head>
             <body>
             <h1>Marathon Exporter</h1>
             <p><a href='` + *metricsPath + `'>Metrics</a></p>
             </body>
             </html>`))
	})

	log.Info("Starting Server: ", *listenAddress)
	log.Fatal(http.ListenAndServe(*listenAddress, nil))
}
Ejemplo n.º 22
0
// checkpointFPMappings persists the fingerprint mappings. The caller has to
// ensure that the provided mappings are not changed concurrently. This method
// is only called upon shutdown or during crash recovery, when no samples are
// ingested.
//
// Description of the file format, v1:
//
// (1) Magic string (const mappingsMagicString).
//
// (2) Uvarint-encoded format version (const mappingsFormatVersion).
//
// (3) Uvarint-encoded number of mappings in fpMappings.
//
// (4) Repeated once per mapping:
//
// (4.1) The raw fingerprint as big-endian uint64.
//
// (4.2) The uvarint-encoded number of sub-mappings for the raw fingerprint.
//
// (4.3) Repeated once per sub-mapping:
//
// (4.3.1) The uvarint-encoded length of the unique metric string.
// (4.3.2) The unique metric string.
// (4.3.3) The mapped fingerprint as big-endian uint64.
func (p *persistence) checkpointFPMappings(fpm fpMappings) (err error) {
	log.Info("Checkpointing fingerprint mappings...")
	begin := time.Now()
	f, err := os.OpenFile(p.mappingsTempFileName(), os.O_WRONLY|os.O_TRUNC|os.O_CREATE, 0640)
	if err != nil {
		return
	}

	defer func() {
		syncErr := f.Sync()
		closeErr := f.Close()
		if err != nil {
			return
		}
		err = syncErr
		if err != nil {
			return
		}
		err = closeErr
		if err != nil {
			return
		}
		err = os.Rename(p.mappingsTempFileName(), p.mappingsFileName())
		duration := time.Since(begin)
		log.Infof("Done checkpointing fingerprint mappings in %v.", duration)
	}()

	w := bufio.NewWriterSize(f, fileBufSize)

	if _, err = w.WriteString(mappingsMagicString); err != nil {
		return
	}
	if _, err = codable.EncodeUvarint(w, mappingsFormatVersion); err != nil {
		return
	}
	if _, err = codable.EncodeUvarint(w, uint64(len(fpm))); err != nil {
		return
	}

	for fp, mappings := range fpm {
		if err = codable.EncodeUint64(w, uint64(fp)); err != nil {
			return
		}
		if _, err = codable.EncodeUvarint(w, uint64(len(mappings))); err != nil {
			return
		}
		for ms, mappedFP := range mappings {
			if _, err = codable.EncodeUvarint(w, uint64(len(ms))); err != nil {
				return
			}
			if _, err = w.WriteString(ms); err != nil {
				return
			}
			if err = codable.EncodeUint64(w, uint64(mappedFP)); err != nil {
				return
			}
		}
	}
	err = w.Flush()
	return
}
Ejemplo n.º 23
0
// Main manages the startup and shutdown lifecycle of the entire Prometheus server.
func Main() int {
	if err := parse(os.Args[1:]); err != nil {
		log.Error(err)
		return 2
	}

	if cfg.printVersion {
		fmt.Fprintln(os.Stdout, version.Print("prometheus"))
		return 0
	}

	log.Infoln("Starting prometheus", version.Info())
	log.Infoln("Build context", version.BuildContext())

	var (
		sampleAppender = storage.Fanout{}
		reloadables    []Reloadable
	)

	var localStorage local.Storage
	switch cfg.localStorageEngine {
	case "persisted":
		localStorage = local.NewMemorySeriesStorage(&cfg.storage)
		sampleAppender = storage.Fanout{localStorage}
	case "none":
		localStorage = &local.NoopStorage{}
	default:
		log.Errorf("Invalid local storage engine %q", cfg.localStorageEngine)
		return 1
	}

	remoteStorage, err := remote.New(&cfg.remote)
	if err != nil {
		log.Errorf("Error initializing remote storage: %s", err)
		return 1
	}
	if remoteStorage != nil {
		sampleAppender = append(sampleAppender, remoteStorage)
		reloadables = append(reloadables, remoteStorage)
	}

	reloadableRemoteStorage := remote.NewConfigurable()
	sampleAppender = append(sampleAppender, reloadableRemoteStorage)
	reloadables = append(reloadables, reloadableRemoteStorage)

	var (
		notifier       = notifier.New(&cfg.notifier)
		targetManager  = retrieval.NewTargetManager(sampleAppender)
		queryEngine    = promql.NewEngine(localStorage, &cfg.queryEngine)
		ctx, cancelCtx = context.WithCancel(context.Background())
	)

	ruleManager := rules.NewManager(&rules.ManagerOptions{
		SampleAppender: sampleAppender,
		Notifier:       notifier,
		QueryEngine:    queryEngine,
		Context:        ctx,
		ExternalURL:    cfg.web.ExternalURL,
	})

	cfg.web.Context = ctx
	cfg.web.Storage = localStorage
	cfg.web.QueryEngine = queryEngine
	cfg.web.TargetManager = targetManager
	cfg.web.RuleManager = ruleManager

	cfg.web.Version = &web.PrometheusVersion{
		Version:   version.Version,
		Revision:  version.Revision,
		Branch:    version.Branch,
		BuildUser: version.BuildUser,
		BuildDate: version.BuildDate,
		GoVersion: version.GoVersion,
	}

	cfg.web.Flags = map[string]string{}
	cfg.fs.VisitAll(func(f *flag.Flag) {
		cfg.web.Flags[f.Name] = f.Value.String()
	})

	webHandler := web.New(&cfg.web)

	reloadables = append(reloadables, targetManager, ruleManager, webHandler, notifier)

	if err := reloadConfig(cfg.configFile, reloadables...); err != nil {
		log.Errorf("Error loading config: %s", err)
		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:
				if err := reloadConfig(cfg.configFile, reloadables...); err != nil {
					log.Errorf("Error reloading config: %s", err)
				}
			case rc := <-webHandler.Reload():
				if err := reloadConfig(cfg.configFile, reloadables...); err != nil {
					log.Errorf("Error reloading config: %s", err)
					rc <- err
				} else {
					rc <- nil
				}
			}
		}
	}()

	// Start all components. The order is NOT arbitrary.

	if err := localStorage.Start(); err != nil {
		log.Errorln("Error opening memory series storage:", err)
		return 1
	}
	defer func() {
		if err := localStorage.Stop(); err != nil {
			log.Errorln("Error stopping storage:", err)
		}
	}()

	if remoteStorage != nil {
		remoteStorage.Start()
		defer remoteStorage.Stop()
	}

	defer reloadableRemoteStorage.Stop()

	// The storage has to be fully initialized before registering.
	if instrumentedStorage, ok := localStorage.(prometheus.Collector); ok {
		prometheus.MustRegister(instrumentedStorage)
	}
	prometheus.MustRegister(notifier)
	prometheus.MustRegister(configSuccess)
	prometheus.MustRegister(configSuccessTime)

	// The notifier is a dependency of the rule manager. It has to be
	// started before and torn down afterwards.
	go notifier.Run()
	defer notifier.Stop()

	go ruleManager.Run()
	defer ruleManager.Stop()

	go targetManager.Run()
	defer targetManager.Stop()

	// Shutting down the query engine before the rule manager will cause pending queries
	// to be canceled and ensures a quick shutdown of the rule manager.
	defer cancelCtx()

	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
}
Ejemplo n.º 24
0
// recoverFromCrash is called by loadSeriesMapAndHeads if the persistence
// appears to be dirty after the loading (either because the loading resulted in
// an error or because the persistence was dirty from the start). Not goroutine
// safe. Only call before anything else is running (except index processing
// queue as started by newPersistence).
func (p *persistence) recoverFromCrash(fingerprintToSeries map[model.Fingerprint]*memorySeries) error {
	// TODO(beorn): We need proper tests for the crash recovery.
	log.Warn("Starting crash recovery. Prometheus is inoperational until complete.")
	log.Warn("To avoid crash recovery in the future, shut down Prometheus with SIGTERM or a HTTP POST to /-/quit.")

	fpsSeen := map[model.Fingerprint]struct{}{}
	count := 0
	seriesDirNameFmt := fmt.Sprintf("%%0%dx", seriesDirNameLen)

	// Delete the fingerprint mapping file as it might be stale or
	// corrupt. We'll rebuild the mappings as we go.
	if err := os.RemoveAll(p.mappingsFileName()); err != nil {
		return fmt.Errorf("couldn't remove old fingerprint mapping file %s: %s", p.mappingsFileName(), err)
	}
	// The mappings to rebuild.
	fpm := fpMappings{}

	log.Info("Scanning files.")
	for i := 0; i < 1<<(seriesDirNameLen*4); i++ {
		dirname := filepath.Join(p.basePath, fmt.Sprintf(seriesDirNameFmt, i))
		dir, err := os.Open(dirname)
		if os.IsNotExist(err) {
			continue
		}
		if err != nil {
			return err
		}
		for fis := []os.FileInfo{}; err != io.EOF; fis, err = dir.Readdir(1024) {
			if err != nil {
				dir.Close()
				return err
			}
			for _, fi := range fis {
				fp, ok := p.sanitizeSeries(dirname, fi, fingerprintToSeries, fpm)
				if ok {
					fpsSeen[fp] = struct{}{}
				}
				count++
				if count%10000 == 0 {
					log.Infof("%d files scanned.", count)
				}
			}
		}
		dir.Close()
	}
	log.Infof("File scan complete. %d series found.", len(fpsSeen))

	log.Info("Checking for series without series file.")
	for fp, s := range fingerprintToSeries {
		if _, seen := fpsSeen[fp]; !seen {
			// fp exists in fingerprintToSeries, but has no representation on disk.
			if s.persistWatermark == len(s.chunkDescs) {
				// Oops, everything including the head chunk was
				// already persisted, but nothing on disk.
				// Thus, we lost that series completely. Clean
				// up the remnants.
				delete(fingerprintToSeries, fp)
				if err := p.purgeArchivedMetric(fp); err != nil {
					// Purging the archived metric didn't work, so try
					// to unindex it, just in case it's in the indexes.
					p.unindexMetric(fp, s.metric)
				}
				log.Warnf("Lost series detected: fingerprint %v, metric %v.", fp, s.metric)
				continue
			}
			// If we are here, the only chunks we have are the chunks in the checkpoint.
			// Adjust things accordingly.
			if s.persistWatermark > 0 || s.chunkDescsOffset != 0 {
				minLostChunks := s.persistWatermark + s.chunkDescsOffset
				if minLostChunks <= 0 {
					log.Warnf(
						"Possible loss of chunks for fingerprint %v, metric %v.",
						fp, s.metric,
					)
				} else {
					log.Warnf(
						"Lost at least %d chunks for fingerprint %v, metric %v.",
						minLostChunks, fp, s.metric,
					)
				}
				s.chunkDescs = append(
					make([]*chunk.Desc, 0, len(s.chunkDescs)-s.persistWatermark),
					s.chunkDescs[s.persistWatermark:]...,
				)
				chunk.NumMemDescs.Sub(float64(s.persistWatermark))
				s.persistWatermark = 0
				s.chunkDescsOffset = 0
			}
			maybeAddMapping(fp, s.metric, fpm)
			fpsSeen[fp] = struct{}{} // Add so that fpsSeen is complete.
		}
	}
	log.Info("Check for series without series file complete.")

	if err := p.cleanUpArchiveIndexes(fingerprintToSeries, fpsSeen, fpm); err != nil {
		return err
	}
	if err := p.rebuildLabelIndexes(fingerprintToSeries); err != nil {
		return err
	}
	// Finally rewrite the mappings file if there are any mappings.
	if len(fpm) > 0 {
		if err := p.checkpointFPMappings(fpm); err != nil {
			return err
		}
	}

	p.dirtyMtx.Lock()
	// Only declare storage clean if it didn't become dirty during crash recovery.
	if !p.becameDirty {
		p.dirty = false
	}
	p.dirtyMtx.Unlock()

	log.Warn("Crash recovery complete.")
	return nil
}
Ejemplo n.º 25
0
func (p *persistence) cleanUpArchiveIndexes(
	fpToSeries map[model.Fingerprint]*memorySeries,
	fpsSeen map[model.Fingerprint]struct{},
	fpm fpMappings,
) error {
	log.Info("Cleaning up archive indexes.")
	var fp codable.Fingerprint
	var m codable.Metric
	count := 0
	if err := p.archivedFingerprintToMetrics.ForEach(func(kv index.KeyValueAccessor) error {
		count++
		if count%10000 == 0 {
			log.Infof("%d archived metrics checked.", count)
		}
		if err := kv.Key(&fp); err != nil {
			return err
		}
		_, fpSeen := fpsSeen[model.Fingerprint(fp)]
		inMemory := false
		if fpSeen {
			_, inMemory = fpToSeries[model.Fingerprint(fp)]
		}
		if !fpSeen || inMemory {
			if inMemory {
				log.Warnf("Archive clean-up: Fingerprint %v is not archived. Purging from archive indexes.", model.Fingerprint(fp))
			}
			if !fpSeen {
				log.Warnf("Archive clean-up: Fingerprint %v is unknown. Purging from archive indexes.", model.Fingerprint(fp))
			}
			// It's fine if the fp is not in the archive indexes.
			if _, err := p.archivedFingerprintToMetrics.Delete(fp); err != nil {
				return err
			}
			// Delete from timerange index, too.
			_, err := p.archivedFingerprintToTimeRange.Delete(fp)
			return err
		}
		// fp is legitimately archived. Now we need the metric to check for a mapped fingerprint.
		if err := kv.Value(&m); err != nil {
			return err
		}
		maybeAddMapping(model.Fingerprint(fp), model.Metric(m), fpm)
		// Make sure it is in timerange index, too.
		has, err := p.archivedFingerprintToTimeRange.Has(fp)
		if err != nil {
			return err
		}
		if has {
			return nil // All good.
		}
		log.Warnf("Archive clean-up: Fingerprint %v is not in time-range index. Unarchiving it for recovery.")
		// Again, it's fine if fp is not in the archive index.
		if _, err := p.archivedFingerprintToMetrics.Delete(fp); err != nil {
			return err
		}
		cds, err := p.loadChunkDescs(model.Fingerprint(fp), 0)
		if err != nil {
			return err
		}
		series, err := newMemorySeries(model.Metric(m), cds, p.seriesFileModTime(model.Fingerprint(fp)))
		if err != nil {
			return err
		}
		fpToSeries[model.Fingerprint(fp)] = series
		return nil
	}); err != nil {
		return err
	}
	count = 0
	if err := p.archivedFingerprintToTimeRange.ForEach(func(kv index.KeyValueAccessor) error {
		count++
		if count%10000 == 0 {
			log.Infof("%d archived time ranges checked.", count)
		}
		if err := kv.Key(&fp); err != nil {
			return err
		}
		has, err := p.archivedFingerprintToMetrics.Has(fp)
		if err != nil {
			return err
		}
		if has {
			return nil // All good.
		}
		log.Warnf("Archive clean-up: Purging unknown fingerprint %v in time-range index.", fp)
		deleted, err := p.archivedFingerprintToTimeRange.Delete(fp)
		if err != nil {
			return err
		}
		if !deleted {
			log.Errorf("Fingerprint %v to be deleted from archivedFingerprintToTimeRange not found. This should never happen.", fp)
		}
		return nil
	}); err != nil {
		return err
	}
	log.Info("Clean-up of archive indexes complete.")
	return nil
}
Ejemplo n.º 26
0
// 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.New(&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)

	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
}
Ejemplo n.º 27
0
// Stop shuts down the notification handler.
func (n *Notifier) Stop() {
	log.Info("Stopping notification handler...")
	n.cancel()
}
Ejemplo n.º 28
0
// Stop shuts down the notification handler.
func (n *Handler) Stop() {
	log.Info("Stopping notification handler...")

	close(n.more)
	n.cancel()
}
Ejemplo n.º 29
0
// checkpointSeriesMapAndHeads persists the fingerprint to memory-series mapping
// and all non persisted chunks. Do not call concurrently with
// loadSeriesMapAndHeads. This method will only write heads format v2, but
// loadSeriesMapAndHeads can also understand v1.
//
// Description of the file format (for both, v1 and v2):
//
// (1) Magic string (const headsMagicString).
//
// (2) Varint-encoded format version (const headsFormatVersion).
//
// (3) Number of series in checkpoint as big-endian uint64.
//
// (4) Repeated once per series:
//
// (4.1) A flag byte, see flag constants above. (Present but unused in v2.)
//
// (4.2) The fingerprint as big-endian uint64.
//
// (4.3) The metric as defined by codable.Metric.
//
// (4.4) The varint-encoded persistWatermark. (Missing in v1.)
//
// (4.5) The modification time of the series file as nanoseconds elapsed since
// January 1, 1970 UTC. -1 if the modification time is unknown or no series file
// exists yet. (Missing in v1.)
//
// (4.6) The varint-encoded chunkDescsOffset.
//
// (4.6) The varint-encoded savedFirstTime.
//
// (4.7) The varint-encoded number of chunk descriptors.
//
// (4.8) Repeated once per chunk descriptor, oldest to most recent, either
// variant 4.8.1 (if index < persistWatermark) or variant 4.8.2 (if index >=
// persistWatermark). In v1, everything is variant 4.8.1 except for a
// non-persisted head-chunk (determined by the flags).
//
// (4.8.1.1) The varint-encoded first time.
// (4.8.1.2) The varint-encoded last time.
//
// (4.8.2.1) A byte defining the chunk type.
// (4.8.2.2) The chunk itself, marshaled with the marshal() method.
//
func (p *persistence) checkpointSeriesMapAndHeads(fingerprintToSeries *seriesMap, fpLocker *fingerprintLocker) (err error) {
	log.Info("Checkpointing in-memory metrics and chunks...")
	begin := time.Now()
	f, err := os.OpenFile(p.headsTempFileName(), os.O_WRONLY|os.O_TRUNC|os.O_CREATE, 0640)
	if err != nil {
		return err
	}

	defer func() {
		syncErr := f.Sync()
		closeErr := f.Close()
		if err != nil {
			return
		}
		err = syncErr
		if err != nil {
			return
		}
		err = closeErr
		if err != nil {
			return
		}
		err = os.Rename(p.headsTempFileName(), p.headsFileName())
		duration := time.Since(begin)
		p.checkpointDuration.Set(duration.Seconds())
		log.Infof("Done checkpointing in-memory metrics and chunks in %v.", duration)
	}()

	w := bufio.NewWriterSize(f, fileBufSize)

	if _, err = w.WriteString(headsMagicString); err != nil {
		return err
	}
	var numberOfSeriesOffset int
	if numberOfSeriesOffset, err = codable.EncodeVarint(w, headsFormatVersion); err != nil {
		return err
	}
	numberOfSeriesOffset += len(headsMagicString)
	numberOfSeriesInHeader := uint64(fingerprintToSeries.length())
	// We have to write the number of series as uint64 because we might need
	// to overwrite it later, and a varint might change byte width then.
	if err = codable.EncodeUint64(w, numberOfSeriesInHeader); err != nil {
		return err
	}

	iter := fingerprintToSeries.iter()
	defer func() {
		// Consume the iterator in any case to not leak goroutines.
		for range iter {
		}
	}()

	var realNumberOfSeries uint64
	for m := range iter {
		func() { // Wrapped in function to use defer for unlocking the fp.
			fpLocker.Lock(m.fp)
			defer fpLocker.Unlock(m.fp)

			if len(m.series.chunkDescs) == 0 {
				// This series was completely purged or archived in the meantime. Ignore.
				return
			}
			realNumberOfSeries++
			// seriesFlags left empty in v2.
			if err = w.WriteByte(0); err != nil {
				return
			}
			if err = codable.EncodeUint64(w, uint64(m.fp)); err != nil {
				return
			}
			var buf []byte
			buf, err = codable.Metric(m.series.metric).MarshalBinary()
			if err != nil {
				return
			}
			if _, err = w.Write(buf); err != nil {
				return
			}
			if _, err = codable.EncodeVarint(w, int64(m.series.persistWatermark)); err != nil {
				return
			}
			if m.series.modTime.IsZero() {
				if _, err = codable.EncodeVarint(w, -1); err != nil {
					return
				}
			} else {
				if _, err = codable.EncodeVarint(w, m.series.modTime.UnixNano()); err != nil {
					return
				}
			}
			if _, err = codable.EncodeVarint(w, int64(m.series.chunkDescsOffset)); err != nil {
				return
			}
			if _, err = codable.EncodeVarint(w, int64(m.series.savedFirstTime)); err != nil {
				return
			}
			if _, err = codable.EncodeVarint(w, int64(len(m.series.chunkDescs))); err != nil {
				return
			}
			for i, chunkDesc := range m.series.chunkDescs {
				if i < m.series.persistWatermark {
					if _, err = codable.EncodeVarint(w, int64(chunkDesc.firstTime())); err != nil {
						return
					}
					lt, err := chunkDesc.lastTime()
					if err != nil {
						return
					}
					if _, err = codable.EncodeVarint(w, int64(lt)); err != nil {
						return
					}
				} else {
					// This is a non-persisted chunk. Fully marshal it.
					if err = w.WriteByte(byte(chunkDesc.c.encoding())); err != nil {
						return
					}
					if err = chunkDesc.c.marshal(w); err != nil {
						return
					}
				}
			}
			// Series is checkpointed now, so declare it clean. In case the entire
			// checkpoint fails later on, this is fine, as the storage's series
			// maintenance will mark these series newly dirty again, continuously
			// increasing the total number of dirty series as seen by the storage.
			// This has the effect of triggering a new checkpoint attempt even
			// earlier than if we hadn't incorrectly set "dirty" to "false" here
			// already.
			m.series.dirty = false
		}()
		if err != nil {
			return err
		}
	}
	if err = w.Flush(); err != nil {
		return err
	}
	if realNumberOfSeries != numberOfSeriesInHeader {
		// The number of series has changed in the meantime.
		// Rewrite it in the header.
		if _, err = f.Seek(int64(numberOfSeriesOffset), os.SEEK_SET); err != nil {
			return err
		}
		if err = codable.EncodeUint64(f, realNumberOfSeries); err != nil {
			return err
		}
	}
	return err
}
Ejemplo n.º 30
0
// Main manages the startup and shutdown lifecycle of the entire Prometheus server.
func Main() int {
	if err := parse(os.Args[1:]); err != nil {
		log.Error(err)
		return 2
	}

	if cfg.printVersion {
		fmt.Fprintln(os.Stdout, version.Print("prometheus"))
		return 0
	}

	log.Infoln("Starting prometheus", version.Info())
	log.Infoln("Build context", version.BuildContext())

	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 (
		notifier      = notifier.New(&cfg.notifier)
		targetManager = retrieval.NewTargetManager(sampleAppender)
		queryEngine   = promql.NewEngine(memStorage, &cfg.queryEngine)
	)

	ruleManager := rules.NewManager(&rules.ManagerOptions{
		SampleAppender: sampleAppender,
		Notifier:       notifier,
		QueryEngine:    queryEngine,
		ExternalURL:    cfg.web.ExternalURL,
	})

	flags := map[string]string{}
	cfg.fs.VisitAll(func(f *flag.Flag) {
		flags[f.Name] = f.Value.String()
	})

	version := &web.PrometheusVersion{
		Version:   version.Version,
		Revision:  version.Revision,
		Branch:    version.Branch,
		BuildUser: version.BuildUser,
		BuildDate: version.BuildDate,
		GoVersion: version.GoVersion,
	}

	webHandler := web.New(memStorage, queryEngine, targetManager, ruleManager, version, flags, &cfg.web)

	reloadables = append(reloadables, targetManager, ruleManager, webHandler, notifier)

	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. The order is NOT arbitrary.

	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(notifier)
	prometheus.MustRegister(configSuccess)
	prometheus.MustRegister(configSuccessTime)

	// The notifieris a dependency of the rule manager. It has to be
	// started before and torn down afterwards.
	go notifier.Run()
	defer notifier.Stop()

	go ruleManager.Run()
	defer ruleManager.Stop()

	go targetManager.Run()
	defer targetManager.Stop()

	// Shutting down the query engine before the rule manager will cause pending queries
	// to be canceled and ensures a quick shutdown of the rule manager.
	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
}