Preserve target state across reloads.

This commit moves Scraper handling into a separate scrapePool type.
TargetSets only manage TargetProvider lifecycles and sync the
retrieved updates to the scrapePool.

TargetProviders are now expected to send a full initial target set
within 5 seconds. The scrapePools preserve target state across reloads
and only drop targets after the initial set was synced.

retrieval/discovery/dns.go

@@ -113,6 +113,7 @@ func (dd *DNSDiscovery) Run(ctx context.Context, ch chan<- []*config.TargetGroup
 
 func (dd *DNSDiscovery) refreshAll(ch chan<- []*config.TargetGroup) {
 	var wg sync.WaitGroup
+
 	wg.Add(len(dd.names))
 	for _, name := range dd.names {
 		go func(n string) {
@@ -122,6 +123,7 @@ func (dd *DNSDiscovery) refreshAll(ch chan<- []*config.TargetGroup) {
 			wg.Done()
 		}(name)
 	}
+
 	wg.Wait()
 }
 

retrieval/target.go

@@ -167,6 +167,7 @@ type Target struct {
 	scraperStopping chan struct{}
 	// Closing scraperStopped signals that scraping has been stopped.
 	scraperStopped chan struct{}
+	running        bool
 
 	// Mutex protects the members below.
 	sync.RWMutex
@@ -411,9 +412,11 @@ func (t *Target) InstanceIdentifier() string {
 func (t *Target) RunScraper(sampleAppender storage.SampleAppender) {
 	defer close(t.scraperStopped)
 
-	lastScrapeInterval := t.interval()
+	t.Lock()
+	t.running = true
+	t.Unlock()
 
-	log.Debugf("Starting scraper for target %v...", t)
+	lastScrapeInterval := t.interval()
 
 	select {
 	case <-time.After(t.offset(lastScrapeInterval)):
@@ -471,6 +474,14 @@ func (t *Target) RunScraper(sampleAppender storage.SampleAppender) {
 
 // StopScraper implements Target.
 func (t *Target) StopScraper() {
+	t.Lock()
+	if !t.running {
+		t.Unlock()
+		return
+	}
+	t.running = false
+	t.Unlock()
+
 	log.Debugf("Stopping scraper for target %v...", t)
 
 	close(t.scraperStopping)

retrieval/targetmanager.go

@@ -17,6 +17,7 @@ import (
 	"fmt"
 	"strings"
 	"sync"
+	"time"
 
 	"github.com/prometheus/common/log"
 	"github.com/prometheus/common/model"
@@ -48,20 +49,22 @@ type TargetProvider interface {
 // target providers.
 type TargetManager struct {
 	appender      storage.SampleAppender
+	scrapeConfigs []*config.ScrapeConfig
 
 	mtx    sync.RWMutex
 	ctx    context.Context
 	cancel func()
 	wg     sync.WaitGroup
 
-	// Providers by the scrape configs they are derived from.
-	scrapeSets []*scrapeSet
+	// Set of unqiue targets by scrape configuration.
+	targetSets map[string]*targetSet
 }
 
 // NewTargetManager creates a new TargetManager.
 func NewTargetManager(app storage.SampleAppender) *TargetManager {
 	return &TargetManager{
 		appender:   app,
+		targetSets: map[string]*targetSet{},
 	}
 }
 
@@ -72,16 +75,39 @@ func (tm *TargetManager) Run() {
 	tm.mtx.Lock()
 	tm.ctx, tm.cancel = context.WithCancel(context.Background())
 
-	for _, ss := range tm.scrapeSets {
+	jobs := map[string]struct{}{}
+
+	// Start new target sets and update existing ones.
+	for _, scfg := range tm.scrapeConfigs {
+		jobs[scfg.JobName] = struct{}{}
+
+		ts, ok := tm.targetSets[scfg.JobName]
+		if !ok {
+			ts = newTargetSet(scfg, tm.appender)
+			tm.targetSets[scfg.JobName] = ts
+		}
+		ts.runProviders(tm.ctx, providersFromConfig(scfg))
+	}
+
+	// Stop old target sets.
+	for name := range tm.targetSets {
+		if _, ok := jobs[name]; !ok {
+			delete(tm.targetSets, name)
+		}
+	}
+
+	// Run target sets.
+	for _, ts := range tm.targetSets {
 		tm.wg.Add(1)
 
-		go func(ss *scrapeSet) {
-			ss.run(tm.ctx)
+		go func(ts *targetSet) {
+			ts.run(tm.ctx)
 			tm.wg.Done()
-		}(ss)
+		}(ts)
 	}
 
 	tm.mtx.Unlock()
+
 	tm.wg.Wait()
 }
 
@@ -90,14 +116,16 @@ func (tm *TargetManager) Stop() {
 	log.Infoln("Stopping target manager...")
 
 	tm.mtx.Lock()
-
-	// Cancel the base context, this will cause all in-flight scrapes to abort immmediately.
+	// Cancel the base context, this will cause all target providers to shut down
+	// and all in-flight scrapes to abort immmediately.
 	// Started inserts will be finished before terminating.
 	tm.cancel()
 	tm.mtx.Unlock()
 
-	// Wait for all provider sets to terminate.
+	// Wait for all scrape inserts to complete.
 	tm.wg.Wait()
+
+	log.Debugln("Target manager stopped")
 }
 
 // Pools returns the targets currently being scraped bucketed by their job name.
@@ -108,8 +136,8 @@ func (tm *TargetManager) Pools() map[string][]*Target {
 	pools := map[string][]*Target{}
 
 	// TODO(fabxc): this is just a hack to maintain compatibility for now.
-	for _, ps := range tm.scrapeSets {
-		for _, ts := range ps.tgroups {
+	for _, ps := range tm.targetSets {
+		for _, ts := range ps.scrapePool.tgroups {
 			for _, t := range ts {
 				job := string(t.Labels()[model.JobLabel])
 				pools[job] = append(pools[job], t)
@@ -135,66 +163,120 @@ func (tm *TargetManager) ApplyConfig(cfg *config.Config) bool {
 	}
 
 	tm.mtx.Lock()
-
-	tm.scrapeSets = tm.scrapeSets[:0]
-
-	for _, scfg := range cfg.ScrapeConfigs {
-		tm.scrapeSets = append(tm.scrapeSets, newScrapeSet(tm.appender, scfg))
-	}
-
+	tm.scrapeConfigs = cfg.ScrapeConfigs
 	tm.mtx.Unlock()
 
 	return true
 }
 
-// scrapeSet holds several TargetProviders for which the same scrape configuration
+// targetSet holds several TargetProviders for which the same scrape configuration
 // is used. It runs the target providers and starts and stops scrapers as it
 // receives target updates.
-type scrapeSet struct {
-	appender storage.SampleAppender
-
-	config  *config.ScrapeConfig
-	tgroups map[string]map[model.Fingerprint]*Target
-
+type targetSet struct {
 	mtx       sync.RWMutex
+	tgroups   map[string]map[model.Fingerprint]*Target
+	providers map[string]TargetProvider
+
+	scrapePool *scrapePool
+	config     *config.ScrapeConfig
+
+	stopProviders func()
+	syncCh        chan struct{}
 }
 
-func newScrapeSet(app storage.SampleAppender, cfg *config.ScrapeConfig) *scrapeSet {
-	return &scrapeSet{
-		appender: app,
-		config:   cfg,
+func newTargetSet(cfg *config.ScrapeConfig, app storage.SampleAppender) *targetSet {
+	ts := &targetSet{
 		tgroups:    map[string]map[model.Fingerprint]*Target{},
+		scrapePool: newScrapePool(app),
+		syncCh:     make(chan struct{}, 1),
+		config:     cfg,
+	}
+	return ts
+}
+
+func (ts *targetSet) run(ctx context.Context) {
+	ts.scrapePool.ctx = ctx
+
+Loop:
+	for {
+		// Throttle syncing to once per five seconds.
+		select {
+		case <-ctx.Done():
+			break Loop
+		case <-time.After(5 * time.Second):
+		}
+
+		select {
+		case <-ctx.Done():
+			break Loop
+		case <-ts.syncCh:
+			ts.sync()
 		}
 	}
 
-// run starts the target providers with the given context and consumes
-// and handles their updates. If the context is done, it blocks until the
-// target scrapers have terminated.
-func (ss *scrapeSet) run(ctx context.Context) {
-	var (
-		providers = providersFromConfig(ss.config)
-		wg        sync.WaitGroup
-	)
+	// We want to wait for all pending target scrapes to complete though to ensure there'll
+	// be no more storage writes after this point.
+	ts.scrapePool.stop()
+}
+
+func (ts *targetSet) sync() {
+	// TODO(fabxc): temporary simple version. For a deduplicating scrape pool we will
+	// submit a list of all targets.
+	ts.scrapePool.sync(ts.tgroups)
+}
+
+func (ts *targetSet) runProviders(ctx context.Context, providers map[string]TargetProvider) {
+	// Lock for the entire time. This may mean up to 5 seconds until the full initial set
+	// is retrieved and applied.
+	// We could release earlier with some tweaks, but this is easier to reason about.
+	ts.mtx.Lock()
+	defer ts.mtx.Unlock()
+
+	var wg sync.WaitGroup
+
+	if ts.stopProviders != nil {
+		ts.stopProviders()
+	}
+	ctx, ts.stopProviders = context.WithCancel(ctx)
 
 	for name, prov := range providers {
-		var (
-			updates = make(chan []*config.TargetGroup)
-		)
-
 		wg.Add(1)
-		// The update and stopping operations for the target provider handling are blocking.
-		// Thus the run method only returns if all background processing is complete.
-		go func(name string, prov TargetProvider) {
-			defer wg.Done()
 
+		updates := make(chan []*config.TargetGroup)
+
+		go func(name string, prov TargetProvider) {
+			var initial []*config.TargetGroup
+
+			select {
+			case <-ctx.Done():
+				wg.Done()
+				return
+			case initial = <-updates:
+				// First set of all targets the provider knows.
+			case <-time.After(5 * time.Second):
+				// Initial set didn't arrive. Act as if it was empty
+				// and wait for updates later on.
+			}
+
+			for _, tgroup := range initial {
+				targets, err := targetsFromGroup(tgroup, ts.config)
+				if err != nil {
+					log.With("target_group", tgroup).Errorf("Target update failed: %s", err)
+					continue
+				}
+				ts.tgroups[name+"/"+tgroup.Source] = targets
+			}
+
+			wg.Done()
+
+			// Start listening for further updates.
 			for {
 				select {
 				case <-ctx.Done():
-					ss.stopScrapers(name)
 					return
 				case tgs := <-updates:
 					for _, tg := range tgs {
-						if err := ss.update(name, tg); err != nil {
+						if err := ts.update(name, tg); err != nil {
 							log.With("target_group", tg).Errorf("Target update failed: %s", err)
 						}
 					}
@@ -206,18 +288,52 @@ func (ss *scrapeSet) run(ctx context.Context) {
 	}
 
 	wg.Wait()
+
+	ts.sync()
 }
 
-// stopScrapers shuts down all active scrapers for a provider.
-func (ss *scrapeSet) stopScrapers(name string) {
+// update handles a target group update from a target provider identified by the name.
+func (ts *targetSet) update(name string, tgroup *config.TargetGroup) error {
+	targets, err := targetsFromGroup(tgroup, ts.config)
+	if err != nil {
+		return err
+	}
+
+	ts.mtx.Lock()
+	defer ts.mtx.Unlock()
+
+	ts.tgroups[name+"/"+tgroup.Source] = targets
+
+	select {
+	case ts.syncCh <- struct{}{}:
+	default:
+	}
+
+	return nil
+}
+
+// scrapePool manages scrapes for sets of targets.
+type scrapePool struct {
+	appender storage.SampleAppender
+
+	ctx     context.Context
+	mtx     sync.RWMutex
+	tgroups map[string]map[model.Fingerprint]*Target
+}
+
+func newScrapePool(app storage.SampleAppender) *scrapePool {
+	return &scrapePool{
+		appender: app,
+		tgroups:  map[string]map[model.Fingerprint]*Target{},
+	}
+}
+
+func (sp *scrapePool) stop() {
 	var wg sync.WaitGroup
 
-	ss.mtx.RLock()
-	// TODO(fabxc): the prefixing is slightly hacky but this will be gone with subsequent changes.
-	for source, tgroup := range ss.tgroups {
-		if !strings.HasPrefix(source, name) {
-			continue
-		}
+	sp.mtx.RLock()
+
+	for _, tgroup := range sp.tgroups {
 		for _, t := range tgroup {
 			wg.Add(1)
 
@@ -227,39 +343,37 @@ func (ss *scrapeSet) stopScrapers(name string) {
 			}(t)
 		}
 	}
-	ss.mtx.RUnlock()
+	sp.mtx.RUnlock()
 
 	wg.Wait()
 }
 
-// update handles a target group update from a target provider identified by the name.
-func (ss *scrapeSet) update(name string, tgroup *config.TargetGroup) error {
+func (sp *scrapePool) sync(tgroups map[string]map[model.Fingerprint]*Target) {
+	sp.mtx.Lock()
+
 	var (
-		source      = name + "/" + tgroup.Source
-		prevTargets = ss.tgroups[source]
+		wg         sync.WaitGroup
+		newTgroups = map[string]map[model.Fingerprint]*Target{}
 	)
 
-	targets, err := targetsFromGroup(tgroup, ss.config)
-	if err != nil {
-		return err
-	}
-
-	ss.mtx.Lock()
-	ss.tgroups[source] = targets
+	for source, targets := range tgroups {
+		var (
+			prevTargets = sp.tgroups[source]
+			newTargets  = map[model.Fingerprint]*Target{}
+		)
+		newTgroups[source] = newTargets
 
 		for fp, tnew := range targets {
 			// If the same target existed before, we let it run and replace
 			// the new one with it.
 			if told, ok := prevTargets[fp]; ok {
-			targets[fp] = told
+				newTargets[fp] = told
 			} else {
-			go tnew.RunScraper(ss.appender)
+				newTargets[fp] = tnew
+				go tnew.RunScraper(sp.appender)
 			}
 		}
-	ss.mtx.Unlock()
-
-	var wg sync.WaitGroup
-	for fp, told := range prevTargets {
+		for fp, told := range targets {
 			// A previous target is no longer in the group.
 			if _, ok := targets[fp]; !ok {
 				wg.Add(1)
@@ -270,13 +384,33 @@ func (ss *scrapeSet) update(name string, tgroup *config.TargetGroup) error {
 				}(told)
 			}
 		}
+	}
+
+	// Stop scrapers for target groups that disappeared completely.
+	for source, targets := range sp.tgroups {
+		if _, ok := tgroups[source]; !ok {
+			continue
+		}
+		for _, told := range targets {
+			wg.Add(1)
+
+			go func(told *Target) {
+				told.StopScraper()
+				wg.Done()
+			}(told)
+		}
+	}
+
+	sp.tgroups = newTgroups
+
 	// Wait for all potentially stopped scrapers to terminate.
 	// This covers the case of flapping targets. If the server is under high load, a new scraper
 	// may be active and tries to insert. The old scraper that didn't terminate yet could still
 	// be inserting a previous sample set.
 	wg.Wait()
 
-	return nil
+	// TODO(fabxc): maybe this can be released earlier with subsequent refactoring.
+	sp.mtx.Unlock()
 }
 
 // providersFromConfig returns all TargetProviders configured in cfg.
@@ -331,6 +465,7 @@ func providersFromConfig(cfg *config.ScrapeConfig) map[string]TargetProvider {
 // targetsFromGroup builds targets based on the given TargetGroup and config.
 func targetsFromGroup(tg *config.TargetGroup, cfg *config.ScrapeConfig) (map[model.Fingerprint]*Target, error) {
 	targets := make(map[model.Fingerprint]*Target, len(tg.Targets))
+
 	for i, labels := range tg.Targets {
 		for k, v := range cfg.Params {
 			if len(v) > 0 {