Remove the remaining races, new and old.

Also, resolve a few other TODOs.

Change-Id: Icb39b5a5e8ca22ebcb48771cd8951c5d9e112691

README.md

@@ -14,7 +14,7 @@ intervals, evaluate rule expressions, display the results, and trigger an
 action if some condition is observed to be true.
 
 TODO: The above description is somewhat esoteric. Rephrase it into
-somethith that tells normal people how they will usually benefit from
+something that tells normal people how they will usually benefit from
 using Prometheus.
 
 ## Install

retrieval/target.go

@@ -19,6 +19,7 @@ import (
 	"net/http"
 	"os"
 	"strings"
+	"sync"
 	"time"
 
 	"github.com/golang/glog"
@@ -136,14 +137,9 @@ type Target interface {
 	RunScraper(extraction.Ingester, time.Duration)
 	// Stop scraping, synchronous.
 	StopScraper()
-	// Do a single scrape.
-	scrape(ingester extraction.Ingester) error
 }
 
 // target is a Target that refers to a singular HTTP or HTTPS endpoint.
-//
-// TODO: The implementation is not yet goroutine safe, but for the web status,
-// methods are called concurrently.
 type target struct {
 	// The current health state of the target.
 	state TargetState
@@ -151,8 +147,10 @@ type target struct {
 	lastError error
 	// The last time a scrape was attempted.
 	lastScrape time.Time
-	// Closing stopScraper signals that scraping should stop.
+	// Closing scraperStopping signals that scraping should stop.
-	stopScraper chan struct{}
+	scraperStopping chan struct{}
+	// Closing scraperStopped signals that scraping has been stopped.
+	scraperStopped chan struct{}
 	// Channel to queue base labels to be replaced.
 	newBaseLabels chan clientmodel.LabelSet
 
@@ -163,17 +161,25 @@ type target struct {
 	baseLabels clientmodel.LabelSet
 	// The HTTP client used to scrape the target's endpoint.
 	httpClient *http.Client
+
+	// Mutex protects lastError, lastScrape, state, and baseLabels.  Writing
+	// the above must only happen in the goroutine running the RunScraper
+	// loop, and it must happen under the lock. In that way, no mutex lock
+	// is required for reading the above in the goroutine running the
+	// RunScraper loop, but only for reading in other goroutines.
+	sync.Mutex
 }
 
 // Furnish a reasonably configured target for querying.
 func NewTarget(address string, deadline time.Duration, baseLabels clientmodel.LabelSet) Target {
 	target := &target{
-		address:       address,
+		address:         address,
-		Deadline:      deadline,
+		Deadline:        deadline,
-		baseLabels:    baseLabels,
+		baseLabels:      baseLabels,
-		httpClient:    utility.NewDeadlineClient(deadline),
+		httpClient:      utility.NewDeadlineClient(deadline),
-		stopScraper:   make(chan struct{}),
+		scraperStopping: make(chan struct{}),
-		newBaseLabels: make(chan clientmodel.LabelSet, 1),
+		scraperStopped:  make(chan struct{}),
+		newBaseLabels:   make(chan clientmodel.LabelSet, 1),
 	}
 
 	return target
@@ -213,6 +219,7 @@ func (t *target) RunScraper(ingester extraction.Ingester, interval time.Duration
 			case <-t.newBaseLabels:
 				// Do nothing.
 			default:
+				close(t.scraperStopped)
 				return
 			}
 		}
@@ -221,7 +228,7 @@ func (t *target) RunScraper(ingester extraction.Ingester, interval time.Duration
 	jitterTimer := time.NewTimer(time.Duration(float64(interval) * rand.Float64()))
 	select {
 	case <-jitterTimer.C:
-	case <-t.stopScraper:
+	case <-t.scraperStopping:
 		jitterTimer.Stop()
 		return
 	}
@@ -230,32 +237,40 @@ func (t *target) RunScraper(ingester extraction.Ingester, interval time.Duration
 	ticker := time.NewTicker(interval)
 	defer ticker.Stop()
 
+	t.Lock() // Writing t.lastScrape requires the lock.
 	t.lastScrape = time.Now()
+	t.Unlock()
 	t.scrape(ingester)
 
 	// Explanation of the contraption below:
 	//
-	// In case t.newBaseLabels or t.stopScraper have something to receive,
+	// In case t.newBaseLabels or t.scraperStopping have something to receive,
 	// we want to read from those channels rather than starting a new scrape
 	// (which might take very long). That's why the outer select has no
-	// ticker.C. Should neither t.newBaseLabels nor t.stopScraper have
+	// ticker.C. Should neither t.newBaseLabels nor t.scraperStopping have
 	// anything to receive, we go into the inner select, where ticker.C is
 	// in the mix.
 	for {
 		select {
 		case newBaseLabels := <-t.newBaseLabels:
+			t.Lock() // Writing t.baseLabels requires the lock.
 			t.baseLabels = newBaseLabels
-		case <-t.stopScraper:
+			t.Unlock()
+		case <-t.scraperStopping:
 			return
 		default:
 			select {
 			case newBaseLabels := <-t.newBaseLabels:
+				t.Lock() // Writing t.baseLabels requires the lock.
 				t.baseLabels = newBaseLabels
-			case <-t.stopScraper:
+				t.Unlock()
+			case <-t.scraperStopping:
 				return
 			case <-ticker.C:
 				targetIntervalLength.WithLabelValues(interval.String()).Observe(float64(time.Since(t.lastScrape) / time.Second))
+				t.Lock() // Write t.lastScrape requires locking.
 				t.lastScrape = time.Now()
+				t.Unlock()
 				t.scrape(ingester)
 			}
 		}
@@ -264,7 +279,8 @@ func (t *target) RunScraper(ingester extraction.Ingester, interval time.Duration
 
 // StopScraper implements Target.
 func (t *target) StopScraper() {
-	close(t.stopScraper)
+	close(t.scraperStopping)
+	<-t.scraperStopped
 }
 
 const acceptHeader = `application/vnd.google.protobuf;proto=io.prometheus.client.MetricFamily;encoding=delimited;q=0.7,text/plain;version=0.0.4;q=0.3,application/json;schema="prometheus/telemetry";version=0.0.2;q=0.2,*/*;q=0.1`
@@ -278,6 +294,7 @@ func (t *target) scrape(ingester extraction.Ingester) (err error) {
 			instance: t.Address(),
 			outcome:  success,
 		}
+		t.Lock() // Writing t.state and t.lastError requires the lock.
 		if err == nil {
 			t.state = ALIVE
 			t.recordScrapeHealth(ingester, timestamp, true)
@@ -288,6 +305,7 @@ func (t *target) scrape(ingester extraction.Ingester) (err error) {
 		}
 		targetOperationLatencies.With(labels).Observe(ms)
 		t.lastError = err
+		t.Unlock()
 	}(time.Now())
 
 	req, err := http.NewRequest("GET", t.Address(), nil)
@@ -310,8 +328,6 @@ func (t *target) scrape(ingester extraction.Ingester) (err error) {
 		return err
 	}
 
-	// TODO: This is a wart; we need to handle this more gracefully down the
-	// road, especially once we have service discovery support.
 	baseLabels := clientmodel.LabelSet{InstanceLabel: clientmodel.LabelValue(t.Address())}
 	for baseLabel, baseValue := range t.baseLabels {
 		baseLabels[baseLabel] = baseValue
@@ -331,16 +347,22 @@ func (t *target) scrape(ingester extraction.Ingester) (err error) {
 
 // LastError implements Target.
 func (t *target) LastError() error {
+	t.Lock()
+	defer t.Unlock()
 	return t.lastError
 }
 
 // State implements Target.
 func (t *target) State() TargetState {
+	t.Lock()
+	defer t.Unlock()
 	return t.state
 }
 
 // LastScrape implements Target.
 func (t *target) LastScrape() time.Time {
+	t.Lock()
+	defer t.Unlock()
 	return t.lastScrape
 }
 
@@ -365,6 +387,8 @@ func (t *target) GlobalAddress() string {
 
 // BaseLabels implements Target.
 func (t *target) BaseLabels() clientmodel.LabelSet {
+	t.Lock()
+	defer t.Unlock()
 	return t.baseLabels
 }
 
@@ -375,5 +399,3 @@ func (t *target) SetBaseLabelsFrom(newTarget Target) {
 	}
 	t.newBaseLabels <- newTarget.BaseLabels()
 }
-
-type targets []Target

retrieval/target_test.go

@@ -100,7 +100,7 @@ func TestTargetScrapeTimeout(t *testing.T) {
 
 	// scrape once without timeout
 	signal <- true
-	if err := testTarget.scrape(ingester); err != nil {
+	if err := testTarget.(*target).scrape(ingester); err != nil {
 		t.Fatal(err)
 	}
 
@@ -109,12 +109,12 @@ func TestTargetScrapeTimeout(t *testing.T) {
 
 	// now scrape again
 	signal <- true
-	if err := testTarget.scrape(ingester); err != nil {
+	if err := testTarget.(*target).scrape(ingester); err != nil {
 		t.Fatal(err)
 	}
 
 	// now timeout
-	if err := testTarget.scrape(ingester); err == nil {
+	if err := testTarget.(*target).scrape(ingester); err == nil {
 		t.Fatal("expected scrape to timeout")
 	} else {
 		signal <- true // let handler continue
@@ -122,7 +122,7 @@ func TestTargetScrapeTimeout(t *testing.T) {
 
 	// now scrape again without timeout
 	signal <- true
-	if err := testTarget.scrape(ingester); err != nil {
+	if err := testTarget.(*target).scrape(ingester); err != nil {
 		t.Fatal(err)
 	}
 }
@@ -138,7 +138,7 @@ func TestTargetScrape404(t *testing.T) {
 	ingester := nopIngester{}
 
 	want := errors.New("server returned HTTP status 404 Not Found")
-	got := testTarget.scrape(ingester)
+	got := testTarget.(*target).scrape(ingester)
 	if got == nil || want.Error() != got.Error() {
 		t.Fatalf("want err %q, got %q", want, got)
 	}
@@ -146,10 +146,11 @@ func TestTargetScrape404(t *testing.T) {
 
 func TestTargetRunScraperScrapes(t *testing.T) {
 	testTarget := target{
-		state:       UNKNOWN,
+		state:           UNKNOWN,
-		address:     "bad schema",
+		address:         "bad schema",
-		httpClient:  utility.NewDeadlineClient(0),
+		httpClient:      utility.NewDeadlineClient(0),
-		stopScraper: make(chan struct{}),
+		scraperStopping: make(chan struct{}),
+		scraperStopped:  make(chan struct{}),
 	}
 	go testTarget.RunScraper(nopIngester{}, time.Duration(time.Millisecond))
 

retrieval/targetmanager.go

@@ -23,16 +23,18 @@ import (
 	"github.com/prometheus/prometheus/config"
 )
 
+// TargetManager manages all scrape targets. All methods are goroutine-safe.
 type TargetManager interface {
 	AddTarget(job config.JobConfig, t Target)
 	ReplaceTargets(job config.JobConfig, newTargets []Target)
 	Remove(t Target)
 	AddTargetsFromConfig(config config.Config)
 	Stop()
-	Pools() map[string]*TargetPool
+	Pools() map[string]*TargetPool // Returns a copy of the name -> TargetPool mapping.
 }
 
 type targetManager struct {
+	sync.Mutex // Protects poolByJob.
 	poolsByJob map[string]*TargetPool
 	ingester   extraction.Ingester
 }
@@ -44,7 +46,7 @@ func NewTargetManager(ingester extraction.Ingester) TargetManager {
 	}
 }
 
-func (m *targetManager) TargetPoolForJob(job config.JobConfig) *TargetPool {
+func (m *targetManager) targetPoolForJob(job config.JobConfig) *TargetPool {
 	targetPool, ok := m.poolsByJob[job.GetName()]
 
 	if !ok {
@@ -58,7 +60,6 @@ func (m *targetManager) TargetPoolForJob(job config.JobConfig) *TargetPool {
 		glog.Infof("Pool for job %s does not exist; creating and starting...", job.GetName())
 
 		m.poolsByJob[job.GetName()] = targetPool
-		// TODO: Investigate whether this auto-goroutine creation is desired.
 		go targetPool.Run()
 	}
 
@@ -66,24 +67,32 @@ func (m *targetManager) TargetPoolForJob(job config.JobConfig) *TargetPool {
 }
 
 func (m *targetManager) AddTarget(job config.JobConfig, t Target) {
-	targetPool := m.TargetPoolForJob(job)
+	m.Lock()
+	defer m.Unlock()
+
+	targetPool := m.targetPoolForJob(job)
 	targetPool.AddTarget(t)
 	m.poolsByJob[job.GetName()] = targetPool
 }
 
 func (m *targetManager) ReplaceTargets(job config.JobConfig, newTargets []Target) {
-	targetPool := m.TargetPoolForJob(job)
+	m.Lock()
+	defer m.Unlock()
+
+	targetPool := m.targetPoolForJob(job)
 	targetPool.ReplaceTargets(newTargets)
 }
 
-func (m targetManager) Remove(t Target) {
+func (m *targetManager) Remove(t Target) {
 	panic("not implemented")
 }
 
 func (m *targetManager) AddTargetsFromConfig(config config.Config) {
 	for _, job := range config.Jobs() {
 		if job.SdName != nil {
-			m.TargetPoolForJob(job)
+			m.Lock()
+			m.targetPoolForJob(job)
+			m.Unlock()
 			continue
 		}
 
@@ -106,6 +115,9 @@ func (m *targetManager) AddTargetsFromConfig(config config.Config) {
 }
 
 func (m *targetManager) Stop() {
+	m.Lock()
+	defer m.Unlock()
+
 	glog.Info("Stopping target manager...")
 	var wg sync.WaitGroup
 	for j, p := range m.poolsByJob {
@@ -121,7 +133,13 @@ func (m *targetManager) Stop() {
 	glog.Info("Target manager stopped.")
 }
 
-// TODO: Not goroutine-safe. Only used in /status page for now.
 func (m *targetManager) Pools() map[string]*TargetPool {
-	return m.poolsByJob
+	m.Lock()
+	defer m.Unlock()
+
+	result := make(map[string]*TargetPool, len(m.poolsByJob))
+	for k, v := range m.poolsByJob {
+		result[k] = v
+	}
+	return result
 }

retrieval/targetpool.go

@@ -122,9 +122,9 @@ func (p *TargetPool) ReplaceTargets(newTargets []Target) {
 				defer wg.Done()
 				glog.V(1).Infof("Stopping scraper for target %s...", k)
 				oldTarget.StopScraper()
-				delete(p.targetsByAddress, k)
 				glog.V(1).Infof("Scraper for target %s stopped.", k)
 			}(k, oldTarget)
+			delete(p.targetsByAddress, k)
 		}
 	}
 	wg.Wait()

retrieval/targetpool_test.go

@@ -114,32 +114,36 @@ func TestTargetPool(t *testing.T) {
 func TestTargetPoolReplaceTargets(t *testing.T) {
 	pool := NewTargetPool(nil, nil, nopIngester{}, time.Duration(1))
 	oldTarget1 := &target{
-		address:       "example1",
+		address:         "example1",
-		state:         UNREACHABLE,
+		state:           UNREACHABLE,
-		stopScraper:   make(chan struct{}),
+		scraperStopping: make(chan struct{}),
-		newBaseLabels: make(chan clientmodel.LabelSet, 1),
+		scraperStopped:  make(chan struct{}),
-		httpClient:    &http.Client{},
+		newBaseLabels:   make(chan clientmodel.LabelSet, 1),
+		httpClient:      &http.Client{},
 	}
 	oldTarget2 := &target{
-		address:       "example2",
+		address:         "example2",
-		state:         UNREACHABLE,
+		state:           UNREACHABLE,
-		stopScraper:   make(chan struct{}),
+		scraperStopping: make(chan struct{}),
-		newBaseLabels: make(chan clientmodel.LabelSet, 1),
+		scraperStopped:  make(chan struct{}),
-		httpClient:    &http.Client{},
+		newBaseLabels:   make(chan clientmodel.LabelSet, 1),
+		httpClient:      &http.Client{},
 	}
 	newTarget1 := &target{
-		address:       "example1",
+		address:         "example1",
-		state:         ALIVE,
+		state:           ALIVE,
-		stopScraper:   make(chan struct{}),
+		scraperStopping: make(chan struct{}),
-		newBaseLabels: make(chan clientmodel.LabelSet, 1),
+		scraperStopped:  make(chan struct{}),
-		httpClient:    &http.Client{},
+		newBaseLabels:   make(chan clientmodel.LabelSet, 1),
+		httpClient:      &http.Client{},
 	}
 	newTarget2 := &target{
-		address:       "example3",
+		address:         "example3",
-		state:         ALIVE,
+		state:           ALIVE,
-		stopScraper:   make(chan struct{}),
+		scraperStopping: make(chan struct{}),
-		newBaseLabels: make(chan clientmodel.LabelSet, 1),
+		scraperStopped:  make(chan struct{}),
-		httpClient:    &http.Client{},
+		newBaseLabels:   make(chan clientmodel.LabelSet, 1),
+		httpClient:      &http.Client{},
 	}
 
 	pool.addTarget(oldTarget1)

rules/ast/functions.go

@@ -255,8 +255,7 @@ func dropCommonLabelsImpl(timestamp clientmodel.Timestamp, args []Node) interfac
 	}
 	common := clientmodel.LabelSet{}
 	for k, v := range vector[0].Metric {
-		// TODO(julius): Revisit this when https://github.com/prometheus/prometheus/issues/380
+		// TODO(julius): Should we also drop common metric names?
-		// is implemented.
 		if k == clientmodel.MetricNameLabel {
 			continue
 		}

storage/local/chunk.go

@@ -125,6 +125,13 @@ func (cd *chunkDesc) contains(t clientmodel.Timestamp) bool {
 	return !t.Before(cd.firstTime()) && !t.After(cd.lastTime())
 }
 
+func (cd *chunkDesc) getChunk() chunk {
+	cd.Lock()
+	defer cd.Unlock()
+
+	return cd.chunk
+}
+
 func (cd *chunkDesc) setChunk(c chunk) {
 	cd.Lock()
 	defer cd.Unlock()

storage/local/series.go

@@ -381,11 +381,11 @@ func (s *memorySeries) preloadChunksForRange(
 func (s *memorySeries) newIterator(lockFunc, unlockFunc func()) SeriesIterator {
 	chunks := make([]chunk, 0, len(s.chunkDescs))
 	for i, cd := range s.chunkDescs {
-		if !cd.isEvicted() {
+		if chunk := cd.getChunk(); chunk != nil {
 			if i == len(s.chunkDescs)-1 && !s.headChunkPersisted {
 				s.headChunkUsedByIterator = true
 			}
-			chunks = append(chunks, cd.chunk)
+			chunks = append(chunks, chunk)
 		}
 	}