Evict based on memory pressure. Evict recently used chunks last.

Change-Id: Ie6168f0cdb3917bdc63b6fe15585dd70c1e42afe

main.go

@@ -43,17 +43,17 @@ const deletionBatchSize = 100
 // Commandline flags.
 var (
 	configFile = flag.String("configFile", "prometheus.conf", "Prometheus configuration file name.")
-	metricsStoragePath = flag.String("metricsStoragePath", "/tmp/metrics", "Base path for metrics storage.")
 
 	alertmanagerURL = flag.String("alertmanager.url", "", "The URL of the alert manager to send notifications to.")
 
+	metricsStoragePath = flag.String("storage.path", "/tmp/metrics", "Base path for metrics storage.")
+
 	remoteTSDBUrl     = flag.String("storage.remote.url", "", "The URL of the OpenTSDB instance to send samples to.")
 	remoteTSDBTimeout = flag.Duration("storage.remote.timeout", 30*time.Second, "The timeout to use when sending samples to OpenTSDB.")
 
 	samplesQueueCapacity = flag.Int("storage.queue.samplesCapacity", 4096, "The size of the unwritten samples queue.")
 
-	memoryEvictionInterval = flag.Duration("storage.memory.evictionInterval", 15*time.Minute, "The period at which old data is evicted from memory.")
-	memoryRetentionPeriod  = flag.Duration("storage.memory.retentionPeriod", time.Hour, "The period of time to retain in memory during evictions.")
+	numMemoryChunks = flag.Int("storage.memoryChunks", 1024*1024, "How many chunks to keep in memory. While the size of a chunk is 1kiB, the total memory usage will be significantly higher than this value * 1kiB. Furthermore, for various reasons, more chunks might have to be kept in memory temporarily.")
 
 	storageRetentionPeriod = flag.Duration("storage.retentionPeriod", 15*24*time.Hour, "The period of time to retain in storage.")
 
@@ -115,8 +115,7 @@ func NewPrometheus() *prometheus {
 	notificationHandler := notification.NewNotificationHandler(*alertmanagerURL, *notificationQueueCapacity)
 
 	o := &local.MemorySeriesStorageOptions{
-		MemoryEvictionInterval:     *memoryEvictionInterval,
-		MemoryRetentionPeriod:      *memoryRetentionPeriod,
+		MemoryChunks:               *numMemoryChunks,
 		PersistenceStoragePath:     *metricsStoragePath,
 		PersistenceRetentionPeriod: *storageRetentionPeriod,
 		CheckpointInterval:         *checkpointInterval,

storage/local/chunk.go

@@ -14,6 +14,7 @@
 package local
 
 import (
+	"container/list"
 	"io"
 	"sync"
 	"sync/atomic"
@@ -27,9 +28,13 @@ type chunkDesc struct {
 	sync.Mutex
 	chunk          chunk
 	refCount       int
-	evict          bool
 	chunkFirstTime clientmodel.Timestamp // Used if chunk is evicted.
 	chunkLastTime  clientmodel.Timestamp // Used if chunk is evicted.
+
+	// evictListElement is nil if the chunk is not in the evict list.
+	// evictListElement is _not_ protected by the chunkDesc mutex.
+	// It must only be touched by the evict list handler in memorySeriesStorage.
+	evictListElement *list.Element
 }
 
 // newChunkDesc creates a new chunkDesc pointing to the given chunk. The
@@ -48,14 +53,18 @@ func (cd *chunkDesc) add(s *metric.SamplePair) []chunk {
 	return cd.chunk.add(s)
 }
 
-func (cd *chunkDesc) pin() {
+func (cd *chunkDesc) pin(evictRequests chan<- evictRequest) {
 	cd.Lock()
 	defer cd.Unlock()
 
+	if cd.refCount == 0 {
+		// Remove ourselves from the evict list.
+		evictRequests <- evictRequest{cd, false}
+	}
 	cd.refCount++
 }
 
-func (cd *chunkDesc) unpin() {
+func (cd *chunkDesc) unpin(evictRequests chan<- evictRequest) {
 	cd.Lock()
 	defer cd.Unlock()
 
@@ -63,8 +72,9 @@ func (cd *chunkDesc) unpin() {
 		panic("cannot unpin already unpinned chunk")
 	}
 	cd.refCount--
-	if cd.refCount == 0 && cd.evict {
-		cd.evictNow()
+	if cd.refCount == 0 {
+		// Add ourselves to the back of the evict list.
+		evictRequests <- evictRequest{cd, true}
 	}
 }
 
@@ -115,36 +125,28 @@ func (cd *chunkDesc) setChunk(c chunk) {
 	if cd.chunk != nil {
 		panic("chunk already set")
 	}
-	cd.evict = false
 	cd.chunk = c
 }
 
-// evictOnUnpin evicts the chunk once unpinned. If it is not pinned when this
-// method is called, it evicts the chunk immediately and returns true. If the
-// chunk is already evicted when this method is called, it returns true, too.
-func (cd *chunkDesc) evictOnUnpin() bool {
+// maybeEvict evicts the chunk if the refCount is 0. It returns wether the chunk
+// is now evicted, which includes the case that the chunk was evicted even
+// before this method was called.
+func (cd *chunkDesc) maybeEvict() bool {
 	cd.Lock()
 	defer cd.Unlock()
 
 	if cd.chunk == nil {
-		// Already evicted.
-		return true
-	}
-	cd.evict = true
-	if cd.refCount == 0 {
-		cd.evictNow()
 		return true
 	}
+	if cd.refCount != 0 {
 		return false
-}
-
-// evictNow is an internal helper method.
-func (cd *chunkDesc) evictNow() {
+	}
 	cd.chunkFirstTime = cd.chunk.firstTime()
 	cd.chunkLastTime = cd.chunk.lastTime()
 	cd.chunk = nil
 	chunkOps.WithLabelValues(evict).Inc()
 	atomic.AddInt64(&numMemChunks, -1)
+	return true
 }
 
 // chunk is the interface for all chunks. Chunks are generally not

storage/local/preload.go

@@ -104,7 +104,7 @@ func (p *memorySeriesPreloader) Close() {
 	// over the pinnedChunkDescs to a manager that will delay the unpinning
 	// based on time and memory pressure.
 	for _, cd := range p.pinnedChunkDescs {
-		cd.unpin()
+		cd.unpin(p.storage.evictRequests)
 	}
 	chunkOps.WithLabelValues(unpin).Add(float64(len(p.pinnedChunkDescs)))
 

storage/local/series.go

@@ -225,42 +225,10 @@ func (s *memorySeries) add(fp clientmodel.Fingerprint, v *metric.SamplePair) []*
 	return chunkDescsToPersist
 }
 
-// evictOlderThan marks for eviction all chunks whose latest sample is older
-// than the given timestamp. It returns allEvicted as true if all chunks in the
-// series were immediately evicted (i.e. all chunks are older than the
-// timestamp, and none of the chunks was pinned).
-//
-// The method also evicts chunkDescs if there are chunkDescEvictionFactor times
-// more chunkDescs in the series than unevicted chunks. (The number of unevicted
-// chunks is considered the number of chunks between (and including) the oldest
-// chunk that could not be evicted immediately and the newest chunk in the
-// series, even if chunks in between were evicted.)
-//
-// Special considerations for the head chunk: If it has not been scheduled to be
-// persisted yet but is old enough for eviction, this method returns a pointer
-// to the descriptor of the head chunk to be persisted. (Otherwise, the method
-// returns nil.) The caller is then responsible for persisting the head
-// chunk. The internal state of this memorySeries is already set accordingly by
-// this method. Calling evictOlderThan for a series with a non-persisted head
-// chunk that is old enough for eviction will never evict all chunks
-// immediately, even if no chunk is pinned for other reasons, because the head
-// chunk is not persisted yet. A series old enough for archiving will require at
-// least two eviction runs to become ready for archiving: In the first run, its
-// head chunk is requested to be persisted. The next call of evictOlderThan will
-// then return true, provided that the series hasn't received new samples in the
-// meantime, the head chunk has now been persisted, and no chunk is pinned for
-// other reasons.
-//
-// The caller must have locked the fingerprint of the series.
-func (s *memorySeries) evictOlderThan(t clientmodel.Timestamp) (allEvicted bool, headChunkToPersist *chunkDesc) {
-	allEvicted = true
-	iOldestNotEvicted := -1
-
-	defer func() {
-		// Evict chunkDescs if there are chunkDescEvictionFactor times
-		// more than non-evicted chunks and we are not going to archive
-		// the whole series anyway.
-		if iOldestNotEvicted != -1 {
+// evictChunkDescs evicts chunkDescs if there are chunkDescEvictionFactor times
+// more than non-evicted chunks. iOldestNotEvicted is the index within the
+// current chunkDescs of the oldest chunk that is not evicted.
+func (s *memorySeries) evictChunkDescs(iOldestNotEvicted int) {
 	lenToKeep := chunkDescEvictionFactor * (len(s.chunkDescs) - iOldestNotEvicted)
 	if lenToKeep < len(s.chunkDescs) {
 		s.savedFirstTime = s.firstTime()
@@ -273,42 +241,10 @@ func (s *memorySeries) evictOlderThan(t clientmodel.Timestamp) (allEvicted bool,
 			s.chunkDescs[lenEvicted:]...,
 		)
 	}
-		}
-	}()
-
-	// For now, always drop the entire range from oldest to t.
-	for i, cd := range s.chunkDescs {
-		if !cd.lastTime().Before(t) {
-			if iOldestNotEvicted == -1 {
-				iOldestNotEvicted = i
-			}
-			return false, nil
-		}
-		if cd.isEvicted() {
-			continue
-		}
-		if !s.headChunkPersisted && i == len(s.chunkDescs)-1 {
-			// This is a non-persisted head chunk that is old enough
-			// for eviction. Request it to be persisted:
-			headChunkToPersist = cd
-			s.headChunkPersisted = true
-			// Since we cannot modify the head chunk from now on, we
-			// don't need to bother with cloning anymore.
-			s.headChunkUsedByIterator = false
-		}
-		if !cd.evictOnUnpin() {
-			if iOldestNotEvicted == -1 {
-				iOldestNotEvicted = i
-			}
-			allEvicted = false
-		}
-	}
-	return allEvicted, headChunkToPersist
 }
 
-// purgeOlderThan removes chunkDescs older than t. It also evicts the chunks of
-// those chunkDescs (although that's probably not even necessary). It returns
-// the number of purged chunkDescs and true if all chunkDescs have been purged.
+// purgeOlderThan removes chunkDescs older than t. It returns the number of
+// purged chunkDescs and true if all chunkDescs have been purged.
 //
 // The caller must have locked the fingerprint of the series.
 func (s *memorySeries) purgeOlderThan(t clientmodel.Timestamp) (int, bool) {
@@ -318,7 +254,6 @@ func (s *memorySeries) purgeOlderThan(t clientmodel.Timestamp) (int, bool) {
 			keepIdx = i
 			break
 		}
-		s.chunkDescs[i].evictOnUnpin()
 	}
 	if keepIdx > 0 {
 		s.chunkDescs = append(make([]*chunkDesc, 0, len(s.chunkDescs)-keepIdx), s.chunkDescs[keepIdx:]...)
@@ -328,13 +263,13 @@ func (s *memorySeries) purgeOlderThan(t clientmodel.Timestamp) (int, bool) {
 }
 
 // preloadChunks is an internal helper method.
-func (s *memorySeries) preloadChunks(indexes []int, p *persistence) ([]*chunkDesc, error) {
+func (s *memorySeries) preloadChunks(indexes []int, mss *memorySeriesStorage) ([]*chunkDesc, error) {
 	loadIndexes := []int{}
 	pinnedChunkDescs := make([]*chunkDesc, 0, len(indexes))
 	for _, idx := range indexes {
 		cd := s.chunkDescs[idx]
 		pinnedChunkDescs = append(pinnedChunkDescs, cd)
-		cd.pin() // Have to pin everything first to prevent concurrent evictOnUnpin later.
+		cd.pin(mss.evictRequests) // Have to pin everything first to prevent immediate eviction on chunk loading.
 		if cd.isEvicted() {
 			loadIndexes = append(loadIndexes, idx)
 		}
@@ -346,11 +281,12 @@ func (s *memorySeries) preloadChunks(indexes []int, p *persistence) ([]*chunkDes
 			panic("requested loading chunks from persistence in a situation where we must not have persisted data for chunk descriptors in memory")
 		}
 		fp := s.metric.Fingerprint()
-		chunks, err := p.loadChunks(fp, loadIndexes, s.chunkDescsOffset)
+		// TODO: Remove law-of-Demeter violation?
+		chunks, err := mss.persistence.loadChunks(fp, loadIndexes, s.chunkDescsOffset)
 		if err != nil {
 			// Unpin the chunks since we won't return them as pinned chunks now.
 			for _, cd := range pinnedChunkDescs {
-				cd.unpin()
+				cd.unpin(mss.evictRequests)
 			}
 			chunkOps.WithLabelValues(unpin).Add(float64(len(pinnedChunkDescs)))
 			return nil, err
@@ -400,14 +336,15 @@ func (s *memorySeries) preloadChunksAtTime(t clientmodel.Timestamp, p *persisten
 // The caller must have locked the fingerprint of the series.
 func (s *memorySeries) preloadChunksForRange(
 	from clientmodel.Timestamp, through clientmodel.Timestamp,
-	fp clientmodel.Fingerprint, p *persistence,
+	fp clientmodel.Fingerprint, mss *memorySeriesStorage,
 ) ([]*chunkDesc, error) {
 	firstChunkDescTime := clientmodel.Timestamp(math.MaxInt64)
 	if len(s.chunkDescs) > 0 {
 		firstChunkDescTime = s.chunkDescs[0].firstTime()
 	}
 	if s.chunkDescsOffset != 0 && from.Before(firstChunkDescTime) {
-		cds, err := p.loadChunkDescs(fp, firstChunkDescTime)
+		// TODO: Remove law-of-demeter violation?
+		cds, err := mss.persistence.loadChunkDescs(fp, firstChunkDescTime)
 		if err != nil {
 			return nil, err
 		}
@@ -438,7 +375,7 @@ func (s *memorySeries) preloadChunksForRange(
 	for i := fromIdx; i <= throughIdx; i++ {
 		pinIndexes = append(pinIndexes, i)
 	}
-	return s.preloadChunks(pinIndexes, p)
+	return s.preloadChunks(pinIndexes, mss)
 }
 
 // newIterator returns a new SeriesIterator. The caller must have locked the
@@ -467,18 +404,6 @@ func (s *memorySeries) head() *chunkDesc {
 	return s.chunkDescs[len(s.chunkDescs)-1]
 }
 
-// values returns all values in the series. The caller must have locked the
-// fingerprint of the memorySeries.
-func (s *memorySeries) values() metric.Values {
-	var values metric.Values
-	for _, cd := range s.chunkDescs {
-		for sample := range cd.chunk.values() {
-			values = append(values, *sample)
-		}
-	}
-	return values
-}
-
 // firstTime returns the timestamp of the first sample in the series. The caller
 // must have locked the fingerprint of the memorySeries.
 func (s *memorySeries) firstTime() clientmodel.Timestamp {

storage/local/storage.go

@@ -15,6 +15,7 @@
 package local
 
 import (
+	"container/list"
 	"sync/atomic"
 	"time"
 
@@ -28,12 +29,16 @@ import (
 
 const (
 	persistQueueCap  = 1024
+	evictRequestsCap = 1024
 	chunkLen         = 1024
 
 	// See waitForNextFP.
 	fpMaxWaitDuration = 10 * time.Second
 	fpMinWaitDuration = 5 * time.Millisecond // ~ hard disk seek time.
 	fpMaxSweepTime    = 6 * time.Hour
+
+	maxEvictInterval = time.Minute
+	headChunkTimeout = time.Hour // Close head chunk if not touched for that long.
 )
 
 type storageState uint
@@ -49,12 +54,17 @@ type persistRequest struct {
 	chunkDesc   *chunkDesc
 }
 
+type evictRequest struct {
+	cd    *chunkDesc
+	evict bool
+}
+
 type memorySeriesStorage struct {
 	fpLocker   *fingerprintLocker
 	fpToSeries *seriesMap
 
 	loopStopping, loopStopped chan struct{}
-	evictInterval, evictAfter time.Duration
+	maxMemoryChunks           int
 	purgeAfter                time.Duration
 	checkpointInterval        time.Duration
 
@@ -62,6 +72,10 @@ type memorySeriesStorage struct {
 	persistStopped chan struct{}
 	persistence    *persistence
 
+	evictList                   *list.List
+	evictRequests               chan evictRequest
+	evictStopping, evictStopped chan struct{}
+
 	persistLatency              prometheus.Summary
 	persistErrors               *prometheus.CounterVec
 	persistQueueLength          prometheus.Gauge
@@ -69,15 +83,14 @@ type memorySeriesStorage struct {
 	seriesOps                   *prometheus.CounterVec
 	ingestedSamplesCount        prometheus.Counter
 	invalidPreloadRequestsCount prometheus.Counter
-	purgeDuration, evictDuration prometheus.Gauge
+	purgeDuration               prometheus.Gauge
 }
 
 // MemorySeriesStorageOptions contains options needed by
 // NewMemorySeriesStorage. It is not safe to leave any of those at their zero
 // values.
 type MemorySeriesStorageOptions struct {
-	MemoryEvictionInterval     time.Duration // How often to check for memory eviction.
-	MemoryRetentionPeriod      time.Duration // Chunks at least that old are evicted from memory.
+	MemoryChunks               int           // How many chunks to keep in memory.
 	PersistenceStoragePath     string        // Location of persistence files.
 	PersistenceRetentionPeriod time.Duration // Chunks at least that old are purged.
 	CheckpointInterval         time.Duration // How often to checkpoint the series map and head chunks.
@@ -111,8 +124,7 @@ func NewMemorySeriesStorage(o *MemorySeriesStorageOptions) (Storage, error) {
 
 		loopStopping:       make(chan struct{}),
 		loopStopped:        make(chan struct{}),
-		evictInterval:      o.MemoryEvictionInterval,
-		evictAfter:         o.MemoryRetentionPeriod,
+		maxMemoryChunks:    o.MemoryChunks,
 		purgeAfter:         o.PersistenceRetentionPeriod,
 		checkpointInterval: o.CheckpointInterval,
 
@@ -120,6 +132,11 @@ func NewMemorySeriesStorage(o *MemorySeriesStorageOptions) (Storage, error) {
 		persistStopped: make(chan struct{}),
 		persistence:    p,
 
+		evictList:     list.New(),
+		evictRequests: make(chan evictRequest, evictRequestsCap),
+		evictStopping: make(chan struct{}),
+		evictStopped:  make(chan struct{}),
+
 		persistLatency: prometheus.NewSummary(prometheus.SummaryOpts{
 			Namespace: namespace,
 			Subsystem: subsystem,
@@ -163,23 +180,12 @@ func NewMemorySeriesStorage(o *MemorySeriesStorageOptions) (Storage, error) {
 			Name:      "invalid_preload_requests_total",
 			Help:      "The total number of preload requests referring to a non-existent series. This is an indication of outdated label indexes.",
 		}),
-		purgeDuration: prometheus.NewGauge(prometheus.GaugeOpts{
-			Namespace: namespace,
-			Subsystem: subsystem,
-			Name:      "purge_duration_milliseconds",
-			Help:      "The duration of the last storage purge iteration in milliseconds.",
-		}),
-		evictDuration: prometheus.NewGauge(prometheus.GaugeOpts{
-			Namespace: namespace,
-			Subsystem: subsystem,
-			Name:      "evict_duration_milliseconds",
-			Help:      "The duration of the last memory eviction iteration in milliseconds.",
-		}),
 	}, nil
 }
 
 // Start implements Storage.
 func (s *memorySeriesStorage) Start() {
+	go s.handleEvictList()
 	go s.handlePersistQueue()
 	go s.loop()
 }
@@ -187,14 +193,19 @@ func (s *memorySeriesStorage) Start() {
 // Stop implements Storage.
 func (s *memorySeriesStorage) Stop() error {
 	glog.Info("Stopping local storage...")
+
 	glog.Info("Stopping maintenance loop...")
 	close(s.loopStopping)
 	<-s.loopStopped
 
-	glog.Info("Stopping persist loop...")
+	glog.Info("Stopping persist queue...")
 	close(s.persistQueue)
 	<-s.persistStopped
 
+	glog.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
@@ -407,7 +418,83 @@ func (s *memorySeriesStorage) preloadChunksForRange(
 			return nil, nil
 		}
 	}
-	return series.preloadChunksForRange(from, through, fp, s.persistence)
+	return series.preloadChunksForRange(from, through, fp, s)
+}
+
+func (s *memorySeriesStorage) handleEvictList() {
+	ticker := time.NewTicker(maxEvictInterval)
+	count := 0
+loop:
+	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 has 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:
+			break loop
+		}
+	}
+	ticker.Stop()
+	glog.Info("Chunk eviction stopped.")
+	close(s.evictStopped)
+}
+
+// maybeEvict is a local helper method. Must only be called by handleEvictList.
+func (s *memorySeriesStorage) maybeEvict() {
+	numChunksToEvict := int(atomic.LoadInt64(&numMemChunks)) - s.maxMemoryChunks
+	if numChunksToEvict <= 0 {
+		return
+	}
+	chunkDescsToEvict := make([]*chunkDesc, numChunksToEvict)
+	for i := range chunkDescsToEvict {
+		e := s.evictList.Front()
+		if e == nil {
+			break
+		}
+		cd := e.Value.(*chunkDesc)
+		cd.evictListElement = nil
+		chunkDescsToEvict[i] = cd
+		s.evictList.Remove(e)
+	}
+	// Do the actual eviction in a goroutine as we might otherwise deadlock,
+	// in the following way: A chunk was unpinned completely and therefore
+	// scheduled for eviction. At the time we actually try to evict it,
+	// another goroutine is pinning the chunk. The pinning goroutine has
+	// currently locked the chunk and tries to send the evict request (to
+	// remove the chunk from the evict list) to the evictRequests
+	// channel. The send blocks because evictRequests is full. However, the
+	// goroutine that is supposed to empty the channel is wating for the
+	// chunkDesc lock to try to evict the chunk.
+	go func() {
+		for _, cd := range chunkDescsToEvict {
+			if cd == nil {
+				break
+			}
+			cd.maybeEvict()
+			// We don't care if the eviction succeeds. If the chunk
+			// was pinned in the meantime, it will be added to the
+			// evict list once it gets unpinned again.
+		}
+	}()
 }
 
 func (s *memorySeriesStorage) handlePersistQueue() {
@@ -430,10 +517,10 @@ func (s *memorySeriesStorage) handlePersistQueue() {
 			s.persistence.setDirty(true)
 			continue
 		}
-		req.chunkDesc.unpin()
+		req.chunkDesc.unpin(s.evictRequests)
 		chunkOps.WithLabelValues(persistAndUnpin).Inc()
 	}
-	glog.Info("Persist loop stopped.")
+	glog.Info("Persist queue drained and stopped.")
 	close(s.persistStopped)
 }
 
@@ -471,11 +558,9 @@ func (s *memorySeriesStorage) waitForNextFP(numberOfFPs int) bool {
 }
 
 func (s *memorySeriesStorage) loop() {
-	evictTicker := time.NewTicker(s.evictInterval)
 	checkpointTicker := time.NewTicker(s.checkpointInterval)
 
 	defer func() {
-		evictTicker.Stop()
 		checkpointTicker.Stop()
 		glog.Info("Maintenance loop stopped.")
 		close(s.loopStopped)
@@ -550,47 +635,9 @@ loop:
 			break loop
 		case <-checkpointTicker.C:
 			s.persistence.checkpointSeriesMapAndHeads(s.fpToSeries, s.fpLocker)
-		case <-evictTicker.C:
-			// TODO: Change this to be based on number of chunks in memory.
-			glog.Info("Evicting chunks...")
-			begin := time.Now()
-
-			for m := range s.fpToSeries.iter() {
-				select {
-				case <-s.loopStopping:
-					glog.Info("Interrupted evicting chunks.")
-					break loop
-				default:
-					// Keep going.
-				}
-				s.fpLocker.Lock(m.fp)
-				allEvicted, headChunkToPersist := m.series.evictOlderThan(
-					clientmodel.TimestampFromTime(time.Now()).Add(-1 * s.evictAfter),
-				)
-				if allEvicted {
-					s.fpToSeries.del(m.fp)
-					s.numSeries.Dec()
-					if err := s.persistence.archiveMetric(
-						m.fp, m.series.metric, m.series.firstTime(), m.series.lastTime(),
-					); err != nil {
-						glog.Errorf("Error archiving metric %v: %v", m.series.metric, err)
-					} else {
-						s.seriesOps.WithLabelValues(archive).Inc()
-					}
-				}
-				s.fpLocker.Unlock(m.fp)
-				// Queue outside of lock!
-				if headChunkToPersist != nil {
-					s.persistQueue <- persistRequest{m.fp, headChunkToPersist}
-				}
-			}
-
-			duration := time.Since(begin)
-			s.evictDuration.Set(float64(duration) / float64(time.Millisecond))
-			glog.Infof("Done evicting chunks in %v.", duration)
 		case fp := <-memoryFingerprints:
 			s.purgeSeries(fp, clientmodel.TimestampFromTime(time.Now()).Add(-1*s.purgeAfter))
-			// TODO: Move chunkdesc eviction, head chunk closing, and archiving here.
+			s.maintainSeries(fp)
 			s.seriesOps.WithLabelValues(memoryMaintenance).Inc()
 		case fp := <-archivedFingerprints:
 			s.purgeSeries(fp, clientmodel.TimestampFromTime(time.Now()).Add(-1*s.purgeAfter))
@@ -604,6 +651,56 @@ loop:
 	}
 }
 
+// maintainSeries closes the head chunk if not touched in a while. It archives a
+// series if all chunks are evicted. It evicts chunkDescs if there are too many.
+func (s *memorySeriesStorage) maintainSeries(fp clientmodel.Fingerprint) {
+	var headChunkToPersist *chunkDesc
+	s.fpLocker.Lock(fp)
+	defer func() {
+		s.fpLocker.Unlock(fp)
+		// Queue outside of lock!
+		if headChunkToPersist != nil {
+			s.persistQueue <- persistRequest{fp, headChunkToPersist}
+		}
+	}()
+
+	series, ok := s.fpToSeries.get(fp)
+	if !ok {
+		return
+	}
+	iOldestNotEvicted := -1
+	for i, cd := range series.chunkDescs {
+		if !cd.isEvicted() {
+			iOldestNotEvicted = i
+			break
+		}
+	}
+
+	// Archive if all chunks are evicted.
+	if iOldestNotEvicted == -1 {
+		s.fpToSeries.del(fp)
+		s.numSeries.Dec()
+		if err := s.persistence.archiveMetric(
+			fp, series.metric, series.firstTime(), series.lastTime(),
+		); err != nil {
+			glog.Errorf("Error archiving metric %v: %v", series.metric, err)
+		} else {
+			s.seriesOps.WithLabelValues(archive).Inc()
+		}
+		return
+	}
+	// If we are here, the series is not archived, so check for chunkDesc
+	// eviction next and then if the head chunk needs to be persisted.
+	series.evictChunkDescs(iOldestNotEvicted)
+	if !series.headChunkPersisted && time.Now().Sub(series.head().firstTime().Time()) > headChunkTimeout {
+		series.headChunkPersisted = true
+		// Since we cannot modify the head chunk from now on, we
+		// don't need to bother with cloning anymore.
+		series.headChunkUsedByIterator = false
+		headChunkToPersist = series.head()
+	}
+}
+
 // purgeSeries purges chunks older than beforeTime from a series. If the series
 // contains no chunks after the purge, it is dropped entirely.
 func (s *memorySeriesStorage) purgeSeries(fp clientmodel.Fingerprint, beforeTime clientmodel.Timestamp) {
@@ -635,7 +732,6 @@ func (s *memorySeriesStorage) purgeSeries(fp clientmodel.Fingerprint, beforeTime
 		}
 		return
 	}
-
 	// Deal with archived series.
 	has, firstTime, lastTime, err := s.persistence.hasArchivedMetric(fp)
 	if err != nil {
@@ -648,6 +744,7 @@ func (s *memorySeriesStorage) purgeSeries(fp clientmodel.Fingerprint, beforeTime
 	}
 
 	newFirstTime, _, allDropped, err := s.persistence.dropChunks(fp, beforeTime)
+	glog.Infoln("DEBUG:", newFirstTime, allDropped)
 	if err != nil {
 		glog.Error("Error purging persisted chunks: ", err)
 	}
@@ -685,8 +782,6 @@ func (s *memorySeriesStorage) Describe(ch chan<- *prometheus.Desc) {
 	s.seriesOps.Describe(ch)
 	ch <- s.ingestedSamplesCount.Desc()
 	ch <- s.invalidPreloadRequestsCount.Desc()
-	ch <- s.purgeDuration.Desc()
-	ch <- s.evictDuration.Desc()
 
 	ch <- persistQueueCapDesc
 
@@ -705,8 +800,6 @@ func (s *memorySeriesStorage) Collect(ch chan<- prometheus.Metric) {
 	s.seriesOps.Collect(ch)
 	ch <- s.ingestedSamplesCount
 	ch <- s.invalidPreloadRequestsCount
-	ch <- s.purgeDuration
-	ch <- s.evictDuration
 
 	ch <- persistQueueCapGauge
 

storage/local/storage_test.go

@@ -20,6 +20,7 @@ import (
 	"testing/quick"
 	"time"
 
+	"github.com/golang/glog"
 	clientmodel "github.com/prometheus/client_golang/model"
 	"github.com/prometheus/prometheus/storage/metric"
 	"github.com/prometheus/prometheus/utility/test"
@@ -42,8 +43,7 @@ func TestLoop(t *testing.T) {
 	directory := test.NewTemporaryDirectory("test_storage", t)
 	defer directory.Close()
 	o := &MemorySeriesStorageOptions{
-		MemoryEvictionInterval:     100 * time.Millisecond,
-		MemoryRetentionPeriod:      time.Hour,
+		MemoryChunks:               50,
 		PersistenceRetentionPeriod: 24 * 7 * time.Hour,
 		PersistenceStoragePath:     directory.Path(),
 		CheckpointInterval:         250 * time.Millisecond,
@@ -73,16 +73,28 @@ func TestChunk(t *testing.T) {
 
 	for m := range s.(*memorySeriesStorage).fpToSeries.iter() {
 		s.(*memorySeriesStorage).fpLocker.Lock(m.fp)
-		for i, v := range m.series.values() {
+
+		var values metric.Values
+		for _, cd := range m.series.chunkDescs {
+			if cd.isEvicted() {
+				continue
+			}
+			for sample := range cd.chunk.values() {
+				values = append(values, *sample)
+			}
+		}
+
+		for i, v := range values {
 			if samples[i].Timestamp != v.Timestamp {
-				t.Fatalf("%d. Got %v; want %v", i, v.Timestamp, samples[i].Timestamp)
+				t.Errorf("%d. Got %v; want %v", i, v.Timestamp, samples[i].Timestamp)
 			}
 			if samples[i].Value != v.Value {
-				t.Fatalf("%d. Got %v; want %v", i, v.Value, samples[i].Value)
+				t.Errorf("%d. Got %v; want %v", i, v.Value, samples[i].Value)
 			}
 		}
 		s.(*memorySeriesStorage).fpLocker.Unlock(m.fp)
 	}
+	glog.Info("test done, closing")
 }
 
 func TestGetValueAtTime(t *testing.T) {
@@ -362,37 +374,12 @@ func TestEvictAndPurgeSeries(t *testing.T) {
 		t.Fatal("could not find series")
 	}
 
-	// Evict everything except head chunk.
-	allEvicted, headChunkToPersist := series.evictOlderThan(1998)
-	// Head chunk not yet old enough, should get false, false:
-	if allEvicted {
-		t.Error("allEvicted with head chunk not yet old enough")
-	}
-	if headChunkToPersist != nil {
-		t.Error("persistHeadChunk is not nil although head chunk is not old enough")
-	}
-	// Evict everything.
-	allEvicted, headChunkToPersist = series.evictOlderThan(10000)
-	// Since the head chunk is not yet persisted, we should get false, true:
-	if allEvicted {
-		t.Error("allEvicted with head chuk not yet persisted")
-	}
-	if headChunkToPersist == nil {
-		t.Error("headChunkToPersist is nil although head chunk is old enough")
-	}
-	// Persist head chunk as requested.
+	// Persist head chunk so we can safely archive.
+	series.headChunkPersisted = true
 	ms.persistQueue <- persistRequest{fp, series.head()}
 	time.Sleep(time.Second) // Give time for persisting to happen.
-	allEvicted, headChunkToPersist = series.evictOlderThan(10000)
-	// Now we should really see everything gone.
-	if !allEvicted {
-		t.Error("not allEvicted")
-	}
-	if headChunkToPersist != nil {
-		t.Error("headChunkToPersist is not nil although already persisted")
-	}
 
-	// Now archive as it would usually be done in the evictTicker loop.
+	// Archive metrics.
 	ms.fpToSeries.del(fp)
 	if err := ms.persistence.archiveMetric(
 		fp, series.metric, series.firstTime(), series.lastTime(),
@@ -491,8 +478,7 @@ func BenchmarkFuzz(b *testing.B) {
 	directory := test.NewTemporaryDirectory("test_storage", b)
 	defer directory.Close()
 	o := &MemorySeriesStorageOptions{
-		MemoryEvictionInterval:     time.Second,
-		MemoryRetentionPeriod:      10 * time.Minute,
+		MemoryChunks:               100,
 		PersistenceRetentionPeriod: time.Hour,
 		PersistenceStoragePath:     directory.Path(),
 		CheckpointInterval:         3 * time.Second,

storage/local/test_helpers.go

@@ -36,8 +36,7 @@ func (t *testStorageCloser) Close() {
 func NewTestStorage(t testing.TB) (Storage, test.Closer) {
 	directory := test.NewTemporaryDirectory("test_storage", t)
 	o := &MemorySeriesStorageOptions{
-		MemoryEvictionInterval:     time.Minute,
-		MemoryRetentionPeriod:      time.Hour,
+		MemoryChunks:               1000000,
 		PersistenceRetentionPeriod: 24 * 7 * time.Hour,
 		PersistenceStoragePath:     directory.Path(),
 		CheckpointInterval:         time.Hour,