Shared in-memory index cache for queriers with blocks storage

CHANGELOG.md

@@ -25,6 +25,7 @@
 * [CHANGE] Updated Prometheus dependency to v2.16.0. This Prometheus version uses Active Query Tracker to limit concurrent queries. In order to keep `-querier.max-concurrent` working, Active Query Tracker is enabled by default, and is configured to store its data to `active-query-tracker` directory (relative to current directory when Cortex started). This can be changed by using `-querier.active-query-tracker-dir` option. Purpose of Active Query Tracker is to log queries that were running when Cortex crashes. This logging happens on next Cortex start. #2088
 * [CHANGE] Experimental TSDB: TSDB head compaction interval and concurrency is now configurable (defaults to 1 min interval and 5 concurrent head compactions). New options: `-experimental.tsdb.head-compaction-interval` and `-experimental.tsdb.head-compaction-concurrency`. #2172
 * [CHANGE] Remove fluentd-based billing infrastructure and flags such as `-distributor.enable-billing`. #1491
+* [CHANGE] Experimental TSDB: the querier in-memory index cache used by the experimental blocks storage shifted from per-tenant to per-querier. The `-experimental.tsdb.bucket-store.index-cache-size-bytes` now configures the per-querier index cache max size instead of a per-tenant cache and its default has been increased to 1GB. #2189
 * [FEATURE] Added a read-only local alertmanager config store using files named corresponding to their tenant id. #2125
 * [FEATURE] Added user sub rings to distribute users to a subset of ingesters. #1947
   * `--experimental.distributor.user-subring-size`

docs/configuration/config-file-reference.md

@@ -2139,10 +2139,10 @@ bucket_store:
   # CLI flag: -experimental.tsdb.bucket-store.sync-interval
   [sync_interval: <duration> | default = 5m0s]
 
-  # Size - in bytes - of a per-tenant in-memory index cache used to speed up
-  # blocks index lookups.
+  # Size in bytes of in-memory index cache used to speed up blocks index lookups
+  # (shared between all tenants).
   # CLI flag: -experimental.tsdb.bucket-store.index-cache-size-bytes
-  [index_cache_size_bytes: <int> | default = 262144000]
+  [index_cache_size_bytes: <int> | default = 1073741824]
 
   # Max size - in bytes - of a per-tenant chunk pool, used to reduce memory
   # allocations.

docs/operations/blocks-storage.md

@@ -133,10 +133,10 @@ tsdb:
     # CLI flag: -experimental.tsdb.bucket-store.sync-interval
     [sync_interval: <duration> | default = 5m0s]
 
-    # Size - in bytes - of a per-tenant in-memory index cache used to speed up
-    # blocks index lookups.
+    # Size in bytes of in-memory index cache used to speed up blocks index
+    # lookups (shared between all tenants).
     # CLI flag: -experimental.tsdb.bucket-store.index-cache-size-bytes
-    [index_cache_size_bytes: <int> | default = 262144000]
+    [index_cache_size_bytes: <int> | default = 1073741824]
 
     # Max size - in bytes - of a per-tenant chunk pool, used to reduce memory
     # allocations.

integration/configs.go

@@ -50,6 +50,7 @@ var (
 		"-experimental.tsdb.bucket-store.sync-interval": "5s",
 		"-experimental.tsdb.retention-period":           "5m",
 		"-experimental.tsdb.ship-interval":              "1m",
+		"-experimental.tsdb.head-compaction-interval":   "1s",
 		"-experimental.tsdb.s3.access-key-id":           e2edb.MinioAccessKey,
 		"-experimental.tsdb.s3.secret-access-key":       e2edb.MinioSecretKey,
 		"-experimental.tsdb.s3.bucket-name":             "cortex",

integration/querier_test.go

@@ -0,0 +1,139 @@
+// +build integration
+
+package main
+
+import (
+	"testing"
+	"time"
+
+	"github.com/prometheus/common/model"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/cortexproject/cortex/integration/e2e"
+	e2edb "github.com/cortexproject/cortex/integration/e2e/db"
+	"github.com/cortexproject/cortex/integration/e2ecortex"
+)
+
+func TestQuerierWithBlocksStorage(t *testing.T) {
+	tests := map[string]struct {
+		flags map[string]string
+	}{
+		"querier running with ingester gRPC streaming disabled": {
+			flags: mergeFlags(BlocksStorageFlags, map[string]string{
+				"-querier.ingester-streaming": "false",
+			}),
+		},
+	}
+
+	for testName, testCfg := range tests {
+		t.Run(testName, func(t *testing.T) {
+			const blockRangePeriod = 5 * time.Second
+
+			s, err := e2e.NewScenario(networkName)
+			require.NoError(t, err)
+			defer s.Close()
+
+			// Configure the blocks storage to frequently compact TSDB head
+			// and ship blocks to the storage.
+			flags := mergeFlags(testCfg.flags, map[string]string{
+				"-experimental.tsdb.block-ranges-period":        blockRangePeriod.String(),
+				"-experimental.tsdb.ship-interval":              "1s",
+				"-experimental.tsdb.bucket-store.sync-interval": "1s",
+				"-experimental.tsdb.retention-period":           ((blockRangePeriod * 2) - 1).String(),
+			})
+
+			// Start dependencies.
+			consul := e2edb.NewConsul()
+			minio := e2edb.NewMinio(9000, flags["-experimental.tsdb.s3.bucket-name"])
+			require.NoError(t, s.StartAndWaitReady(consul, minio))
+
+			// Start Cortex components.
+			distributor := e2ecortex.NewDistributor("distributor", consul.NetworkHTTPEndpoint(), flags, "")
+			ingester := e2ecortex.NewIngester("ingester", consul.NetworkHTTPEndpoint(), flags, "")
+			querier := e2ecortex.NewQuerier("querier", consul.NetworkHTTPEndpoint(), flags, "")
+			require.NoError(t, s.StartAndWaitReady(distributor, ingester, querier))
+
+			// Wait until both the distributor and querier have updated the ring.
+			require.NoError(t, distributor.WaitSumMetrics(e2e.Equals(512), "cortex_ring_tokens_total"))
+			require.NoError(t, querier.WaitSumMetrics(e2e.Equals(512), "cortex_ring_tokens_total"))
+
+			c, err := e2ecortex.NewClient(distributor.HTTPEndpoint(), querier.HTTPEndpoint(), "", "user-1")
+			require.NoError(t, err)
+
+			// Push some series to Cortex.
+			series1Timestamp := time.Now()
+			series2Timestamp := series1Timestamp.Add(blockRangePeriod * 2)
+			series1, expectedVector1 := generateSeries("series_1", series1Timestamp)
+			series2, expectedVector2 := generateSeries("series_2", series2Timestamp)
+
+			res, err := c.Push(series1)
+			require.NoError(t, err)
+			require.Equal(t, 200, res.StatusCode)
+
+			res, err = c.Push(series2)
+			require.NoError(t, err)
+			require.Equal(t, 200, res.StatusCode)
+
+			// Wait until the TSDB head is compacted and shipped to the storage.
+			// The shipped block contains the 1st series, while the 2ns series in in the head.
+			require.NoError(t, ingester.WaitSumMetrics(e2e.Equals(1), "cortex_ingester_shipper_uploads_total"))
+			require.NoError(t, ingester.WaitSumMetrics(e2e.Equals(2), "cortex_ingester_memory_series_created_total"))
+			require.NoError(t, ingester.WaitSumMetrics(e2e.Equals(1), "cortex_ingester_memory_series_removed_total"))
+			require.NoError(t, ingester.WaitSumMetrics(e2e.Equals(1), "cortex_ingester_memory_series"))
+
+			// Push another series to further compact another block and delete the first block
+			// due to expired retention.
+			series3Timestamp := series2Timestamp.Add(blockRangePeriod * 2)
+			series3, expectedVector3 := generateSeries("series_3", series3Timestamp)
+
+			res, err = c.Push(series3)
+			require.NoError(t, err)
+			require.Equal(t, 200, res.StatusCode)
+
+			require.NoError(t, ingester.WaitSumMetrics(e2e.Equals(2), "cortex_ingester_shipper_uploads_total"))
+			require.NoError(t, ingester.WaitSumMetrics(e2e.Equals(3), "cortex_ingester_memory_series_created_total"))
+			require.NoError(t, ingester.WaitSumMetrics(e2e.Equals(2), "cortex_ingester_memory_series_removed_total"))
+			require.NoError(t, ingester.WaitSumMetrics(e2e.Equals(1), "cortex_ingester_memory_series"))
+
+			// Wait until the querier has synched the new uploaded blocks.
+			require.NoError(t, querier.WaitSumMetrics(e2e.Equals(2), "cortex_querier_bucket_store_blocks_loaded"))
+
+			// Query back the series (1 only in the storage, 1 only in the ingesters, 1 on both).
+			// TODO: apparently Thanos has a bug which cause a block to not be considered if the
+			//       query timetamp matches the block max timestamp
+			series1Timestamp = series1Timestamp.Add(time.Duration(time.Millisecond))
+			expectedVector1[0].Timestamp = model.Time(e2e.TimeToMilliseconds(series1Timestamp))
+
+			result, err := c.Query("series_1", series1Timestamp)
+			require.NoError(t, err)
+			require.Equal(t, model.ValVector, result.Type())
+			assert.Equal(t, expectedVector1, result.(model.Vector))
+
+			result, err = c.Query("series_2", series2Timestamp)
+			require.NoError(t, err)
+			require.Equal(t, model.ValVector, result.Type())
+			assert.Equal(t, expectedVector2, result.(model.Vector))
+
+			result, err = c.Query("series_3", series3Timestamp)
+			require.NoError(t, err)
+			require.Equal(t, model.ValVector, result.Type())
+			assert.Equal(t, expectedVector3, result.(model.Vector))
+
+			// Check the in-memory index cache metrics (in the querier).
+			require.NoError(t, querier.WaitSumMetrics(e2e.Equals(2*2), "cortex_querier_blocks_index_cache_items"))             // 2 series both for postings and series cache
+			require.NoError(t, querier.WaitSumMetrics(e2e.Equals(2*2), "cortex_querier_blocks_index_cache_items_added_total")) // 2 series both for postings and series cache
+			require.NoError(t, querier.WaitSumMetrics(e2e.Equals(0), "cortex_querier_blocks_index_cache_hits_total"))          // no cache hit cause the cache was empty
+
+			// Query back again the 1st series from storage. This time it should use the index cache.
+			result, err = c.Query("series_1", series1Timestamp)
+			require.NoError(t, err)
+			require.Equal(t, model.ValVector, result.Type())
+			assert.Equal(t, expectedVector1, result.(model.Vector))
+
+			require.NoError(t, querier.WaitSumMetrics(e2e.Equals(2*2), "cortex_querier_blocks_index_cache_items"))             // as before
+			require.NoError(t, querier.WaitSumMetrics(e2e.Equals(2*2), "cortex_querier_blocks_index_cache_items_added_total")) // as before
+			require.NoError(t, querier.WaitSumMetrics(e2e.Equals(2), "cortex_querier_blocks_index_cache_hits_total"))          // this time has used the index cache
+		})
+	}
+}

pkg/querier/block_store.go

@@ -12,6 +12,7 @@ import (
 
 	"github.com/go-kit/kit/log"
 	"github.com/go-kit/kit/log/level"
+	"github.com/pkg/errors"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/thanos-io/thanos/pkg/block"
 	"github.com/thanos-io/thanos/pkg/model"
@@ -31,16 +32,20 @@ import (
 
 // UserStore is a multi-tenant version of Thanos BucketStore
 type UserStore struct {
-	logger      log.Logger
-	cfg         tsdb.Config
-	bucket      objstore.Bucket
-	client      storepb.StoreClient
-	logLevel    logging.Level
-	tsdbMetrics *tsdbBucketStoreMetrics
+	logger             log.Logger
+	cfg                tsdb.Config
+	bucket             objstore.Bucket
+	client             storepb.StoreClient
+	logLevel           logging.Level
+	bucketStoreMetrics *tsdbBucketStoreMetrics
+	indexCacheMetrics  *tsdbIndexCacheMetrics
 
 	syncMint model.TimeOrDurationValue
 	syncMaxt model.TimeOrDurationValue
 
+	// Index cache shared across all tenants.
+	indexCache storecache.IndexCache
+
 	// Keeps a bucket store for each tenant.
 	storesMu sync.RWMutex
 	stores   map[string]*store.BucketStore
@@ -55,16 +60,20 @@ type UserStore struct {
 
 // NewUserStore returns a new UserStore
 func NewUserStore(cfg tsdb.Config, bucketClient objstore.Bucket, logLevel logging.Level, logger log.Logger, registerer prometheus.Registerer) (*UserStore, error) {
+	var err error
+
 	workersCtx, workersCancel := context.WithCancel(context.Background())
+	indexCacheRegistry := prometheus.NewRegistry()
 
 	u := &UserStore{
-		logger:        logger,
-		cfg:           cfg,
-		bucket:        bucketClient,
-		stores:        map[string]*store.BucketStore{},
-		logLevel:      logLevel,
-		tsdbMetrics:   newTSDBBucketStoreMetrics(),
-		workersCancel: workersCancel,
+		logger:             logger,
+		cfg:                cfg,
+		bucket:             bucketClient,
+		stores:             map[string]*store.BucketStore{},
+		logLevel:           logLevel,
+		bucketStoreMetrics: newTSDBBucketStoreMetrics(),
+		indexCacheMetrics:  newTSDBIndexCacheMetrics(indexCacheRegistry),
+		workersCancel:      workersCancel,
 		syncTimes: prometheus.NewHistogram(prometheus.HistogramOpts{
 			Name:    "cortex_querier_blocks_sync_seconds",
 			Help:    "The total time it takes to perform a sync stores",
@@ -73,15 +82,20 @@ func NewUserStore(cfg tsdb.Config, bucketClient objstore.Bucket, logLevel loggin
 	}
 
 	// Configure the time range to sync all blocks.
-	if err := u.syncMint.Set("0000-01-01T00:00:00Z"); err != nil {
+	if err = u.syncMint.Set("0000-01-01T00:00:00Z"); err != nil {
 		return nil, err
 	}
-	if err := u.syncMaxt.Set("9999-12-31T23:59:59Z"); err != nil {
+	if err = u.syncMaxt.Set("9999-12-31T23:59:59Z"); err != nil {
 		return nil, err
 	}
 
+	// Init the index cache.
+	if u.indexCache, err = tsdb.NewIndexCache(cfg.BucketStore, logger, indexCacheRegistry); err != nil {
+		return nil, errors.Wrap(err, "create index cache")
+	}
+
 	if registerer != nil {
-		registerer.MustRegister(u.syncTimes, u.tsdbMetrics)
+		registerer.MustRegister(u.syncTimes, u.bucketStoreMetrics, u.indexCacheMetrics)
 	}
 
 	serv := grpc.NewServer()
@@ -357,16 +371,6 @@ func (u *UserStore) getOrCreateStore(userID string) (*store.BucketStore, error)
 	userBkt := tsdb.NewUserBucketClient(userID, u.bucket)
 
 	reg := prometheus.NewRegistry()
-	indexCacheSizeBytes := u.cfg.BucketStore.IndexCacheSizeBytes
-	maxItemSizeBytes := indexCacheSizeBytes / 2
-	indexCache, err := storecache.NewInMemoryIndexCacheWithConfig(userLogger, reg, storecache.InMemoryIndexCacheConfig{
-		MaxSize:     storecache.Bytes(indexCacheSizeBytes),
-		MaxItemSize: storecache.Bytes(maxItemSizeBytes),
-	})
-	if err != nil {
-		return nil, err
-	}
-
 	fetcher, err := block.NewMetaFetcher(
 		userLogger,
 		u.cfg.BucketStore.MetaSyncConcurrency,
@@ -385,7 +389,7 @@ func (u *UserStore) getOrCreateStore(userID string) (*store.BucketStore, error)
 		userBkt,
 		fetcher,
 		filepath.Join(u.cfg.BucketStore.SyncDir, userID),
-		indexCache,
+		u.indexCache,
 		uint64(u.cfg.BucketStore.MaxChunkPoolBytes),
 		u.cfg.BucketStore.MaxSampleCount,
 		u.cfg.BucketStore.MaxConcurrent,
@@ -402,7 +406,7 @@ func (u *UserStore) getOrCreateStore(userID string) (*store.BucketStore, error)
 	}
 
 	u.stores[userID] = bs
-	u.tsdbMetrics.addUserRegistry(userID, reg)
+	u.bucketStoreMetrics.addUserRegistry(userID, reg)
 
 	return bs, nil
 }