Add latency histogram for fetching index cache (thanos-io#6749)

* add latency histogram for fetching index cache Signed-off-by: Ben Ye <benye@amazon.com> * update changelog Signed-off-by: Ben Ye <benye@amazon.com> * use timer Signed-off-by: Ben Ye <benye@amazon.com> --------- Signed-off-by: Ben Ye <benye@amazon.com>
Preethivika · Sep 26, 2023 · 20d2900 · 20d2900
1 parent 686dd60
commit 20d2900
Show file tree

Hide file tree

Showing 4 changed files with 41 additions and 9 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -16,6 +16,7 @@ We use *breaking :warning:* to mark changes that are not backward compatible (re
 
 - [#6605](https://github.com/thanos-io/thanos/pull/6605) Query Frontend: Support vertical sharding binary expression with metric name when no matching labels specified.
 - [#6308](https://github.com/thanos-io/thanos/pull/6308) Ruler: Support configuration flag that allows customizing template for alert message.
+- [#6749](https://github.com/thanos-io/thanos/pull/6308) Store Gateway: Added `thanos_store_index_cache_fetch_duration_seconds` histogram for tracking latency of fetching data from index cache. 
 
 ### Changed
 

diff --git a/pkg/store/cache/cache.go b/pkg/store/cache/cache.go
@@ -61,6 +61,7 @@ type commonMetrics struct {
 	requestTotal  *prometheus.CounterVec
 	hitsTotal     *prometheus.CounterVec
 	dataSizeBytes *prometheus.HistogramVec
+	fetchLatency  *prometheus.HistogramVec
 }
 
 func newCommonMetrics(reg prometheus.Registerer) *commonMetrics {
@@ -80,6 +81,11 @@ func newCommonMetrics(reg prometheus.Registerer) *commonMetrics {
 				32, 256, 512, 1024, 32 * 1024, 256 * 1024, 512 * 1024, 1024 * 1024, 32 * 1024 * 1024, 64 * 1024 * 1024, 128 * 1024 * 1024, 256 * 1024 * 1024, 512 * 1024 * 1024,
 			},
 		}, []string{"item_type"}),
+		fetchLatency: promauto.With(reg).NewHistogramVec(prometheus.HistogramOpts{
+			Name:    "thanos_store_index_cache_fetch_duration_seconds",
+			Help:    "Histogram to track latency to fetch items from index cache",
+			Buckets: []float64{0.01, 0.1, 0.3, 0.6, 1, 3, 6, 10, 15, 20, 30, 45, 60, 90, 120},
+		}, []string{"item_type"}),
 	}
 }
 

diff --git a/pkg/store/cache/inmemory.go b/pkg/store/cache/inmemory.go
@@ -302,6 +302,9 @@ func (c *InMemoryIndexCache) StorePostings(blockID ulid.ULID, l labels.Label, v
 // FetchMultiPostings fetches multiple postings - each identified by a label -
 // and returns a map containing cache hits, along with a list of missing keys.
 func (c *InMemoryIndexCache) FetchMultiPostings(_ context.Context, blockID ulid.ULID, keys []labels.Label) (hits map[labels.Label][]byte, misses []labels.Label) {
+	timer := prometheus.NewTimer(c.commonMetrics.fetchLatency.WithLabelValues(cacheTypePostings))
+	defer timer.ObserveDuration()
+
 	hits = map[labels.Label][]byte{}
 
 	blockIDKey := blockID.String()
@@ -325,6 +328,9 @@ func (c *InMemoryIndexCache) StoreExpandedPostings(blockID ulid.ULID, matchers [
 
 // FetchExpandedPostings fetches expanded postings and returns cached data and a boolean value representing whether it is a cache hit or not.
 func (c *InMemoryIndexCache) FetchExpandedPostings(_ context.Context, blockID ulid.ULID, matchers []*labels.Matcher) ([]byte, bool) {
+	timer := prometheus.NewTimer(c.commonMetrics.fetchLatency.WithLabelValues(cacheTypeExpandedPostings))
+	defer timer.ObserveDuration()
+
 	if b, ok := c.get(cacheTypeExpandedPostings, cacheKey{blockID.String(), cacheKeyExpandedPostings(labelMatchersToString(matchers)), ""}); ok {
 		return b, true
 	}
@@ -341,6 +347,9 @@ func (c *InMemoryIndexCache) StoreSeries(blockID ulid.ULID, id storage.SeriesRef
 // FetchMultiSeries fetches multiple series - each identified by ID - from the cache
 // and returns a map containing cache hits, along with a list of missing IDs.
 func (c *InMemoryIndexCache) FetchMultiSeries(_ context.Context, blockID ulid.ULID, ids []storage.SeriesRef) (hits map[storage.SeriesRef][]byte, misses []storage.SeriesRef) {
+	timer := prometheus.NewTimer(c.commonMetrics.fetchLatency.WithLabelValues(cacheTypeSeries))
+	defer timer.ObserveDuration()
+
 	hits = map[storage.SeriesRef][]byte{}
 
 	blockIDKey := blockID.String()

diff --git a/pkg/store/cache/memcached.go b/pkg/store/cache/memcached.go
@@ -33,15 +33,18 @@ type RemoteIndexCache struct {
 	compressionScheme string
 
 	// Metrics.
-	postingRequests              prometheus.Counter
-	seriesRequests               prometheus.Counter
-	expandedPostingRequests      prometheus.Counter
-	postingHits                  prometheus.Counter
-	seriesHits                   prometheus.Counter
-	expandedPostingHits          prometheus.Counter
-	postingDataSizeBytes         prometheus.Observer
-	expandedPostingDataSizeBytes prometheus.Observer
-	seriesDataSizeBytes          prometheus.Observer
+	postingRequests               prometheus.Counter
+	seriesRequests                prometheus.Counter
+	expandedPostingRequests       prometheus.Counter
+	postingHits                   prometheus.Counter
+	seriesHits                    prometheus.Counter
+	expandedPostingHits           prometheus.Counter
+	postingDataSizeBytes          prometheus.Observer
+	expandedPostingDataSizeBytes  prometheus.Observer
+	seriesDataSizeBytes           prometheus.Observer
+	postingsFetchDuration         prometheus.Observer
+	expandedPostingsFetchDuration prometheus.Observer
+	seriesFetchDuration           prometheus.Observer
 }
 
 // NewRemoteIndexCache makes a new RemoteIndexCache.
@@ -68,6 +71,10 @@ func NewRemoteIndexCache(logger log.Logger, cacheClient cacheutil.RemoteCacheCli
 	c.seriesDataSizeBytes = commonMetrics.dataSizeBytes.WithLabelValues(cacheTypeSeries)
 	c.expandedPostingDataSizeBytes = commonMetrics.dataSizeBytes.WithLabelValues(cacheTypeExpandedPostings)
 
+	c.postingsFetchDuration = commonMetrics.fetchLatency.WithLabelValues(cacheTypePostings)
+	c.seriesFetchDuration = commonMetrics.fetchLatency.WithLabelValues(cacheTypeSeries)
+	c.expandedPostingsFetchDuration = commonMetrics.fetchLatency.WithLabelValues(cacheTypeExpandedPostings)
+
 	level.Info(logger).Log("msg", "created index cache")
 
 	return c, nil
@@ -88,6 +95,9 @@ func (c *RemoteIndexCache) StorePostings(blockID ulid.ULID, l labels.Label, v []
 // and returns a map containing cache hits, along with a list of missing keys.
 // In case of error, it logs and return an empty cache hits map.
 func (c *RemoteIndexCache) FetchMultiPostings(ctx context.Context, blockID ulid.ULID, lbls []labels.Label) (hits map[labels.Label][]byte, misses []labels.Label) {
+	timer := prometheus.NewTimer(c.postingsFetchDuration)
+	defer timer.ObserveDuration()
+
 	keys := make([]string, 0, len(lbls))
 
 	blockIDKey := blockID.String()
@@ -138,6 +148,9 @@ func (c *RemoteIndexCache) StoreExpandedPostings(blockID ulid.ULID, keys []*labe
 // and returns a map containing cache hits, along with a list of missing keys.
 // In case of error, it logs and return an empty cache hits map.
 func (c *RemoteIndexCache) FetchExpandedPostings(ctx context.Context, blockID ulid.ULID, lbls []*labels.Matcher) ([]byte, bool) {
+	timer := prometheus.NewTimer(c.postingsFetchDuration)
+	defer timer.ObserveDuration()
+
 	key := cacheKey{blockID.String(), cacheKeyExpandedPostings(labelMatchersToString(lbls)), c.compressionScheme}.string()
 
 	// Fetch the keys from memcached in a single request.
@@ -169,6 +182,9 @@ func (c *RemoteIndexCache) StoreSeries(blockID ulid.ULID, id storage.SeriesRef,
 // and returns a map containing cache hits, along with a list of missing IDs.
 // In case of error, it logs and return an empty cache hits map.
 func (c *RemoteIndexCache) FetchMultiSeries(ctx context.Context, blockID ulid.ULID, ids []storage.SeriesRef) (hits map[storage.SeriesRef][]byte, misses []storage.SeriesRef) {
+	timer := prometheus.NewTimer(c.postingsFetchDuration)
+	defer timer.ObserveDuration()
+
 	keys := make([]string, 0, len(ids))
 
 	blockIDKey := blockID.String()