Implementy max_fetched_data_bytes_per_query limit

Signed-off-by: 🌲 Harry 🌊 John 🏔 <johrry@amazon.com>

Author: harry671003

CHANGELOG.md

@@ -55,6 +55,7 @@
 * [FEATURE] Added `-api.http-request-headers-to-log` allowing for the addition of HTTP Headers to logs #4803
 * [FEATURE] Distributor: Added a new limit `-validation.max-labels-size-bytes` allowing to limit the combined size of labels for each timeseries. #4848
 * [FEATURE] Storage/Bucket: Added `-*.s3.bucket-lookup-type` allowing to configure the s3 bucket lookup type. #4794
+* [FEATURE] Querier: Added a new limit `-querier.max-fetched-data-bytes-per-query` allowing to limit the maximum size of all data in bytes that a query can fetch from each ingester and storage. #4854
 * [BUGFIX] Memberlist: Add join with no retrying when starting service. #4804
 * [BUGFIX] Ruler: Fix /ruler/rule_groups returns YAML with extra fields. #4767
 

docs/configuration/config-file-reference.md

@@ -2673,12 +2673,19 @@ The `limits_config` configures default and per-tenant limits imposed by Cortex s
 # CLI flag: -querier.max-fetched-series-per-query
 [max_fetched_series_per_query: <int> | default = 0]
 
-# The maximum size of all chunks in bytes that a query can fetch from each
-# ingester and storage. This limit is enforced in the querier and ruler only
-# when running Cortex with blocks storage. 0 to disable.
+# Deprecated (user max-fetched-data-bytes-per-query instead): The maximum size
+# of all chunks in bytes that a query can fetch from each ingester and storage.
+# This limit is enforced in the querier and ruler only when running Cortex with
+# blocks storage. 0 to disable.
 # CLI flag: -querier.max-fetched-chunk-bytes-per-query
 [max_fetched_chunk_bytes_per_query: <int> | default = 0]
 
+# The maximum combined size of all data that a query can fetch from each
+# ingester and storage. This limit is enforced in the querier and ruler only
+# when running Cortex with blocks storage. 0 to disable.
+# CLI flag: -querier.max-fetched-data-bytes-per-query
+[max_fetched_data_bytes_per_query: <int> | default = 0]
+
 # Limit how long back data (series and metadata) can be queried, up until
 # <lookback> duration ago. This limit is enforced in the query-frontend, querier
 # and ruler. If the requested time range is outside the allowed range, the

pkg/distributor/distributor_test.go

@@ -1032,7 +1032,7 @@ func TestDistributor_QueryStream_ShouldReturnErrorIfMaxChunksPerQueryLimitIsReac
 		limits:           limits,
 	})
 
-	ctx = limiter.AddQueryLimiterToContext(ctx, limiter.NewQueryLimiter(0, 0, maxChunksLimit))
+	ctx = limiter.AddQueryLimiterToContext(ctx, limiter.NewQueryLimiter(0, 0, maxChunksLimit, 0))
 
 	// Push a number of series below the max chunks limit. Each series has 1 sample,
 	// so expect 1 chunk per series when querying back.
@@ -1077,7 +1077,7 @@ func TestDistributor_QueryStream_ShouldReturnErrorIfMaxSeriesPerQueryLimitIsReac
 	ctx := user.InjectOrgID(context.Background(), "user")
 	limits := &validation.Limits{}
 	flagext.DefaultValues(limits)
-	ctx = limiter.AddQueryLimiterToContext(ctx, limiter.NewQueryLimiter(maxSeriesLimit, 0, 0))
+	ctx = limiter.AddQueryLimiterToContext(ctx, limiter.NewQueryLimiter(maxSeriesLimit, 0, 0, 0))
 
 	// Prepare distributors.
 	ds, _, _, _ := prepare(t, prepConfig{
@@ -1161,7 +1161,7 @@ func TestDistributor_QueryStream_ShouldReturnErrorIfMaxChunkBytesPerQueryLimitIs
 	var maxBytesLimit = (seriesToAdd) * responseChunkSize
 
 	// Update the limiter with the calculated limits.
-	ctx = limiter.AddQueryLimiterToContext(ctx, limiter.NewQueryLimiter(0, maxBytesLimit, 0))
+	ctx = limiter.AddQueryLimiterToContext(ctx, limiter.NewQueryLimiter(0, maxBytesLimit, 0, 0))
 
 	// Push a number of series below the max chunk bytes limit. Subtract one for the series added above.
 	writeReq = makeWriteRequest(0, seriesToAdd-1, 0)
@@ -1192,6 +1192,75 @@ func TestDistributor_QueryStream_ShouldReturnErrorIfMaxChunkBytesPerQueryLimitIs
 	assert.Equal(t, err, validation.LimitError(fmt.Sprintf(limiter.ErrMaxChunkBytesHit, maxBytesLimit)))
 }
 
+func TestDistributor_QueryStream_ShouldReturnErrorIfMaxDataBytesPerQueryLimitIsReached(t *testing.T) {
+	const seriesToAdd = 10
+
+	ctx := user.InjectOrgID(context.Background(), "user")
+	limits := &validation.Limits{}
+	flagext.DefaultValues(limits)
+
+	// Prepare distributors.
+	// Use replication factor of 2 to always read all the chunks from both ingesters,
+	// this guarantees us to always read the same chunks and have a stable test.
+	ds, _, _, _ := prepare(t, prepConfig{
+		numIngesters:      2,
+		happyIngesters:    2,
+		numDistributors:   1,
+		shardByAllLabels:  true,
+		limits:            limits,
+		replicationFactor: 2,
+	})
+
+	allSeriesMatchers := []*labels.Matcher{
+		labels.MustNewMatcher(labels.MatchRegexp, model.MetricNameLabel, ".+"),
+	}
+	// Push a single series to allow us to calculate the label size to calculate the limit for the test.
+	writeReq := &cortexpb.WriteRequest{}
+	writeReq.Timeseries = append(writeReq.Timeseries,
+		makeWriteRequestTimeseries([]cortexpb.LabelAdapter{{Name: model.MetricNameLabel, Value: "another_series"}}, 0, 0),
+	)
+	writeRes, err := ds[0].Push(ctx, writeReq)
+	assert.Equal(t, &cortexpb.WriteResponse{}, writeRes)
+	assert.Nil(t, err)
+	labelSizeResponse, err := ds[0].QueryStream(ctx, math.MinInt32, math.MaxInt32, allSeriesMatchers...)
+	require.NoError(t, err)
+
+	// Use the resulting chunks size to calculate the limit as (series to add + our test series) * the response chunk size.
+	var responseLabelSize = labelSizeResponse.LabelsSize()
+	var maxBytesLimit = (seriesToAdd) * responseLabelSize * 2 // Multiplying by RF because the limit is applied before de-duping.
+
+	// Update the limiter with the calculated limits.
+	ctx = limiter.AddQueryLimiterToContext(ctx, limiter.NewQueryLimiter(0, 0, 0, maxBytesLimit))
+
+	// Push a number of series below the max chunk bytes limit. Subtract one for the series added above.
+	writeReq = makeWriteRequest(0, seriesToAdd-1, 0)
+	writeRes, err = ds[0].Push(ctx, writeReq)
+	assert.Equal(t, &cortexpb.WriteResponse{}, writeRes)
+	assert.Nil(t, err)
+
+	// Since the number of chunk bytes is equal to the limit (but doesn't
+	// exceed it), we expect a query running on all series to succeed.
+	queryRes, err := ds[0].QueryStream(ctx, math.MinInt32, math.MaxInt32, allSeriesMatchers...)
+	require.NoError(t, err)
+	assert.Len(t, queryRes.Chunkseries, seriesToAdd)
+
+	// Push another series to exceed the chunk bytes limit once we'll query back all series.
+	writeReq = &cortexpb.WriteRequest{}
+	writeReq.Timeseries = append(writeReq.Timeseries,
+		makeWriteRequestTimeseries([]cortexpb.LabelAdapter{{Name: model.MetricNameLabel, Value: "another_series_1"}}, 0, 0),
+	)
+
+	writeRes, err = ds[0].Push(ctx, writeReq)
+	assert.Equal(t, &cortexpb.WriteResponse{}, writeRes)
+	assert.Nil(t, err)
+
+	// Since the aggregated chunk size is exceeding the limit, we expect
+	// a query running on all series to fail.
+	_, err = ds[0].QueryStream(ctx, math.MinInt32, math.MaxInt32, allSeriesMatchers...)
+	require.Error(t, err)
+	assert.Equal(t, err, validation.LimitError(fmt.Sprintf(limiter.ErrMaxDataBytesHit, maxBytesLimit)))
+}
+
 func TestDistributor_Push_LabelRemoval(t *testing.T) {
 	ctx := user.InjectOrgID(context.Background(), "user")
 
@@ -1930,7 +1999,7 @@ func TestDistributor_MetricsForLabelMatchers(t *testing.T) {
 			},
 			expectedResult:    []metric.Metric{},
 			expectedIngesters: numIngesters,
-			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0),
+			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0, 0),
 			expectedErr:       nil,
 		},
 		"should filter metrics by single matcher": {
@@ -1942,7 +2011,7 @@ func TestDistributor_MetricsForLabelMatchers(t *testing.T) {
 				{Metric: util.LabelsToMetric(fixtures[1].lbls)},
 			},
 			expectedIngesters: numIngesters,
-			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0),
+			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0, 0),
 			expectedErr:       nil,
 		},
 		"should filter metrics by multiple matchers": {
@@ -1954,7 +2023,7 @@ func TestDistributor_MetricsForLabelMatchers(t *testing.T) {
 				{Metric: util.LabelsToMetric(fixtures[0].lbls)},
 			},
 			expectedIngesters: numIngesters,
-			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0),
+			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0, 0),
 			expectedErr:       nil,
 		},
 		"should return all matching metrics even if their FastFingerprint collide": {
@@ -1966,7 +2035,7 @@ func TestDistributor_MetricsForLabelMatchers(t *testing.T) {
 				{Metric: util.LabelsToMetric(fixtures[4].lbls)},
 			},
 			expectedIngesters: numIngesters,
-			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0),
+			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0, 0),
 			expectedErr:       nil,
 		},
 		"should query only ingesters belonging to tenant's subring if shuffle sharding is enabled": {
@@ -1980,7 +2049,7 @@ func TestDistributor_MetricsForLabelMatchers(t *testing.T) {
 				{Metric: util.LabelsToMetric(fixtures[1].lbls)},
 			},
 			expectedIngesters: 3,
-			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0),
+			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0, 0),
 			expectedErr:       nil,
 		},
 		"should query all ingesters if shuffle sharding is enabled but shard size is 0": {
@@ -1994,7 +2063,7 @@ func TestDistributor_MetricsForLabelMatchers(t *testing.T) {
 				{Metric: util.LabelsToMetric(fixtures[1].lbls)},
 			},
 			expectedIngesters: numIngesters,
-			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0),
+			queryLimiter:      limiter.NewQueryLimiter(0, 0, 0, 0),
 			expectedErr:       nil,
 		},
 		"should return err if series limit is exhausted": {
@@ -2005,7 +2074,7 @@ func TestDistributor_MetricsForLabelMatchers(t *testing.T) {
 			},
 			expectedResult:    nil,
 			expectedIngesters: numIngesters,
-			queryLimiter:      limiter.NewQueryLimiter(1, 0, 0),
+			queryLimiter:      limiter.NewQueryLimiter(1, 0, 0, 0),
 			expectedErr:       validation.LimitError(fmt.Sprintf(limiter.ErrMaxSeriesHit, 1)),
 		},
 		"should not exhaust series limit when only one series is fetched": {
@@ -2016,7 +2085,7 @@ func TestDistributor_MetricsForLabelMatchers(t *testing.T) {
 				{Metric: util.LabelsToMetric(fixtures[2].lbls)},
 			},
 			expectedIngesters: numIngesters,
-			queryLimiter:      limiter.NewQueryLimiter(1, 0, 0),
+			queryLimiter:      limiter.NewQueryLimiter(1, 0, 0, 0),
 			expectedErr:       nil,
 		},
 	}
@@ -2116,7 +2185,7 @@ func BenchmarkDistributor_MetricsForLabelMatchers(b *testing.B) {
 			matchers: []*labels.Matcher{
 				mustNewMatcher(labels.MatchRegexp, model.MetricNameLabel, "foo.+"),
 			},
-			queryLimiter: limiter.NewQueryLimiter(100, 0, 0),
+			queryLimiter: limiter.NewQueryLimiter(100, 0, 0, 0),
 			expectedErr:  nil,
 		},
 	}

pkg/distributor/query.go

@@ -335,6 +335,10 @@ func (d *Distributor) queryIngesterStream(ctx context.Context, replicationSet ri
 				return nil, validation.LimitError(chunkBytesLimitErr.Error())
 			}
 
+			if labelBytesLimitErr := queryLimiter.AddDataBytes(resp.ChunksSize() + resp.LabelsSize()); labelBytesLimitErr != nil {
+				return nil, validation.LimitError(labelBytesLimitErr.Error())
+			}
+
 			for _, series := range resp.Timeseries {
 				if limitErr := queryLimiter.AddSeries(series.Labels); limitErr != nil {
 					return nil, validation.LimitError(limitErr.Error())
@@ -392,6 +396,7 @@ func (d *Distributor) queryIngesterStream(ctx context.Context, replicationSet ri
 
 	reqStats.AddFetchedSeries(uint64(len(resp.Chunkseries) + len(resp.Timeseries)))
 	reqStats.AddFetchedChunkBytes(uint64(resp.ChunksSize()))
+	reqStats.AddFetchedDataBytes(uint64(resp.LabelsSize() + resp.ChunksSize()))
 
 	return resp, nil
 }

pkg/frontend/transport/handler.go

@@ -60,10 +60,11 @@ type Handler struct {
 	roundTripper http.RoundTripper
 
 	// Metrics.
-	querySeconds *prometheus.CounterVec
-	querySeries  *prometheus.CounterVec
-	queryBytes   *prometheus.CounterVec
-	activeUsers  *util.ActiveUsersCleanupService
+	querySeconds   *prometheus.CounterVec
+	querySeries    *prometheus.CounterVec
+	queryBytes     *prometheus.CounterVec
+	queryDataBytes *prometheus.CounterVec
+	activeUsers    *util.ActiveUsersCleanupService
 }
 
 // NewHandler creates a new frontend handler.
@@ -90,10 +91,16 @@ func NewHandler(cfg HandlerConfig, roundTripper http.RoundTripper, log log.Logge
 			Help: "Size of all chunks fetched to execute a query in bytes.",
 		}, []string{"user"})
 
+		h.queryDataBytes = promauto.With(reg).NewCounterVec(prometheus.CounterOpts{
+			Name: "cortex_query_fetched_data_bytes_total",
+			Help: "Size of all data fetched to execute a query in bytes.",
+		}, []string{"user"})
+
 		h.activeUsers = util.NewActiveUsersCleanupWithDefaultValues(func(user string) {
 			h.querySeconds.DeleteLabelValues(user)
 			h.querySeries.DeleteLabelValues(user)
 			h.queryBytes.DeleteLabelValues(user)
+			h.queryDataBytes.DeleteLabelValues(user)
 		})
 		// If cleaner stops or fail, we will simply not clean the metrics for inactive users.
 		_ = h.activeUsers.StartAsync(context.Background())
@@ -186,11 +193,13 @@ func (f *Handler) reportQueryStats(r *http.Request, queryString url.Values, quer
 	wallTime := stats.LoadWallTime()
 	numSeries := stats.LoadFetchedSeries()
 	numBytes := stats.LoadFetchedChunkBytes()
+	numDataBytes := stats.LoadFetchedDataBytes()
 
 	// Track stats.
 	f.querySeconds.WithLabelValues(userID).Add(wallTime.Seconds())
 	f.querySeries.WithLabelValues(userID).Add(float64(numSeries))
 	f.queryBytes.WithLabelValues(userID).Add(float64(numBytes))
+	f.queryDataBytes.WithLabelValues(userID).Add(float64(numDataBytes))
 	f.activeUsers.UpdateUserTimestamp(userID, time.Now())
 
 	// Log stats.
@@ -203,6 +212,7 @@ func (f *Handler) reportQueryStats(r *http.Request, queryString url.Values, quer
 		"query_wall_time_seconds", wallTime.Seconds(),
 		"fetched_series_count", numSeries,
 		"fetched_chunks_bytes", numBytes,
+		"fetched_data_bytes", numDataBytes,
 	}, formatQueryString(queryString)...)
 
 	level.Info(util_log.WithContext(r.Context(), f.log)).Log(logMessage...)

pkg/ingester/client/custom.go

@@ -27,3 +27,24 @@ func (m *QueryStreamResponse) ChunksSize() int {
 	}
 	return size
 }
+
+// ChunksSize returns the size of all chunks in the response.
+func (m *QueryStreamResponse) LabelsSize() int {
+	if len(m.Timeseries) == 0 && len(m.Chunkseries) == 0 {
+		return 0
+	}
+
+	size := 0
+	for _, entry := range m.Chunkseries {
+		for _, label := range entry.Labels {
+			size += label.Size()
+		}
+	}
+
+	for _, entry := range m.Timeseries {
+		for _, label := range entry.Labels {
+			size += label.Size()
+		}
+	}
+	return size
+}

pkg/querier/blocks_store_queryable.go

@@ -652,12 +652,16 @@ func (q *blocksStoreQuerier) fetchSeriesFromStores(
 						}
 					}
 					chunksSize := countChunkBytes(s)
+					labelsSize := countLabelBytes(s)
 					if chunkBytesLimitErr := queryLimiter.AddChunkBytes(chunksSize); chunkBytesLimitErr != nil {
 						return validation.LimitError(chunkBytesLimitErr.Error())
 					}
 					if chunkLimitErr := queryLimiter.AddChunks(len(s.Chunks)); chunkLimitErr != nil {
 						return validation.LimitError(chunkLimitErr.Error())
 					}
+					if dataBytesLimitErr := queryLimiter.AddDataBytes(labelsSize + chunksSize); dataBytesLimitErr != nil {
+						return validation.LimitError(dataBytesLimitErr.Error())
+					}
 				}
 
 				if w := resp.GetWarning(); w != "" {
@@ -681,14 +685,18 @@ func (q *blocksStoreQuerier) fetchSeriesFromStores(
 
 			numSeries := len(mySeries)
 			chunkBytes := countChunkBytes(mySeries...)
+			labelBytes := countLabelBytes(mySeries...)
+			dataBytes := labelBytes + chunkBytes
 
 			reqStats.AddFetchedSeries(uint64(numSeries))
 			reqStats.AddFetchedChunkBytes(uint64(chunkBytes))
+			reqStats.AddFetchedDataBytes(uint64(dataBytes))
 
 			level.Debug(spanLog).Log("msg", "received series from store-gateway",
 				"instance", c.RemoteAddress(),
 				"fetched series", numSeries,
 				"fetched chunk bytes", chunkBytes,
+				"fetched data bytes", dataBytes,
 				"requested blocks", strings.Join(convertULIDsToString(blockIDs), " "),
 				"queried blocks", strings.Join(convertULIDsToString(myQueriedBlocks), " "))
 
@@ -993,6 +1001,17 @@ func countChunkBytes(series ...*storepb.Series) (count int) {
 	return count
 }
 
+// countChunkBytes returns the size of the chunks making up the provided series in bytes
+func countLabelBytes(series ...*storepb.Series) (count int) {
+	for _, s := range series {
+		for _, l := range s.Labels {
+			count += l.Size()
+		}
+	}
+
+	return count
+}
+
 // only retry connection issues
 func isRetryableError(err error) bool {
 	switch status.Code(err) {