Support peakSamples in query frontend response (#6188)

Author: SungJin1212

CHANGELOG.md

@@ -3,6 +3,7 @@
 ## master / unreleased
 
 * [FEATURE] Ruler: Minimize chances of missed rule group evaluations that can occur due to OOM kills, bad underlying nodes, or due to an unhealthy ruler that appears in the ring as healthy. This feature is enabled via `-ruler.enable-ha-evaluation` flag. #6129
+* [ENHANCEMENT] Query Frontend: Add peakSample in query stats response. #6188
 * [ENHANCEMENT] Ruler: Add new ruler metric `cortex_ruler_rule_groups_in_store` that is the total rule groups per tenant in store, which can be used to compare with `cortex_prometheus_rule_group_rules` to count the number of rule groups that are not loaded by a ruler. #5869
 * [ENHANCEMENT] Ingester/Ring: New `READONLY` status on ring to be used by Ingester. New ingester API to change mode of ingester #6163
 * [ENHANCEMENT] Ruler: Add query statistics metrics when --ruler.query-stats-enabled=true. #6173

pkg/querier/tripperware/instantquery/instant_query_test.go

@@ -174,10 +174,10 @@ func TestResponse(t *testing.T) {
 			body: `{"status":"success","data":{"resultType":"string","result":[1,"foo"]}}`,
 		},
 		{
-			body: `{"status":"success","data":{"resultType":"string","result":[1,"foo"],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1536673680,5],[1536673780,5]]}}}}`,
+			body: `{"status":"success","data":{"resultType":"string","result":[1,"foo"],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1536673680,5],[1536673780,5]],"peakSamples":10}}}}`,
 		},
 		{
-			body: `{"status":"success","data":{"resultType":"matrix","result":[{"metric":{"foo":"bar"},"values":[[1,"137"],[2,"137"]]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1536673680,5],[1536673780,5]]}}}}`,
+			body: `{"status":"success","data":{"resultType":"matrix","result":[{"metric":{"foo":"bar"},"values":[[1,"137"],[2,"137"]]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1536673680,5],[1536673780,5]],"peakSamples":10}}}}`,
 		},
 		{
 			body: `{"status":"success","data":{"resultType":"matrix","result":[{"metric":{"foo":"bar"},"values":[[1,"137"],[2,"137"]]}]}}`,
@@ -242,8 +242,8 @@ func TestMergeResponse(t *testing.T) {
 		{
 			name:         "empty response with stats",
 			req:          defaultReq,
-			resps:        []string{`{"status":"success","data":{"resultType":"vector","result":[],"stats":{"samples":{"totalQueryableSamples":0,"totalQueryableSamplesPerStep":[]}}}}`},
-			expectedResp: `{"status":"success","data":{"resultType":"vector","result":[],"stats":{"samples":{"totalQueryableSamples":0,"totalQueryableSamplesPerStep":[]}}}}`,
+			resps:        []string{`{"status":"success","data":{"resultType":"vector","result":[],"stats":{"samples":{"totalQueryableSamples":0,"totalQueryableSamplesPerStep":[],"peakSamples":0}}}}`},
+			expectedResp: `{"status":"success","data":{"resultType":"vector","result":[],"stats":{"samples":{"totalQueryableSamples":0,"totalQueryableSamplesPerStep":[],"peakSamples":0}}}}`,
 		},
 		{
 			name:         "single response",
@@ -254,8 +254,8 @@ func TestMergeResponse(t *testing.T) {
 		{
 			name:         "single response with stats",
 			req:          defaultReq,
-			resps:        []string{`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]]}}}}`},
-			expectedResp: `{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]]}}}}`,
+			resps:        []string{`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]],"peakSamples":10}}}}`},
+			expectedResp: `{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]],"peakSamples":10}}}}`,
 		},
 		{
 			name: "duplicated response",
@@ -276,10 +276,10 @@ func TestMergeResponse(t *testing.T) {
 			name: "duplicated response with stats",
 			req:  defaultReq,
 			resps: []string{
-				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]]}}}}`,
-				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]]}}}}`,
+				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]],"peakSamples":10}}}}`,
+				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]],"peakSamples":10}}}}`,
 			},
-			expectedResp: `{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":20,"totalQueryableSamplesPerStep":[[1,20]]}}}}`,
+			expectedResp: `{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":20,"totalQueryableSamplesPerStep":[[1,20]],"peakSamples":10}}}}`,
 		},
 		{
 			name: "merge two responses",
@@ -366,10 +366,19 @@ func TestMergeResponse(t *testing.T) {
 			name: "merge two responses with stats",
 			req:  defaultReq,
 			resps: []string{
-				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up","job":"foo"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]]}}}}`,
-				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up","job":"bar"},"value":[2,"2"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]]}}}}`,
+				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up","job":"foo"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]],"peakSamples":10}}}}`,
+				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up","job":"bar"},"value":[2,"2"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]],"peakSamples":10}}}}`,
 			},
-			expectedResp: `{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up","job":"bar"},"value":[2,"2"]},{"metric":{"__name__":"up","job":"foo"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":20,"totalQueryableSamplesPerStep":[[1,20]]}}}}`,
+			expectedResp: `{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up","job":"bar"},"value":[2,"2"]},{"metric":{"__name__":"up","job":"foo"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":20,"totalQueryableSamplesPerStep":[[1,20]],"peakSamples":10}}}}`,
+		},
+		{
+			name: "merge two responses with stats, the peak samples should be larger one among the responses",
+			req:  defaultReq,
+			resps: []string{
+				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up","job":"foo"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]],"peakSamples":10}}}}`,
+				`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up","job":"bar"},"value":[2,"2"]}],"stats":{"samples":{"totalQueryableSamples":10,"totalQueryableSamplesPerStep":[[1,10]],"peakSamples":15}}}}`,
+			},
+			expectedResp: `{"status":"success","data":{"resultType":"vector","result":[{"metric":{"__name__":"up","job":"bar"},"value":[2,"2"]},{"metric":{"__name__":"up","job":"foo"},"value":[1,"1"]}],"stats":{"samples":{"totalQueryableSamples":20,"totalQueryableSamplesPerStep":[[1,20]],"peakSamples":15}}}}`,
 		},
 		{
 			name: "responses don't contain vector, should return an error",

pkg/querier/tripperware/merge.go

@@ -211,6 +211,7 @@ func vectorMerge(ctx context.Context, req Request, resps []*PrometheusResponse)
 func statsMerge(shouldSumStats bool, resps []*PrometheusResponse) *PrometheusResponseStats {
 	output := map[int64]*PrometheusResponseQueryableSamplesStatsPerStep{}
 	hasStats := false
+	var peakSamples int64
 	for _, resp := range resps {
 		if resp.Data.Stats == nil {
 			continue
@@ -232,6 +233,7 @@ func statsMerge(shouldSumStats bool, resps []*PrometheusResponse) *PrometheusRes
 				output[s.GetTimestampMs()] = s
 			}
 		}
+		peakSamples = max(peakSamples, resp.Data.Stats.Samples.PeakSamples)
 	}
 
 	if !hasStats {
@@ -249,6 +251,7 @@ func statsMerge(shouldSumStats bool, resps []*PrometheusResponse) *PrometheusRes
 		result.Samples.TotalQueryableSamplesPerStep = append(result.Samples.TotalQueryableSamplesPerStep, output[key])
 		result.Samples.TotalQueryableSamples += output[key].Value
 	}
+	result.Samples.PeakSamples = peakSamples
 
 	return result
 }