Query: add query metrics to calls going through the Store API

CHANGELOG.md

@@ -29,6 +29,7 @@ We use *breaking :warning:* to mark changes that are not backward compatible (re
 - [#5734](https://github.com/thanos-io/thanos/pull/5734) Store: Support disable block viewer UI.
 - [#5411](https://github.com/thanos-io/thanos/pull/5411) Tracing: Add OpenTelemetry Protocol exporter.
 - [#5779](https://github.com/thanos-io/thanos/pull/5779) Objstore: Support specifying S3 storage class.
+- [#5741](https://github.com/thanos-io/thanos/pull/5741) Query: add metrics on how much data is being selected by downstream Store APIs.
 
 ### Changed
 

cmd/thanos/query.go

@@ -25,6 +25,9 @@ import (
 	"github.com/prometheus/prometheus/discovery/targetgroup"
 	"github.com/prometheus/prometheus/model/labels"
 	"github.com/prometheus/prometheus/promql"
+	"github.com/thanos-io/thanos/pkg/store/telemetry"
+	"google.golang.org/grpc"
+
 	v1 "github.com/prometheus/prometheus/web/api/v1"
 	"github.com/thanos-community/promql-engine/engine"
 	apiv1 "github.com/thanos-io/thanos/pkg/api/query"
@@ -54,7 +57,6 @@ import (
 	"github.com/thanos-io/thanos/pkg/targets"
 	"github.com/thanos-io/thanos/pkg/tls"
 	"github.com/thanos-io/thanos/pkg/ui"
-	"google.golang.org/grpc"
 )
 
 const (
@@ -194,6 +196,10 @@ func registerQuery(app *extkingpin.App) {
 	alertQueryURL := cmd.Flag("alert.query-url", "The external Thanos Query URL that would be set in all alerts 'Source' field.").String()
 	grpcProxyStrategy := cmd.Flag("grpc.proxy-strategy", "Strategy to use when proxying Series requests to leaf nodes. Hidden and only used for testing, will be removed after lazy becomes the default.").Default(string(store.EagerRetrieval)).Hidden().Enum(string(store.EagerRetrieval), string(store.LazyRetrieval))
 
+	queryTelemetryDurationQuantiles := cmd.Flag("query.telemetry.request-duration-seconds-quantiles", "The quantiles for exporting metrics about the request duration quantiles.").Default("0.1", "0.25", "0.75", "1.25", "1.75", "2.5", "3", "5", "10").Float64List()
+	queryTelemetrySamplesQuantiles := cmd.Flag("query.telemetry.request-samples-quantiles", "The quantiles for exporting metrics about the samples count quantiles.").Default("100", "1000", "10000", "100000", "1000000").Int64List()
+	queryTelemetrySeriesQuantiles := cmd.Flag("query.telemetry.request-series-seconds-quantiles", "The quantiles for exporting metrics about the series count quantiles.").Default("10", "100", "1000", "10000", "100000").Int64List()
+
 	cmd.Setup(func(g *run.Group, logger log.Logger, reg *prometheus.Registry, tracer opentracing.Tracer, _ <-chan struct{}, _ bool) error {
 		selectorLset, err := parseFlagLabels(*selectorLabels)
 		if err != nil {
@@ -305,6 +311,9 @@ func registerQuery(app *extkingpin.App) {
 			*alertQueryURL,
 			*grpcProxyStrategy,
 			component.Query,
+			*queryTelemetryDurationQuantiles,
+			*queryTelemetrySamplesQuantiles,
+			*queryTelemetrySeriesQuantiles,
 			promqlEngineType(*promqlEngine),
 		)
 	})
@@ -377,6 +386,9 @@ func runQuery(
 	alertQueryURL string,
 	grpcProxyStrategy string,
 	comp component.Component,
+	queryTelemetryDurationQuantiles []float64,
+	queryTelemetrySamplesQuantiles []int64,
+	queryTelemetrySeriesQuantiles []int64,
 	promqlEngine promqlEngineType,
 ) error {
 	if alertQueryURL == "" {
@@ -680,6 +692,12 @@ func runQuery(
 				extprom.WrapRegistererWithPrefix("thanos_query_concurrent_", reg),
 				maxConcurrentQueries,
 			),
+			telemetry.NewSeriesStatsAggregator(
+				reg,
+				queryTelemetryDurationQuantiles,
+				queryTelemetrySamplesQuantiles,
+				queryTelemetrySeriesQuantiles,
+			),
 			reg,
 		)
 

docs/components/query.md

@@ -378,6 +378,15 @@ Flags:
                                  be able to query without deduplication using
                                  'dedup=false' parameter. Data includes time
                                  series, recording rules, and alerting rules.
+      --query.telemetry.request-duration-seconds-quantiles=0.1... ...
+                                 The quantiles for exporting metrics about the
+                                 request duration quantiles.
+      --query.telemetry.request-samples-quantiles=100... ...
+                                 The quantiles for exporting metrics about the
+                                 samples count quantiles.
+      --query.telemetry.request-series-seconds-quantiles=10... ...
+                                 The quantiles for exporting metrics about the
+                                 series count quantiles.
       --query.timeout=2m         Maximum time to process query by query node.
       --request.logging-config=<content>
                                  Alternative to 'request.logging-config-file'
@@ -460,3 +469,13 @@ Flags:
                                  of Prometheus.
 
 ```
+
+## Exported metrics
+
+Thanos Query also exports metrics about its own performance. You can find a list with these metrics below.
+
+**Disclaimer**: this list is incomplete. The remaining metrics will be added over time.
+
+| Name                                    | Type      | Labels                | Description                                                                                                       |
+|-----------------------------------------|-----------|-----------------------|-------------------------------------------------------------------------------------------------------------------|
+| thanos_store_api_query_duration_seconds | Histogram | samples_le, series_le | Duration of the Thanos Store API select phase for a query according to the amount of samples and series selected. |

pkg/api/query/grpc.go

@@ -94,6 +94,7 @@ func (g *GRPCAPI) Query(request *querypb.QueryRequest, server querypb.Query_Quer
 		request.EnableQueryPushdown,
 		false,
 		request.ShardInfo,
+		query.NoopSeriesStatsReporter,
 	)
 	qry, err := g.queryEngine.NewInstantQuery(queryable, &promql.QueryOpts{LookbackDelta: lookbackDelta}, request.Query, ts)
 	if err != nil {
@@ -168,6 +169,7 @@ func (g *GRPCAPI) QueryRange(request *querypb.QueryRangeRequest, srv querypb.Que
 		request.EnableQueryPushdown,
 		false,
 		request.ShardInfo,
+		query.NoopSeriesStatsReporter,
 	)
 
 	startTime := time.Unix(request.StartTimeSeconds, 0)

pkg/api/query/v1.go

@@ -41,9 +41,9 @@ import (
 	"github.com/prometheus/prometheus/promql"
 	"github.com/prometheus/prometheus/promql/parser"
 	"github.com/prometheus/prometheus/storage"
-	v1 "github.com/prometheus/prometheus/web/api/v1"
-
 	"github.com/prometheus/prometheus/util/stats"
+	v1 "github.com/prometheus/prometheus/web/api/v1"
+	"github.com/thanos-io/thanos/pkg/store/telemetry"
 
 	"github.com/thanos-io/thanos/pkg/api"
 	"github.com/thanos-io/thanos/pkg/exemplars"
@@ -107,6 +107,13 @@ type QueryAPI struct {
 	defaultMetadataTimeRange               time.Duration
 
 	queryRangeHist prometheus.Histogram
+
+	seriesStatsAggregator seriesQueryPerformanceMetricsAggregator
+}
+
+type seriesQueryPerformanceMetricsAggregator interface {
+	Aggregate(seriesStats storepb.SeriesStatsCounter)
+	Observe(duration float64)
 }
 
 // NewQueryAPI returns an initialized QueryAPI type.
@@ -134,8 +141,12 @@ func NewQueryAPI(
 	defaultMetadataTimeRange time.Duration,
 	disableCORS bool,
 	gate gate.Gate,
+	statsAggregator seriesQueryPerformanceMetricsAggregator,
 	reg *prometheus.Registry,
 ) *QueryAPI {
+	if statsAggregator == nil {
+		statsAggregator = &telemetry.NoopSeriesStatsAggregator{}
+	}
 	return &QueryAPI{
 		baseAPI:                                api.NewBaseAPI(logger, disableCORS, flagsMap),
 		logger:                                 logger,
@@ -160,6 +171,7 @@ func NewQueryAPI(
 		defaultInstantQueryMaxSourceResolution: defaultInstantQueryMaxSourceResolution,
 		defaultMetadataTimeRange:               defaultMetadataTimeRange,
 		disableCORS:                            disableCORS,
+		seriesStatsAggregator:                  statsAggregator,
 
 		queryRangeHist: promauto.With(reg).NewHistogram(prometheus.HistogramOpts{
 			Name:    "thanos_query_range_requested_timespan_duration_seconds",
@@ -396,7 +408,24 @@ func (qapi *QueryAPI) query(r *http.Request) (interface{}, []error, *api.ApiErro
 	span, ctx := tracing.StartSpan(ctx, "promql_instant_query")
 	defer span.Finish()
 
-	qry, err := qapi.queryEngine.NewInstantQuery(qapi.queryableCreate(enableDedup, replicaLabels, storeDebugMatchers, maxSourceResolution, enablePartialResponse, qapi.enableQueryPushdown, false, shardInfo), &promql.QueryOpts{LookbackDelta: lookbackDelta}, r.FormValue("query"), ts)
+	var seriesStats []storepb.SeriesStatsCounter
+	qry, err := qapi.queryEngine.NewInstantQuery(
+		qapi.queryableCreate(
+			enableDedup,
+			replicaLabels,
+			storeDebugMatchers,
+			maxSourceResolution,
+			enablePartialResponse,
+			qapi.enableQueryPushdown,
+			false,
+			shardInfo,
+			query.NewAggregateStatsReporter(&seriesStats),
+		),
+		&promql.QueryOpts{LookbackDelta: lookbackDelta},
+		r.FormValue("query"),
+		ts,
+	)
+
 	if err != nil {
 		return nil, nil, &api.ApiError{Typ: api.ErrorBadData, Err: err}, func() {}
 	}
@@ -409,6 +438,7 @@ func (qapi *QueryAPI) query(r *http.Request) (interface{}, []error, *api.ApiErro
 	}
 	defer qapi.gate.Done()
 
+	beforeRange := time.Now()
 	res := qry.Exec(ctx)
 	if res.Err != nil {
 		switch res.Err.(type) {
@@ -421,6 +451,10 @@ func (qapi *QueryAPI) query(r *http.Request) (interface{}, []error, *api.ApiErro
 		}
 		return nil, nil, &api.ApiError{Typ: api.ErrorExec, Err: res.Err}, qry.Close
 	}
+	for i := range seriesStats {
+		qapi.seriesStatsAggregator.Aggregate(seriesStats[i])
+	}
+	qapi.seriesStatsAggregator.Observe(time.Since(beforeRange).Seconds())
 
 	// Optional stats field in response if parameter "stats" is not empty.
 	var qs stats.QueryStats
@@ -525,8 +559,19 @@ func (qapi *QueryAPI) queryRange(r *http.Request) (interface{}, []error, *api.Ap
 	span, ctx := tracing.StartSpan(ctx, "promql_range_query")
 	defer span.Finish()
 
+	var seriesStats []storepb.SeriesStatsCounter
 	qry, err := qapi.queryEngine.NewRangeQuery(
-		qapi.queryableCreate(enableDedup, replicaLabels, storeDebugMatchers, maxSourceResolution, enablePartialResponse, qapi.enableQueryPushdown, false, shardInfo),
+		qapi.queryableCreate(
+			enableDedup,
+			replicaLabels,
+			storeDebugMatchers,
+			maxSourceResolution,
+			enablePartialResponse,
+			qapi.enableQueryPushdown,
+			false,
+			shardInfo,
+			query.NewAggregateStatsReporter(&seriesStats),
+		),
 		&promql.QueryOpts{LookbackDelta: lookbackDelta},
 		r.FormValue("query"),
 		start,
@@ -545,6 +590,7 @@ func (qapi *QueryAPI) queryRange(r *http.Request) (interface{}, []error, *api.Ap
 	}
 	defer qapi.gate.Done()
 
+	beforeRange := time.Now()
 	res := qry.Exec(ctx)
 	if res.Err != nil {
 		switch res.Err.(type) {
@@ -555,6 +601,10 @@ func (qapi *QueryAPI) queryRange(r *http.Request) (interface{}, []error, *api.Ap
 		}
 		return nil, nil, &api.ApiError{Typ: api.ErrorExec, Err: res.Err}, qry.Close
 	}
+	for i := range seriesStats {
+		qapi.seriesStatsAggregator.Aggregate(seriesStats[i])
+	}
+	qapi.seriesStatsAggregator.Observe(time.Since(beforeRange).Seconds())
 
 	// Optional stats field in response if parameter "stats" is not empty.
 	var qs stats.QueryStats
@@ -600,8 +650,17 @@ func (qapi *QueryAPI) labelValues(r *http.Request) (interface{}, []error, *api.A
 		matcherSets = append(matcherSets, matchers)
 	}
 
-	q, err := qapi.queryableCreate(true, nil, storeDebugMatchers, 0, enablePartialResponse, qapi.enableQueryPushdown, true, nil).
-		Querier(ctx, timestamp.FromTime(start), timestamp.FromTime(end))
+	q, err := qapi.queryableCreate(
+		true,
+		nil,
+		storeDebugMatchers,
+		0,
+		enablePartialResponse,
+		qapi.enableQueryPushdown,
+		true,
+		nil,
+		query.NoopSeriesStatsReporter,
+	).Querier(ctx, timestamp.FromTime(start), timestamp.FromTime(end))
 	if err != nil {
 		return nil, nil, &api.ApiError{Typ: api.ErrorExec, Err: err}, func() {}
 	}
@@ -687,8 +746,18 @@ func (qapi *QueryAPI) series(r *http.Request) (interface{}, []error, *api.ApiErr
 		return nil, nil, apiErr, func() {}
 	}
 
-	q, err := qapi.queryableCreate(enableDedup, replicaLabels, storeDebugMatchers, math.MaxInt64, enablePartialResponse, qapi.enableQueryPushdown, true, nil).
-		Querier(r.Context(), timestamp.FromTime(start), timestamp.FromTime(end))
+	q, err := qapi.queryableCreate(
+		enableDedup,
+		replicaLabels,
+		storeDebugMatchers,
+		math.MaxInt64,
+		enablePartialResponse,
+		qapi.enableQueryPushdown,
+		true,
+		nil,
+		query.NoopSeriesStatsReporter,
+	).Querier(r.Context(), timestamp.FromTime(start), timestamp.FromTime(end))
+
 	if err != nil {
 		return nil, nil, &api.ApiError{Typ: api.ErrorExec, Err: err}, func() {}
 	}
@@ -737,8 +806,17 @@ func (qapi *QueryAPI) labelNames(r *http.Request) (interface{}, []error, *api.Ap
 		matcherSets = append(matcherSets, matchers)
 	}
 
-	q, err := qapi.queryableCreate(true, nil, storeDebugMatchers, 0, enablePartialResponse, qapi.enableQueryPushdown, true, nil).
-		Querier(r.Context(), timestamp.FromTime(start), timestamp.FromTime(end))
+	q, err := qapi.queryableCreate(
+		true,
+		nil,
+		storeDebugMatchers,
+		0,
+		enablePartialResponse,
+		qapi.enableQueryPushdown,
+		true,
+		nil,
+		query.NoopSeriesStatsReporter,
+	).Querier(r.Context(), timestamp.FromTime(start), timestamp.FromTime(end))
 	if err != nil {
 		return nil, nil, &api.ApiError{Typ: api.ErrorExec, Err: err}, func() {}
 	}

pkg/api/query/v1_test.go

@@ -45,6 +45,7 @@ import (
 	promgate "github.com/prometheus/prometheus/util/gate"
 	"github.com/prometheus/prometheus/util/stats"
 	"github.com/thanos-io/thanos/pkg/compact"
+	"github.com/thanos-io/thanos/pkg/store/telemetry"
 
 	baseAPI "github.com/thanos-io/thanos/pkg/api"
 	"github.com/thanos-io/thanos/pkg/component"
@@ -198,6 +199,7 @@ func TestQueryEndpoints(t *testing.T) {
 		queryRangeHist: promauto.With(prometheus.NewRegistry()).NewHistogram(prometheus.HistogramOpts{
 			Name: "query_range_hist",
 		}),
+		seriesStatsAggregator: &telemetry.NoopSeriesStatsAggregator{},
 	}
 
 	start := time.Unix(0, 0)
@@ -737,6 +739,7 @@ func TestMetadataEndpoints(t *testing.T) {
 		queryRangeHist: promauto.With(prometheus.NewRegistry()).NewHistogram(prometheus.HistogramOpts{
 			Name: "query_range_hist",
 		}),
+		seriesStatsAggregator: &telemetry.NoopSeriesStatsAggregator{},
 	}
 	apiWithLabelLookback := &QueryAPI{
 		baseAPI: &baseAPI.BaseAPI{
@@ -750,6 +753,7 @@ func TestMetadataEndpoints(t *testing.T) {
 		queryRangeHist: promauto.With(prometheus.NewRegistry()).NewHistogram(prometheus.HistogramOpts{
 			Name: "query_range_hist",
 		}),
+		seriesStatsAggregator: &telemetry.NoopSeriesStatsAggregator{},
 	}
 
 	var tests = []endpointTestCase{