receive,ruler: Upgraded TSDB and used ChunkIterator in Series TSDB St…

…ore.

Signed-off-by: Bartlomiej Plotka <bwplotka@gmail.com>

CHANGELOG.md

@@ -43,6 +43,7 @@ We use *breaking* word for marking changes that are not backward compatible (rel
 - [#2902](https://github.com/thanos-io/thanos/pull/2902) ui: React: Separate dedupe and partial response checkboxes per panel.
 - [#2931](https://github.com/thanos-io/thanos/pull/2931) Query: Allow passing a `storeMatcher[]` to select matching stores when debugging the querier. See [documentation](https://thanos.io/components/query.md/#store-filtering)
 - [#2991](https://github.com/thanos-io/thanos/pull/2991) store: `operation` label value `getrange` changed to `get_range` for `thanos_store_bucket_cache_operation_requests_total` and `thanos_store_bucket_cache_operation_hits_total` to be consistent with bucket operation metrics.
+- [#2876](https://github.com/thanos-io/thanos/pull/2876) Receive,Ruler: Updated TSDB and switched to ChunkIterators instead of sample one, which avoids unnecessary decoding / encoding.
 
 ## [v0.14.0](https://github.com/thanos-io/thanos/releases/tag/v0.14.0) - 2020.07.10
 

cmd/thanos/query.go

@@ -151,7 +151,6 @@ func registerQuery(m map[string]setupFunc, app *kingpin.Application) {
 		}
 
 		flagsMap := getFlagsMap(cmd.Model().Flags)
-
 		return runQuery(
 			g,
 			logger,
@@ -193,6 +192,7 @@ func registerQuery(m map[string]setupFunc, app *kingpin.Application) {
 			time.Duration(*instantDefaultMaxSourceResolution),
 			*strictStores,
 			component.Query,
+			time.Duration(*defaultEvaluationInterval),
 		)
 	}
 }
@@ -240,6 +240,7 @@ func runQuery(
 	instantDefaultMaxSourceResolution time.Duration,
 	strictStores []string,
 	comp component.Component,
+	defaultEvaluationInterval time.Duration,
 ) error {
 	// TODO(bplotka in PR #513 review): Move arguments into struct.
 	duplicatedStores := promauto.With(reg).NewCounter(prometheus.CounterOpts{

go.mod

@@ -74,8 +74,8 @@ require (
 // so that we don't get errors about being incompatible with the Go proxies.
 // See https://github.com/thanos-io/thanos/issues/1415
 replace (
-	// Make sure Cortex is not forcing us to some other Prometheus version.
-	github.com/prometheus/prometheus => github.com/prometheus/prometheus v1.8.2-0.20200805082714-e0cf219f0de2
+	// Make sure Prometheus version is pinned as Prometheus semver does not include Go APIs.
+	github.com/prometheus/prometheus => github.com/prometheus/prometheus v1.8.2-0.20200807135816-2899773b0159
 	google.golang.org/grpc => google.golang.org/grpc v1.29.1
 	k8s.io/klog => k8s.io/klog v0.3.1
 	k8s.io/kube-openapi => k8s.io/kube-openapi v0.0.0-20190228160746-b3a7cee44a30

go.sum

@@ -857,7 +857,6 @@ github.com/prometheus/common v0.6.0/go.mod h1:eBmuwkDJBwy6iBfxCBob6t6dR6ENT/y+J+
 github.com/prometheus/common v0.7.0/go.mod h1:DjGbpBbp5NYNiECxcL/VnbXCCaQpKd3tt26CguLLsqA=
 github.com/prometheus/common v0.8.0/go.mod h1:PC/OgXc+UN7B4ALwvn1yzVZmVwvhXp5JsbBv6wSv6i0=
 github.com/prometheus/common v0.9.1/go.mod h1:yhUN8i9wzaXS3w1O07YhxHEBxD+W35wd8bs7vj7HSQ4=
-github.com/prometheus/common v0.10.0 h1:RyRA7RzGXQZiW+tGMr7sxa85G1z0yOpM1qq5c8lNawc=
 github.com/prometheus/common v0.10.0/go.mod h1:Tlit/dnDKsSWFlCLTWaA1cyBgKHSMdTB80sz/V91rCo=
 github.com/prometheus/common v0.11.1 h1:0ZISXCMRuCZcxF77aT1BXY5m74mX2vrGYl1dSwBI0Jo=
 github.com/prometheus/common v0.11.1/go.mod h1:U+gB1OBLb1lF3O42bTCL+FK18tX9Oar16Clt/msog/s=
@@ -876,8 +875,8 @@ github.com/prometheus/procfs v0.0.8/go.mod h1:7Qr8sr6344vo1JqZ6HhLceV9o3AJ1Ff+Gx
 github.com/prometheus/procfs v0.0.11/go.mod h1:lV6e/gmhEcM9IjHGsFOCxxuZ+z1YqCvr4OA4YeYWdaU=
 github.com/prometheus/procfs v0.1.3 h1:F0+tqvhOksq22sc6iCHF5WGlWjdwj92p0udFh1VFBS8=
 github.com/prometheus/procfs v0.1.3/go.mod h1:lV6e/gmhEcM9IjHGsFOCxxuZ+z1YqCvr4OA4YeYWdaU=
-github.com/prometheus/prometheus v1.8.2-0.20200805082714-e0cf219f0de2 h1:eQ88y1vfbXuclr6B04jYTmhc6ydXlBUSIaXCjEs0osk=
-github.com/prometheus/prometheus v1.8.2-0.20200805082714-e0cf219f0de2/go.mod h1:i1KZsZmyDTJRvnR7zE8z/u2v+tkpPjoiPpnWp6nwhr0=
+github.com/prometheus/prometheus v1.8.2-0.20200807135816-2899773b0159 h1:mT66e//l/+QugUat5A42YvIPxlsS6O/yr9UtjlDhPYw=
+github.com/prometheus/prometheus v1.8.2-0.20200807135816-2899773b0159/go.mod h1:zfAqy/MwhMFajB9E2n12/9gG2fvofIE9uKDtlZCDxqs=
 github.com/rafaeljusto/redigomock v0.0.0-20190202135759-257e089e14a1/go.mod h1:JaY6n2sDr+z2WTsXkOmNRUfDy6FN0L6Nk7x06ndm4tY=
 github.com/rcrowley/go-metrics v0.0.0-20181016184325-3113b8401b8a/go.mod h1:bCqnVzQkZxMG4s8nGwiZ5l3QUCyqpo9Y+/ZMZ9VjZe4=
 github.com/retailnext/hllpp v1.0.1-0.20180308014038-101a6d2f8b52/go.mod h1:RDpi1RftBQPUCDRw6SmxeaREsAaRKnOclghuzp/WRzc=

pkg/store/tsdb.go

@@ -13,7 +13,6 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/prometheus/pkg/labels"
 	"github.com/prometheus/prometheus/storage"
-	"github.com/prometheus/prometheus/tsdb/chunkenc"
 	"google.golang.org/grpc/codes"
 	"google.golang.org/grpc/status"
 
@@ -24,18 +23,19 @@ import (
 )
 
 type TSDBReader interface {
-	storage.Queryable
+	storage.ChunkQueryable
 	StartTime() (int64, error)
 }
 
 // TSDBStore implements the store API against a local TSDB instance.
 // It attaches the provided external labels to all results. It only responds with raw data
 // and does not support downsampling.
 type TSDBStore struct {
-	logger         log.Logger
-	db             TSDBReader
-	component      component.StoreAPI
-	externalLabels labels.Labels
+	logger           log.Logger
+	db               TSDBReader
+	component        component.StoreAPI
+	externalLabels   labels.Labels
+	maxBytesPerFrame int
 }
 
 // ReadWriteTSDBStore is a TSDBStore that can also be written to.
@@ -50,10 +50,11 @@ func NewTSDBStore(logger log.Logger, _ prometheus.Registerer, db TSDBReader, com
 		logger = log.NewNopLogger()
 	}
 	return &TSDBStore{
-		logger:         logger,
-		db:             db,
-		component:      component,
-		externalLabels: externalLabels,
+		logger:           logger,
+		db:               db,
+		component:        component,
+		externalLabels:   externalLabels,
+		maxBytesPerFrame: 1024 * 1024, // 1MB as recommended by gRPC.
 	}
 }
 
@@ -109,7 +110,7 @@ func (s *TSDBStore) Series(r *storepb.SeriesRequest, srv storepb.Store_SeriesSer
 		return status.Error(codes.InvalidArgument, err.Error())
 	}
 
-	q, err := s.db.Querier(context.Background(), r.MinTime, r.MaxTime)
+	q, err := s.db.ChunkQuerier(context.Background(), r.MinTime, r.MaxTime)
 	if err != nil {
 		return status.Error(codes.Internal, err.Error())
 	}
@@ -119,72 +120,67 @@ func (s *TSDBStore) Series(r *storepb.SeriesRequest, srv storepb.Store_SeriesSer
 		set        = q.Select(false, nil, matchers...)
 		respSeries storepb.Series
 	)
+
+	// Stream at most one series per frame; series may be split over multiple frames according to maxBytesInFrame.
 	for set.Next() {
 		series := set.At()
-
 		respSeries.Labels = s.translateAndExtendLabels(series.Labels(), s.externalLabels)
-
-		if !r.SkipChunks {
-			// TODO(fabxc): An improvement over this trivial approach would be to directly
-			// use the chunks provided by TSDB in the response.
-			c, err := s.encodeChunks(series.Iterator(), MaxSamplesPerChunk)
-			if err != nil {
-				return status.Errorf(codes.Internal, "encode chunk: %s", err)
+		respSeries.Chunks = respSeries.Chunks[:0]
+		if r.SkipChunks {
+			if err := srv.Send(storepb.NewSeriesResponse(&respSeries)); err != nil {
+				return status.Error(codes.Aborted, err.Error())
 			}
-
-			respSeries.Chunks = append(respSeries.Chunks[:0], c...)
+			continue
 		}
 
-		if err := srv.Send(storepb.NewSeriesResponse(&respSeries)); err != nil {
-			return status.Error(codes.Aborted, err.Error())
+		frameBytesLeft := s.maxBytesPerFrame
+		for _, lbl := range respSeries.Labels {
+			frameBytesLeft -= lbl.Size()
 		}
-	}
-	if err := set.Err(); err != nil {
-		return status.Error(codes.Internal, err.Error())
-	}
-	return nil
-}
-
-func (s *TSDBStore) encodeChunks(it chunkenc.Iterator, maxSamplesPerChunk int) (chks []storepb.AggrChunk, err error) {
-	var (
-		chkMint int64
-		chk     *chunkenc.XORChunk
-		app     chunkenc.Appender
-		isNext  = it.Next()
-	)
 
-	for isNext {
-		if chk == nil {
-			chk = chunkenc.NewXORChunk()
-			app, err = chk.Appender()
-			if err != nil {
-				return nil, err
+		chIter := series.Iterator()
+		isNext := chIter.Next()
+		for isNext {
+			chk := chIter.At()
+			if chk.Chunk == nil {
+				return status.Errorf(codes.Internal, "TSDBStore: found not populated chunk returned by SeriesSet at ref: %v", chk.Ref)
 			}
-			chkMint, _ = it.At()
-		}
 
-		app.Append(it.At())
-		chkMaxt, _ := it.At()
+			respSeries.Chunks = append(respSeries.Chunks, storepb.AggrChunk{
+				MinTime: chk.MinTime,
+				MaxTime: chk.MaxTime,
+				Raw: &storepb.Chunk{
+					Type: storepb.Chunk_Encoding(chk.Chunk.Encoding() - 1), // Proto chunk encoding is one off to TSDB one.
+					Data: chk.Chunk.Bytes(),
+				},
+			})
+			frameBytesLeft -= respSeries.Chunks[len(respSeries.Chunks)-1].Size()
+
+			// We are fine with minor inaccuracy of max bytes per frame. The inaccuracy will be max of full chunk size.
+			isNext = chIter.Next()
+			if frameBytesLeft > 0 && isNext {
+				continue
+			}
 
-		isNext = it.Next()
-		if isNext && chk.NumSamples() < maxSamplesPerChunk {
-			continue
+			if err := srv.Send(storepb.NewSeriesResponse(&respSeries)); err != nil {
+				return status.Error(codes.Aborted, err.Error())
+			}
+			respSeries.Chunks = respSeries.Chunks[:0]
+		}
+		if err := chIter.Err(); err != nil {
+			return status.Error(codes.Internal, errors.Wrap(err, "chunk iter").Error())
 		}
 
-		// Cut the chunk.
-		chks = append(chks, storepb.AggrChunk{
-			MinTime: chkMint,
-			MaxTime: chkMaxt,
-			Raw:     &storepb.Chunk{Type: storepb.Chunk_XOR, Data: chk.Bytes()},
-		})
-		chk = nil
 	}
-	if it.Err() != nil {
-		return nil, errors.Wrap(it.Err(), "read TSDB series")
+	if err := set.Err(); err != nil {
+		return status.Error(codes.Internal, err.Error())
+	}
+	for _, w := range set.Warnings() {
+		if err := srv.Send(storepb.NewWarnSeriesResponse(w)); err != nil {
+			return status.Error(codes.Aborted, err.Error())
+		}
 	}
-
-	return chks, nil
-
+	return nil
 }
 
 // translateAndExtendLabels transforms a metrics into a protobuf label set. It additionally