support out of order samples ingestion feature

Signed-off-by: Ben Ye <[email protected]>

Author: yeya24

CHANGELOG.md

@@ -27,6 +27,7 @@
 * [FEATURE] Querier/Ruler: Support the new thanos promql engine. This is an experimental feature and might change in the future. #5093
 * [FEATURE] Added zstd as an option for grpc compression #5092
 * [FEATURE] Ring: Add new kv store option `dynamodb`. #5026
+* [FEATURE] Enable experimental out-of-order samples support. Added 2 new configs `ingester.out_of_order_time_window` and `blocks-storage.tsdb.out_of_order_cap_max`. #4964
 * [BUGFIX] Updated `golang.org/x/net` dependency to fix CVE-2022-27664. #5008
 * [BUGFIX] Fix panic when otel and xray tracing is enabled. #5044
 * [BUGFIX] Fixed no compact block got grouped in shuffle sharding grouper. #5055

docs/blocks-storage/querier.md

@@ -896,4 +896,9 @@ blocks_storage:
     # down.
     # CLI flag: -blocks-storage.tsdb.memory-snapshot-on-shutdown
     [memory_snapshot_on_shutdown: <boolean> | default = false]
+
+    # [EXPERIMENTAL] Configures the maximum number of samples per chunk that can
+    # be out-of-order.
+    # CLI flag: -blocks-storage.tsdb.out-of-order-cap-max
+    [out_of_order_cap_max: <int> | default = 32]
 ```

docs/blocks-storage/store-gateway.md

@@ -966,4 +966,9 @@ blocks_storage:
     # down.
     # CLI flag: -blocks-storage.tsdb.memory-snapshot-on-shutdown
     [memory_snapshot_on_shutdown: <boolean> | default = false]
+
+    # [EXPERIMENTAL] Configures the maximum number of samples per chunk that can
+    # be out-of-order.
+    # CLI flag: -blocks-storage.tsdb.out-of-order-cap-max
+    [out_of_order_cap_max: <int> | default = 32]
 ```

docs/configuration/config-file-reference.md

@@ -2772,6 +2772,11 @@ The `limits_config` configures default and per-tenant limits imposed by Cortex s
 # CLI flag: -ingester.max-global-metadata-per-metric
 [max_global_metadata_per_metric: <int> | default = 0]
 
+# [Experimental] Configures the allowed time window for ingestion of
+# out-of-order samples. Disabled (0s) by default.
+# CLI flag: -ingester.out-of-order-time-window
+[out_of_order_time_window: <duration> | default = 0s]
+
 # Maximum number of chunks that can be fetched in a single query from ingesters
 # and long-term storage. This limit is enforced in the querier, ruler and
 # store-gateway. 0 to disable.
@@ -2784,8 +2789,8 @@ 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]
 
-# 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.
+# Deprecated (use 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, ruler and store-gateway. 0 to disable.
 # CLI flag: -querier.max-fetched-chunk-bytes-per-query
 [max_fetched_chunk_bytes_per_query: <int> | default = 0]
@@ -3835,6 +3840,11 @@ tsdb:
   # down.
   # CLI flag: -blocks-storage.tsdb.memory-snapshot-on-shutdown
   [memory_snapshot_on_shutdown: <boolean> | default = false]
+
+  # [EXPERIMENTAL] Configures the maximum number of samples per chunk that can
+  # be out-of-order.
+  # CLI flag: -blocks-storage.tsdb.out-of-order-cap-max
+  [out_of_order_cap_max: <int> | default = 32]
 ```
 
 ### `compactor_config`

docs/configuration/v1-guarantees.md

@@ -100,4 +100,8 @@ Currently experimental features are:
   - `-frontend.query-vertical-shard-size` (int) CLI flag
   - `query_vertical_shard_size` (int) field in runtime config file
 - Snapshotting of in-memory TSDB on disk during shutdown
-  - `-blocks-storage.tsdb.memory-snapshot-on-shutdown` (boolean) CLI flag
+  - `-blocks-storage.tsdb.memory-snapshot-on-shutdown` (boolean) CLI flag
+- Out of order samples support
+  - `-blocks-storage.tsdb.out-of-order-cap-max` (int) CLI flag
+  - `-ingester.out-of-order-time-window` (duration) CLI flag
+  - `out_of_order_time_window` (duration) field in runtime config file

pkg/distributor/distributor_test.go

@@ -3032,7 +3032,6 @@ func TestDistributorValidation(t *testing.T) {
 			}},
 			err: httpgrpc.Errorf(http.StatusBadRequest, `timestamp too old: %d metric: "testmetric"`, past),
 		},
-
 		// Test validation fails for samples from the future.
 		{
 			labels: []labels.Labels{{{Name: labels.MetricName, Value: "testmetric"}, {Name: "foo", Value: "bar"}}},

pkg/ingester/ingester.go

@@ -841,11 +841,13 @@ func (i *Ingester) updateUserTSDBConfigs() {
 				ExemplarsConfig: &config.ExemplarsConfig{
 					MaxExemplars: i.getMaxExemplars(userID),
 				},
+				TSDBConfig: &config.TSDBConfig{
+					OutOfOrderTimeWindow: time.Duration(i.limits.OutOfOrderTimeWindow(userID)).Milliseconds(),
+				},
 			},
 		}
 
-		// This method currently updates the MaxExemplars and OutOfOrderTimeWindow. Invoking this method
-		// with a 0 value of OutOfOrderTimeWindow simply updates Max Exemplars.
+		// This method currently updates the MaxExemplars and OutOfOrderTimeWindow.
 		err := userDB.db.ApplyConfig(cfg)
 		if err != nil {
 			level.Error(logutil.WithUserID(userID, i.logger)).Log("msg", "failed to update user tsdb configuration.")
@@ -994,6 +996,7 @@ func (i *Ingester) Push(ctx context.Context, req *cortexpb.WriteRequest) (*corte
 		startAppend               = time.Now()
 		sampleOutOfBoundsCount    = 0
 		sampleOutOfOrderCount     = 0
+		sampleTooOldCount         = 0
 		newValueForTimestampCount = 0
 		perUserSeriesLimitCount   = 0
 		perMetricSeriesLimitCount = 0
@@ -1062,6 +1065,11 @@ func (i *Ingester) Push(ctx context.Context, req *cortexpb.WriteRequest) (*corte
 				updateFirstPartial(func() error { return wrappedTSDBIngestErr(err, model.Time(s.TimestampMs), ts.Labels) })
 				continue
 
+			case storage.ErrTooOldSample:
+				sampleTooOldCount++
+				updateFirstPartial(func() error { return wrappedTSDBIngestErr(err, model.Time(s.TimestampMs), ts.Labels) })
+				continue
+
 			case errMaxSeriesPerUserLimitExceeded:
 				perUserSeriesLimitCount++
 				updateFirstPartial(func() error { return makeLimitError(perUserSeriesLimit, i.limiter.FormatError(userID, cause)) })
@@ -1152,6 +1160,9 @@ func (i *Ingester) Push(ctx context.Context, req *cortexpb.WriteRequest) (*corte
 	if sampleOutOfOrderCount > 0 {
 		validation.DiscardedSamples.WithLabelValues(sampleOutOfOrder, userID).Add(float64(sampleOutOfOrderCount))
 	}
+	if sampleTooOldCount > 0 {
+		validation.DiscardedSamples.WithLabelValues(sampleTooOld, userID).Add(float64(sampleTooOldCount))
+	}
 	if newValueForTimestampCount > 0 {
 		validation.DiscardedSamples.WithLabelValues(newValueForTimestamp, userID).Add(float64(newValueForTimestampCount))
 	}
@@ -1944,6 +1955,8 @@ func (i *Ingester) createTSDB(userID string) (*userTSDB, error) {
 		MaxExemplars:                   maxExemplarsForUser,
 		HeadChunksWriteQueueSize:       i.cfg.BlocksStorageConfig.TSDB.HeadChunksWriteQueueSize,
 		EnableMemorySnapshotOnShutdown: i.cfg.BlocksStorageConfig.TSDB.MemorySnapshotOnShutdown,
+		OutOfOrderTimeWindow:           time.Duration(i.limits.OutOfOrderTimeWindow(userID)).Milliseconds(),
+		OutOfOrderCapMax:               i.cfg.BlocksStorageConfig.TSDB.OutOfOrderCapMax,
 	}, nil)
 	if err != nil {
 		return nil, errors.Wrapf(err, "failed to open TSDB: %s", udir)

pkg/ingester/ingester_test.go

@@ -341,6 +341,7 @@ func TestIngester_Push(t *testing.T) {
 		additionalMetrics         []string
 		disableActiveSeries       bool
 		maxExemplars              int
+		oooTimeWindow             time.Duration
 	}{
 		"should succeed on valid series and metadata": {
 			reqs: []*cortexpb.WriteRequest{
@@ -553,7 +554,7 @@ func TestIngester_Push(t *testing.T) {
 				cortex_ingester_memory_series_removed_total{user="test"} 0
 			`,
 		},
-		"should soft fail on sample out of order": {
+		"ooo disabled, should soft fail on sample out of order": {
 			reqs: []*cortexpb.WriteRequest{
 				cortexpb.ToWriteRequest(
 					[]labels.Labels{metricLabels},
@@ -597,7 +598,7 @@ func TestIngester_Push(t *testing.T) {
 				cortex_ingester_active_series{user="test"} 1
 			`,
 		},
-		"should soft fail on sample out of bound": {
+		"ooo disabled, should soft fail on sample out of bound": {
 			reqs: []*cortexpb.WriteRequest{
 				cortexpb.ToWriteRequest(
 					[]labels.Labels{metricLabels},
@@ -641,6 +642,92 @@ func TestIngester_Push(t *testing.T) {
 				cortex_ingester_active_series{user="test"} 1
 			`,
 		},
+		"ooo enabled, should soft fail on sample too old": {
+			reqs: []*cortexpb.WriteRequest{
+				cortexpb.ToWriteRequest(
+					[]labels.Labels{metricLabels},
+					[]cortexpb.Sample{{Value: 2, TimestampMs: 1575043969}},
+					nil,
+					cortexpb.API),
+				cortexpb.ToWriteRequest(
+					[]labels.Labels{metricLabels},
+					[]cortexpb.Sample{{Value: 1, TimestampMs: 1575043969 - (600 * 1000)}},
+					nil,
+					cortexpb.API),
+			},
+			oooTimeWindow: 5 * time.Minute,
+			expectedErr:   httpgrpc.Errorf(http.StatusBadRequest, wrapWithUser(wrappedTSDBIngestErr(storage.ErrTooOldSample, model.Time(1575043969-(600*1000)), cortexpb.FromLabelsToLabelAdapters(metricLabels)), userID).Error()),
+			expectedIngested: []cortexpb.TimeSeries{
+				{Labels: metricLabelAdapters, Samples: []cortexpb.Sample{{Value: 2, TimestampMs: 1575043969}}},
+			},
+			expectedMetrics: `
+				# HELP cortex_ingester_ingested_samples_total The total number of samples ingested.
+				# TYPE cortex_ingester_ingested_samples_total counter
+				cortex_ingester_ingested_samples_total 1
+				# HELP cortex_ingester_ingested_samples_failures_total The total number of samples that errored on ingestion.
+				# TYPE cortex_ingester_ingested_samples_failures_total counter
+				cortex_ingester_ingested_samples_failures_total 1
+				# HELP cortex_ingester_memory_users The current number of users in memory.
+				# TYPE cortex_ingester_memory_users gauge
+				cortex_ingester_memory_users 1
+				# HELP cortex_ingester_memory_series The current number of series in memory.
+				# TYPE cortex_ingester_memory_series gauge
+				cortex_ingester_memory_series 1
+				# HELP cortex_ingester_memory_series_created_total The total number of series that were created per user.
+				# TYPE cortex_ingester_memory_series_created_total counter
+				cortex_ingester_memory_series_created_total{user="test"} 1
+				# HELP cortex_ingester_memory_series_removed_total The total number of series that were removed per user.
+				# TYPE cortex_ingester_memory_series_removed_total counter
+				cortex_ingester_memory_series_removed_total{user="test"} 0
+				# HELP cortex_discarded_samples_total The total number of samples that were discarded.
+				# TYPE cortex_discarded_samples_total counter
+				cortex_discarded_samples_total{reason="sample-too-old",user="test"} 1
+				# HELP cortex_ingester_active_series Number of currently active series per user.
+				# TYPE cortex_ingester_active_series gauge
+				cortex_ingester_active_series{user="test"} 1
+			`,
+		},
+		"ooo enabled, should succeed": {
+			reqs: []*cortexpb.WriteRequest{
+				cortexpb.ToWriteRequest(
+					[]labels.Labels{metricLabels},
+					[]cortexpb.Sample{{Value: 2, TimestampMs: 1575043969}},
+					nil,
+					cortexpb.API),
+				cortexpb.ToWriteRequest(
+					[]labels.Labels{metricLabels},
+					[]cortexpb.Sample{{Value: 1, TimestampMs: 1575043969 - (60 * 1000)}},
+					nil,
+					cortexpb.API),
+			},
+			oooTimeWindow: 5 * time.Minute,
+			expectedIngested: []cortexpb.TimeSeries{
+				{Labels: metricLabelAdapters, Samples: []cortexpb.Sample{{Value: 1, TimestampMs: 1575043969 - (60 * 1000)}, {Value: 2, TimestampMs: 1575043969}}},
+			},
+			expectedMetrics: `
+				# HELP cortex_ingester_ingested_samples_total The total number of samples ingested.
+				# TYPE cortex_ingester_ingested_samples_total counter
+				cortex_ingester_ingested_samples_total 2
+				# HELP cortex_ingester_ingested_samples_failures_total The total number of samples that errored on ingestion.
+				# TYPE cortex_ingester_ingested_samples_failures_total counter
+				cortex_ingester_ingested_samples_failures_total 0
+				# HELP cortex_ingester_memory_users The current number of users in memory.
+				# TYPE cortex_ingester_memory_users gauge
+				cortex_ingester_memory_users 1
+				# HELP cortex_ingester_memory_series The current number of series in memory.
+				# TYPE cortex_ingester_memory_series gauge
+				cortex_ingester_memory_series 1
+				# HELP cortex_ingester_memory_series_created_total The total number of series that were created per user.
+				# TYPE cortex_ingester_memory_series_created_total counter
+				cortex_ingester_memory_series_created_total{user="test"} 1
+				# HELP cortex_ingester_memory_series_removed_total The total number of series that were removed per user.
+				# TYPE cortex_ingester_memory_series_removed_total counter
+				cortex_ingester_memory_series_removed_total{user="test"} 0
+				# HELP cortex_ingester_active_series Number of currently active series per user.
+				# TYPE cortex_ingester_active_series gauge
+				cortex_ingester_active_series{user="test"} 1
+			`,
+		},
 		"should soft fail on two different sample values at the same timestamp": {
 			reqs: []*cortexpb.WriteRequest{
 				cortexpb.ToWriteRequest(
@@ -777,6 +864,7 @@ func TestIngester_Push(t *testing.T) {
 
 			limits := defaultLimitsTestConfig()
 			limits.MaxExemplars = testData.maxExemplars
+			limits.OutOfOrderTimeWindow = model.Duration(testData.oooTimeWindow)
 			i, err := prepareIngesterWithBlocksStorageAndLimits(t, cfg, limits, "", registry)
 			require.NoError(t, err)
 			require.NoError(t, services.StartAndAwaitRunning(context.Background(), i))

pkg/ingester/series.go

@@ -4,4 +4,5 @@ const (
 	sampleOutOfOrder     = "sample-out-of-order"
 	newValueForTimestamp = "new-value-for-timestamp"
 	sampleOutOfBounds    = "sample-out-of-bounds"
+	sampleTooOld         = "sample-too-old"
 )

pkg/storage/tsdb/config.go

@@ -8,6 +8,7 @@ import (
 
 	"github.com/alecthomas/units"
 	"github.com/pkg/errors"
+	"github.com/prometheus/prometheus/tsdb"
 	"github.com/prometheus/prometheus/tsdb/chunks"
 	"github.com/prometheus/prometheus/tsdb/wlog"
 	"github.com/thanos-io/thanos/pkg/store"
@@ -45,6 +46,7 @@ var (
 	errInvalidCompactionConcurrency = errors.New("invalid TSDB compaction concurrency")
 	errInvalidWALSegmentSizeBytes   = errors.New("invalid TSDB WAL segment size bytes")
 	errInvalidStripeSize            = errors.New("invalid TSDB stripe size")
+	errInvalidOutOfOrderCapMax      = errors.New("invalid TSDB OOO chunks capacity (in samples)")
 	errEmptyBlockranges             = errors.New("empty block ranges for TSDB")
 )
 
@@ -145,11 +147,14 @@ type TSDBConfig struct {
 	// How often to check for idle TSDBs for closing. DefaultCloseIdleTSDBInterval is not suitable for testing, so tests can override.
 	CloseIdleTSDBInterval time.Duration `yaml:"-"`
 
-	// Positive value enables experiemental support for exemplars. 0 or less to disable.
+	// Positive value enables experimental support for exemplars. 0 or less to disable.
 	MaxExemplars int `yaml:"max_exemplars"`
 
 	// Enable snapshotting of in-memory TSDB data on disk when shutting down.
 	MemorySnapshotOnShutdown bool `yaml:"memory_snapshot_on_shutdown"`
+
+	// OutOfOrderCapMax is maximum capacity for OOO chunks (in samples).
+	OutOfOrderCapMax int64 `yaml:"out_of_order_cap_max"`
 }
 
 // RegisterFlags registers the TSDBConfig flags.
@@ -176,6 +181,7 @@ func (cfg *TSDBConfig) RegisterFlags(f *flag.FlagSet) {
 	f.IntVar(&cfg.HeadChunksWriteQueueSize, "blocks-storage.tsdb.head-chunks-write-queue-size", chunks.DefaultWriteQueueSize, "The size of the in-memory queue used before flushing chunks to the disk.")
 	f.IntVar(&cfg.MaxExemplars, "blocks-storage.tsdb.max-exemplars", 0, "Deprecated, use maxExemplars in limits instead. If the MaxExemplars value in limits is set to zero, cortex will fallback on this value. This setting enables support for exemplars in TSDB and sets the maximum number that will be stored. 0 or less means disabled.")
 	f.BoolVar(&cfg.MemorySnapshotOnShutdown, "blocks-storage.tsdb.memory-snapshot-on-shutdown", false, "True to enable snapshotting of in-memory TSDB data on disk when shutting down.")
+	f.Int64Var(&cfg.OutOfOrderCapMax, "blocks-storage.tsdb.out-of-order-cap-max", tsdb.DefaultOutOfOrderCapMax, "[EXPERIMENTAL] Configures the maximum number of samples per chunk that can be out-of-order.")
 }
 
 // Validate the config.
@@ -212,6 +218,10 @@ func (cfg *TSDBConfig) Validate() error {
 		return errInvalidWALSegmentSizeBytes
 	}
 
+	if cfg.OutOfOrderCapMax <= 0 {
+		return errInvalidOutOfOrderCapMax
+	}
+
 	return nil
 }