Allow overlapping blocks by default (#11331)

Signed-off-by: Ganesh Vernekar <ganeshvern@gmail.com>

cmd/prometheus/main.go

@@ -312,8 +312,10 @@ func main() {
 	serverOnlyFlag(a, "storage.tsdb.no-lockfile", "Do not create lockfile in data directory.").
 		Default("false").BoolVar(&cfg.tsdb.NoLockfile)
 
-	serverOnlyFlag(a, "storage.tsdb.allow-overlapping-blocks", "Allow overlapping blocks, which in turn enables vertical compaction and vertical query merge.").
+	// TODO: Remove in Prometheus 3.0.
-		Default("false").BoolVar(&cfg.tsdb.AllowOverlappingBlocks)
+	var b bool
+	serverOnlyFlag(a, "storage.tsdb.allow-overlapping-blocks", "[DEPRECATED] This flag has no effect. Overlapping blocks are enabled by default now.").
+		Default("true").Hidden().BoolVar(&b)
 
 	serverOnlyFlag(a, "storage.tsdb.wal-compression", "Compress the tsdb WAL.").
 		Hidden().Default("true").BoolVar(&cfg.tsdb.WALCompression)
@@ -1008,7 +1010,6 @@ func main() {
 					"NoLockfile", cfg.tsdb.NoLockfile,
 					"RetentionDuration", cfg.tsdb.RetentionDuration,
 					"WALSegmentSize", cfg.tsdb.WALSegmentSize,
-					"AllowOverlappingBlocks", cfg.tsdb.AllowOverlappingBlocks,
 					"WALCompression", cfg.tsdb.WALCompression,
 				)
 
@@ -1531,7 +1532,6 @@ type tsdbOptions struct {
 	RetentionDuration              model.Duration
 	MaxBytes                       units.Base2Bytes
 	NoLockfile                     bool
-	AllowOverlappingBlocks         bool
 	WALCompression                 bool
 	HeadChunksWriteQueueSize       int
 	StripeSize                     int
@@ -1550,8 +1550,7 @@ func (opts tsdbOptions) ToTSDBOptions() tsdb.Options {
 		RetentionDuration:              int64(time.Duration(opts.RetentionDuration) / time.Millisecond),
 		MaxBytes:                       int64(opts.MaxBytes),
 		NoLockfile:                     opts.NoLockfile,
-		AllowOverlappingCompaction:     opts.AllowOverlappingBlocks,
+		AllowOverlappingCompaction:     true,
-		AllowOverlappingQueries:        opts.AllowOverlappingBlocks,
 		WALCompression:                 opts.WALCompression,
 		HeadChunksWriteQueueSize:       opts.HeadChunksWriteQueueSize,
 		StripeSize:                     opts.StripeSize,

cmd/promtool/rules_test.go

@@ -117,8 +117,6 @@ func TestBackfillRuleIntegration(t *testing.T) {
 				}
 
 				opts := tsdb.DefaultOptions()
-				opts.AllowOverlappingQueries = true
-				opts.AllowOverlappingCompaction = true
 				db, err := tsdb.Open(tmpDir, nil, nil, opts, nil)
 				require.NoError(t, err)
 
@@ -246,8 +244,6 @@ func TestBackfillLabels(t *testing.T) {
 	}
 
 	opts := tsdb.DefaultOptions()
-	opts.AllowOverlappingQueries = true
-	opts.AllowOverlappingCompaction = true
 	db, err := tsdb.Open(tmpDir, nil, nil, opts, nil)
 	require.NoError(t, err)
 

docs/storage.md

@@ -155,7 +155,7 @@ Backfilling can be used via the Promtool command line. Promtool will write the b
 promtool tsdb create-blocks-from openmetrics <input file> [<output directory>]
 ```
 
-After the creation of the blocks, move it to the data directory of Prometheus. If there is an overlap with the existing blocks in Prometheus, the flag `--storage.tsdb.allow-overlapping-blocks` needs to be set. Note that any backfilled data is subject to the retention configured for your Prometheus server (by time or size).
+After the creation of the blocks, move it to the data directory of Prometheus. If there is an overlap with the existing blocks in Prometheus, the flag `--storage.tsdb.allow-overlapping-blocks` needs to be set for Prometheus versions v2.38 and below. Note that any backfilled data is subject to the retention configured for your Prometheus server (by time or size).
 
 #### Longer Block Durations
 
@@ -189,7 +189,7 @@ $ promtool tsdb create-blocks-from rules \
 
 The recording rule files provided should be a normal [Prometheus rules file](https://prometheus.io/docs/prometheus/latest/configuration/recording_rules/).
 
-The output of `promtool tsdb create-blocks-from rules` command is a directory that contains blocks with the historical rule data for all rules in the recording rule files. By default the output directory is `data/`. In order to make use of this new block data, the blocks must be moved to a running Prometheus instance data dir `storage.tsdb.path` that has the flag `--storage.tsdb.allow-overlapping-blocks` enabled. Once moved, the new blocks will merge with existing blocks when the next compaction runs.
+The output of `promtool tsdb create-blocks-from rules` command is a directory that contains blocks with the historical rule data for all rules in the recording rule files. By default, the output directory is `data/`. In order to make use of this new block data, the blocks must be moved to a running Prometheus instance data dir `storage.tsdb.path` (for Prometheus versions v2.38 and below, the flag `--storage.tsdb.allow-overlapping-blocks` must be enabled). Once moved, the new blocks will merge with existing blocks when the next compaction runs.
 
 ### Limitations
 

tsdb/db.go

@@ -76,8 +76,7 @@ func DefaultOptions() *Options {
 		MinBlockDuration:           DefaultBlockDuration,
 		MaxBlockDuration:           DefaultBlockDuration,
 		NoLockfile:                 false,
-		AllowOverlappingCompaction: false,
+		AllowOverlappingCompaction: true,
-		AllowOverlappingQueries:    false,
 		WALCompression:             false,
 		StripeSize:                 DefaultStripeSize,
 		HeadChunksWriteBufferSize:  chunks.DefaultWriteBufferSize,
@@ -114,18 +113,12 @@ type Options struct {
 	// NoLockfile disables creation and consideration of a lock file.
 	NoLockfile bool
 
-	// Querying on overlapping blocks are allowed if AllowOverlappingQueries is true.
-	// Since querying is a required operation for TSDB, if there are going to be
-	// overlapping blocks, then this should be set to true.
-	// NOTE: Do not use this directly in DB. Use it via DB.AllowOverlappingQueries().
-	AllowOverlappingQueries bool
-
 	// Compaction of overlapping blocks are allowed if AllowOverlappingCompaction is true.
 	// This is an optional flag for overlapping blocks.
 	// The reason why this flag exists is because there are various users of the TSDB
 	// that do not want vertical compaction happening on ingest time. Instead,
 	// they'd rather keep overlapping blocks and let another component do the overlapping compaction later.
-	// For Prometheus, this will always be enabled if overlapping queries is enabled.
+	// For Prometheus, this will always be true.
 	AllowOverlappingCompaction bool
 
 	// WALCompression will turn on Snappy compression for records on the WAL.
@@ -642,9 +635,6 @@ func validateOpts(opts *Options, rngs []int64) (*Options, []int64) {
 	if opts.MinBlockDuration > opts.MaxBlockDuration {
 		opts.MaxBlockDuration = opts.MinBlockDuration
 	}
-	if opts.OutOfOrderTimeWindow > 0 {
-		opts.AllowOverlappingQueries = true
-	}
 	if opts.OutOfOrderCapMax <= 0 {
 		opts.OutOfOrderCapMax = DefaultOutOfOrderCapMax
 	}
@@ -1368,11 +1358,6 @@ func (db *DB) reloadBlocks() (err error) {
 	sort.Slice(toLoad, func(i, j int) bool {
 		return toLoad[i].Meta().MinTime < toLoad[j].Meta().MinTime
 	})
-	if !db.AllowOverlappingQueries() {
-		if err := validateBlockSequence(toLoad); err != nil {
-			return errors.Wrap(err, "invalid block sequence")
-		}
-	}
 
 	// Swap new blocks first for subsequently created readers to be seen.
 	oldBlocks := db.blocks
@@ -1398,10 +1383,6 @@ func (db *DB) reloadBlocks() (err error) {
 	return nil
 }
 
-func (db *DB) AllowOverlappingQueries() bool {
-	return db.opts.AllowOverlappingQueries || db.oooWasEnabled.Load()
-}
-
 func openBlocks(l log.Logger, dir string, loaded []*Block, chunkPool chunkenc.Pool) (blocks []*Block, corrupted map[ulid.ULID]error, err error) {
 	bDirs, err := blockDirs(dir)
 	if err != nil {
@@ -1548,25 +1529,6 @@ func (db *DB) deleteBlocks(blocks map[ulid.ULID]*Block) error {
 	return nil
 }
 
-// validateBlockSequence returns error if given block meta files indicate that some blocks overlaps within sequence.
-func validateBlockSequence(bs []*Block) error {
-	if len(bs) <= 1 {
-		return nil
-	}
-
-	var metas []BlockMeta
-	for _, b := range bs {
-		metas = append(metas, b.meta)
-	}
-
-	overlaps := OverlappingBlocks(metas)
-	if len(overlaps) > 0 {
-		return errors.Errorf("block time ranges overlap: %s", overlaps)
-	}
-
-	return nil
-}
-
 // TimeRange specifies minTime and maxTime range.
 type TimeRange struct {
 	Min, Max int64

tsdb/db_test.go

@@ -3976,8 +3976,6 @@ func TestOOOCompaction(t *testing.T) {
 	opts := DefaultOptions()
 	opts.OutOfOrderCapMax = 30
 	opts.OutOfOrderTimeWindow = 300 * time.Minute.Milliseconds()
-	opts.AllowOverlappingQueries = true
-	opts.AllowOverlappingCompaction = true
 
 	db, err := Open(dir, nil, nil, opts, nil)
 	require.NoError(t, err)
@@ -4158,8 +4156,6 @@ func TestOOOCompactionWithNormalCompaction(t *testing.T) {
 	opts := DefaultOptions()
 	opts.OutOfOrderCapMax = 30
 	opts.OutOfOrderTimeWindow = 300 * time.Minute.Milliseconds()
-	opts.AllowOverlappingQueries = true
-	opts.AllowOverlappingCompaction = true
 
 	db, err := Open(dir, nil, nil, opts, nil)
 	require.NoError(t, err)
@@ -4255,7 +4251,6 @@ func Test_Querier_OOOQuery(t *testing.T) {
 	opts := DefaultOptions()
 	opts.OutOfOrderCapMax = 30
 	opts.OutOfOrderTimeWindow = 24 * time.Hour.Milliseconds()
-	opts.AllowOverlappingQueries = true
 	opts.AllowOverlappingCompaction = false
 
 	series1 := labels.FromStrings("foo", "bar1")
@@ -4341,7 +4336,6 @@ func Test_ChunkQuerier_OOOQuery(t *testing.T) {
 	opts := DefaultOptions()
 	opts.OutOfOrderCapMax = 30
 	opts.OutOfOrderTimeWindow = 24 * time.Hour.Milliseconds()
-	opts.AllowOverlappingQueries = true
 	opts.AllowOverlappingCompaction = false
 
 	series1 := labels.FromStrings("foo", "bar1")
@@ -4435,7 +4429,6 @@ func TestOOOAppendAndQuery(t *testing.T) {
 	opts := DefaultOptions()
 	opts.OutOfOrderCapMax = 30
 	opts.OutOfOrderTimeWindow = 4 * time.Hour.Milliseconds()
-	opts.AllowOverlappingQueries = true
 
 	db := openTestDB(t, opts, nil)
 	db.DisableCompactions()
@@ -4627,7 +4620,6 @@ func TestWBLAndMmapReplay(t *testing.T) {
 	opts := DefaultOptions()
 	opts.OutOfOrderCapMax = 30
 	opts.OutOfOrderTimeWindow = 4 * time.Hour.Milliseconds()
-	opts.AllowOverlappingQueries = true
 
 	db := openTestDB(t, opts, nil)
 	db.DisableCompactions()
@@ -4815,8 +4807,6 @@ func TestOOOCompactionFailure(t *testing.T) {
 	opts := DefaultOptions()
 	opts.OutOfOrderCapMax = 30
 	opts.OutOfOrderTimeWindow = 300 * time.Minute.Milliseconds()
-	opts.AllowOverlappingQueries = true
-	opts.AllowOverlappingCompaction = true
 
 	db, err := Open(dir, nil, nil, opts, nil)
 	require.NoError(t, err)
@@ -4956,8 +4946,6 @@ func TestWBLCorruption(t *testing.T) {
 	opts := DefaultOptions()
 	opts.OutOfOrderCapMax = 30
 	opts.OutOfOrderTimeWindow = 300 * time.Minute.Milliseconds()
-	opts.AllowOverlappingQueries = true
-	opts.AllowOverlappingCompaction = true
 
 	db, err := Open(dir, nil, nil, opts, nil)
 	require.NoError(t, err)
@@ -5104,8 +5092,6 @@ func TestOOOMmapCorruption(t *testing.T) {
 	opts := DefaultOptions()
 	opts.OutOfOrderCapMax = 10
 	opts.OutOfOrderTimeWindow = 300 * time.Minute.Milliseconds()
-	opts.AllowOverlappingQueries = true
-	opts.AllowOverlappingCompaction = true
 
 	db, err := Open(dir, nil, nil, opts, nil)
 	require.NoError(t, err)
@@ -5563,8 +5549,6 @@ func TestWblReplayAfterOOODisableAndRestart(t *testing.T) {
 
 	opts := DefaultOptions()
 	opts.OutOfOrderTimeWindow = 60 * time.Minute.Milliseconds()
-	opts.AllowOverlappingQueries = true
-	opts.AllowOverlappingCompaction = true
 
 	db, err := Open(dir, nil, nil, opts, nil)
 	require.NoError(t, err)
@@ -5624,7 +5608,6 @@ func TestPanicOnApplyConfig(t *testing.T) {
 
 	opts := DefaultOptions()
 	opts.OutOfOrderTimeWindow = 60 * time.Minute.Milliseconds()
-	opts.AllowOverlappingQueries = true
 
 	db, err := Open(dir, nil, nil, opts, nil)
 	require.NoError(t, err)
@@ -5673,7 +5656,6 @@ func TestDiskFillingUpAfterDisablingOOO(t *testing.T) {
 
 	opts := DefaultOptions()
 	opts.OutOfOrderTimeWindow = 60 * time.Minute.Milliseconds()
-	opts.AllowOverlappingQueries = true
 
 	db, err := Open(dir, nil, nil, opts, nil)
 	require.NoError(t, err)