Revert Compression Updates

CHANGELOG.md

@@ -10,8 +10,8 @@
 * [ENHANCEMENT] Added traceid to `trace too large message`. [#888](https://github.com/grafana/tempo/pull/888) (@mritunjaysharma394)
 * [ENHANCEMENT] Add support to tempo workloads to `overrides` from single configmap in microservice mode. [#896](https://github.com/grafana/tempo/pull/896) (@kavirajk)
 * [ENHANCEMENT] Make `overrides_config` block name consistent with Loki and Cortex in microservice mode. [#906](https://github.com/grafana/tempo/pull/906) (@kavirajk)
-* [ENHANCEMENT] Changes the metrics name from `cortex_runtime_config_last_reload_successful` to `tempo_runtime_config_last_reload_successful` [#945](https://github.com/grafana/tempo/pull/945) (@kavirajk)
 * [ENHANCEMENT] Updated config defaults to reflect better capture operational knowledge. [#913](https://github.com/grafana/tempo/pull/913) (@joe-elliott)
+* [ENHANCEMENT] Changes the metrics name from `cortex_runtime_config_last_reload_successful` to `tempo_runtime_config_last_reload_successful` [#945](https://github.com/grafana/tempo/pull/945) (@kavirajk)
   ```
   ingester:
     trace_idle_period: 30s => 10s  # reduce ingester memory requirements with little impact on querying
@@ -37,7 +37,6 @@
 * [ENHANCEMENT] Improve zstd read throughput using zstd.Decoder [#948](https://github.com/grafana/tempo/pull/948) (@joe-elliott)
 * [ENHANCEMENT] Dedupe search records while replaying WAL [#940](https://github.com/grafana/tempo/pull/940) (@annanay25)
 * [ENHANCEMENT] Add status endpoint to list the available endpoints [#938](https://github.com/grafana/tempo/pull/938) (@zalegrala)
-* [ENHANCEMENT] Compression updates: Added s2, improved snappy performance [#961](https://github.com/grafana/tempo/pull/961) (@joe-elliott)
 * [ENHANCEMENT] Add search block headers [#943](https://github.com/grafana/tempo/pull/943) (@mdisibio)
 * [ENHANCEMENT] Add search block headers for wal blocks [#963](https://github.com/grafana/tempo/pull/963) (@mdisibio)
 * [CHANGE] Renamed CLI flag from `--storage.trace.maintenance-cycle` to `--storage.trace.blocklist_poll`. This is a **breaking change**  [#897](https://github.com/grafana/tempo/pull/897) (@mritunjaysharma394)

docs/tempo/website/configuration/_index.md

@@ -523,7 +523,7 @@ storage:
             [path: <string>] 
 
             # wal encoding/compression.
-            # options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+            # options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
             # (default: snappy)
             [encoding: <string>]
 
@@ -542,7 +542,7 @@ storage:
             # (default: 1MiB)
             [index_downsample_bytes: <uint64>]
 
-            # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+            # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
             [encoding: <string>]
 ```
 

docs/tempo/website/configuration/compression.md

@@ -29,7 +29,6 @@ The following options are supported:
 - lz4
 - snappy
 - zstd
-- s2
 
 It is important to note that although all of these compression formats are supported in Tempo, at Grafana
 we use zstd and it's possible/probable that the other compression algorithms may have issue at scale.  Please 

example/docker-compose/azure/tempo-azure.yaml

@@ -34,10 +34,10 @@ storage:
     block:
       bloom_filter_false_positive: .05 # bloom filter false positive rate.  lower values create larger filters but fewer false positives
       index_downsample_bytes: 1000     # number of bytes per index record
-      encoding: zstd                   # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+      encoding: zstd                   # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
     wal:
       path: /tmp/tempo/wal             # where to store the the wal locally
-      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
     azure:
       container-name: tempo                    # how to store data in azure
       endpoint-suffix: azurite:10000

example/docker-compose/distributed/tempo-distributed.yaml

@@ -55,7 +55,7 @@ storage:
       insecure: true
     wal:
       path: /tmp/tempo/wal             # where to store the the wal locally
-      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
     local:
       path: /tmp/tempo/blocks
     pool:

example/docker-compose/gcs/tempo-gcs.yaml

@@ -35,10 +35,10 @@ storage:
     block:
       bloom_filter_false_positive: .05 # bloom filter false positive rate.  lower values create larger filters but fewer false positives
       index_downsample_bytes: 1000     # number of bytes per index record
-      encoding: zstd                   # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+      encoding: zstd                   # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
     wal:
       path: /tmp/tempo/wal             # where to store the the wal locally
-      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
     gcs:
       bucket_name: tempo
       endpoint: https://gcs:4443/storage/v1/

example/docker-compose/local/tempo-local.yaml

@@ -34,10 +34,10 @@ storage:
     block:
       bloom_filter_false_positive: .05 # bloom filter false positive rate.  lower values create larger filters but fewer false positives
       index_downsample_bytes: 1000     # number of bytes per index record
-      encoding: zstd                   # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+      encoding: zstd                   # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
     wal:
       path: /tmp/tempo/wal             # where to store the the wal locally
-      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
     local:
       path: /tmp/tempo/blocks
     pool:

example/docker-compose/s3/tempo-s3.yaml

@@ -35,10 +35,10 @@ storage:
     block:
       bloom_filter_false_positive: .05 # bloom filter false positive rate.  lower values create larger filters but fewer false positives
       index_downsample_bytes: 1000     # number of bytes per index record
-      encoding: zstd                   # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+      encoding: zstd                   # block encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
     wal:
       path: /tmp/tempo/wal             # where to store the the wal locally
-      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd, s2
+      encoding: snappy                 # wal encoding/compression.  options: none, gzip, lz4-64k, lz4-256k, lz4-1M, lz4, snappy, zstd
     s3:
       bucket: tempo                    # how to store data in s3
       endpoint: minio:9000

go.mod

@@ -19,6 +19,7 @@ require (
 	github.com/gogo/protobuf v1.3.2
 	github.com/gogo/status v1.1.0
 	github.com/golang/protobuf v1.5.2
+	github.com/golang/snappy v0.0.4
 	github.com/google/flatbuffers v2.0.0+incompatible
 	github.com/google/go-cmp v0.5.6
 	github.com/google/uuid v1.2.0
@@ -140,7 +141,6 @@ require (
 	github.com/gogo/googleapis v1.4.0 // indirect
 	github.com/golang-migrate/migrate/v4 v4.7.0 // indirect
 	github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da // indirect
-	github.com/golang/snappy v0.0.4 // indirect
 	github.com/google/btree v1.0.1 // indirect
 	github.com/google/pprof v0.0.0-20210609004039-a478d1d731e9 // indirect
 	github.com/googleapis/gax-go/v2 v2.0.5 // indirect

tempodb/backend/encoding.go

@@ -21,7 +21,6 @@ const (
 	EncLZ4_4M
 	EncSnappy
 	EncZstd
-	EncS2
 )
 
 // SupportedEncoding is a slice of all supported encodings
@@ -34,7 +33,6 @@ var SupportedEncoding = []Encoding{
 	EncLZ4_4M,
 	EncSnappy,
 	EncZstd,
-	EncS2,
 }
 
 func (e Encoding) String() string {
@@ -55,8 +53,6 @@ func (e Encoding) String() string {
 		return "snappy"
 	case EncZstd:
 		return "zstd"
-	case EncS2:
-		return "s2"
 	default:
 		return "unsupported"
 	}

tempodb/encoding/streaming_block_test.go

@@ -323,9 +323,6 @@ func BenchmarkWriteNone(b *testing.B) {
 func BenchmarkWriteZstd(b *testing.B) {
 	benchmarkCompressBlock(b, backend.EncZstd, benchDownsample, false)
 }
-func BenchmarkWriteS2(b *testing.B) {
-	benchmarkCompressBlock(b, backend.EncS2, benchDownsample, false)
-}
 
 func BenchmarkReadGzip(b *testing.B) {
 	benchmarkCompressBlock(b, backend.EncGZIP, benchDownsample, true)
@@ -345,9 +342,6 @@ func BenchmarkReadNone(b *testing.B) {
 func BenchmarkReadZstd(b *testing.B) {
 	benchmarkCompressBlock(b, backend.EncZstd, benchDownsample, true)
 }
-func BenchmarkReadS2(b *testing.B) {
-	benchmarkCompressBlock(b, backend.EncS2, benchDownsample, true)
-}
 
 // Download a block from your backend and place in ./benchmark_block/<tenant id>/<guid>
 //nolint:unparam
@@ -362,7 +356,7 @@ func benchmarkCompressBlock(b *testing.B, encoding backend.Encoding, indexDownsa
 	require.NoError(b, err, "error creating backend")
 
 	r := backend.NewReader(rawR)
-	meta, err := r.BlockMeta(context.Background(), uuid.MustParse("20a614f8-8cda-4b9d-9789-cb626f9fab28"), "1")
+	meta, err := r.BlockMeta(context.Background(), uuid.MustParse("00006e9d-94f0-4487-8e62-99f951be9349"), "1")
 	require.NoError(b, err)
 
 	backendBlock, err := NewBackendBlock(meta, r)

tempodb/encoding/v2/pool.go

@@ -5,16 +5,15 @@ import (
 	"io"
 	"sync"
 
+	"github.com/golang/snappy"
 	"github.com/grafana/tempo/tempodb/backend"
 	"github.com/klauspost/compress/gzip"
-	"github.com/klauspost/compress/s2"
-	"github.com/klauspost/compress/snappy"
 	"github.com/klauspost/compress/zstd"
 	"github.com/pierrec/lz4/v4"
 	"github.com/prometheus/prometheus/pkg/pool"
 )
 
-const maxEncoding = backend.EncS2
+const maxEncoding = backend.EncZstd
 
 // WriterPool is a pool of io.Writer
 // This is used by every chunk to avoid unnecessary allocations.
@@ -50,8 +49,6 @@ var (
 	Noop NoopPool
 	// Zstd Pool
 	Zstd = ZstdPool{}
-	// S2 Pool
-	S2 = S2Pool{}
 
 	// BytesBufferPool is a bytes buffer used for lines decompressed.
 	// Buckets [0.5KB,1KB,2KB,4KB,8KB]
@@ -85,8 +82,6 @@ func getReaderPool(enc backend.Encoding) (ReaderPool, error) {
 		return &Snappy, nil
 	case backend.EncZstd:
 		return &Zstd, nil
-	case backend.EncS2:
-		return &S2, nil
 	default:
 		return nil, fmt.Errorf("Unknown pool encoding %d", enc)
 	}
@@ -262,7 +257,7 @@ func (pool *SnappyPool) Encoding() backend.Encoding {
 // GetReader gets or creates a new CompressionReader and reset it to read from src
 func (pool *SnappyPool) GetReader(src io.Reader) (io.Reader, error) {
 	if r := pool.readers.Get(); r != nil {
-		reader := r.(*s2.Reader)
+		reader := r.(*snappy.Reader)
 		reader.Reset(src)
 		return reader, nil
 	}
@@ -276,15 +271,15 @@ func (pool *SnappyPool) PutReader(reader io.Reader) {
 
 // ResetReader implements ReaderPool
 func (pool *SnappyPool) ResetReader(src io.Reader, resetReader io.Reader) (io.Reader, error) {
-	reader := resetReader.(*s2.Reader)
+	reader := resetReader.(*snappy.Reader)
 	reader.Reset(src)
 	return reader, nil
 }
 
 // GetWriter gets or creates a new CompressionWriter and reset it to write to dst
 func (pool *SnappyPool) GetWriter(dst io.Writer) (io.WriteCloser, error) {
 	if w := pool.writers.Get(); w != nil {
-		writer := w.(*s2.Writer)
+		writer := w.(*snappy.Writer)
 		writer.Reset(dst)
 		return writer, nil
 	}
@@ -399,58 +394,3 @@ func (pool *ZstdPool) ResetWriter(dst io.Writer, resetWriter io.WriteCloser) (io
 	writer.Reset(dst)
 	return writer, nil
 }
-
-// S2Pool is one s short of s3
-type S2Pool struct {
-	readers sync.Pool
-	writers sync.Pool
-}
-
-// Encoding implements WriterPool and ReaderPool
-func (pool *S2Pool) Encoding() backend.Encoding {
-	return backend.EncS2
-}
-
-// GetReader gets or creates a new CompressionReader and reset it to read from src
-func (pool *S2Pool) GetReader(src io.Reader) (io.Reader, error) {
-	if r := pool.readers.Get(); r != nil {
-		reader := r.(*s2.Reader)
-		reader.Reset(src)
-		return reader, nil
-	}
-	return s2.NewReader(src), nil
-}
-
-// PutReader places back in the pool a CompressionReader
-func (pool *S2Pool) PutReader(reader io.Reader) {
-	pool.readers.Put(reader)
-}
-
-// ResetReader implements ReaderPool
-func (pool *S2Pool) ResetReader(src io.Reader, resetReader io.Reader) (io.Reader, error) {
-	reader := resetReader.(*s2.Reader)
-	reader.Reset(src)
-	return reader, nil
-}
-
-// GetWriter gets or creates a new CompressionWriter and reset it to write to dst
-func (pool *S2Pool) GetWriter(dst io.Writer) (io.WriteCloser, error) {
-	if w := pool.writers.Get(); w != nil {
-		writer := w.(*s2.Writer)
-		writer.Reset(dst)
-		return writer, nil
-	}
-	return s2.NewWriter(dst), nil
-}
-
-// PutWriter places back in the pool a CompressionWriter
-func (pool *S2Pool) PutWriter(writer io.WriteCloser) {
-	pool.writers.Put(writer)
-}
-
-// ResetWriter implements WriterPool
-func (pool *S2Pool) ResetWriter(dst io.Writer, resetWriter io.WriteCloser) (io.WriteCloser, error) {
-	writer := resetWriter.(*s2.Writer)
-	writer.Reset(dst)
-	return writer, nil
-}