sstable: only support MinLZ for v6+ tables

We fall back to Snappy for older formats.

Fixes #4633

Author: RaduBerinde

options.go

@@ -49,7 +49,9 @@ const (
 	NoCompression      = block.NoCompression
 	SnappyCompression  = block.SnappyCompression
 	ZstdCompression    = block.ZstdCompression
-	MinLZCompression   = block.MinLZCompression
+	// MinLZCompression is only supported with table formats v6+. Older formats
+	// fall back to snappy.
+	MinLZCompression = block.MinLZCompression
 )
 
 // FilterType exports the base.FilterType type.

sstable/block/compression.go

@@ -25,6 +25,8 @@ const (
 	NoCompression
 	SnappyCompression
 	ZstdCompression
+	// MinLZCompression is only supported with table formats v6+. Older formats
+	// fall back to snappy.
 	MinLZCompression
 	NCompression
 )

sstable/format.go

@@ -29,7 +29,7 @@ const (
 	TableFormatPebblev3 // Value blocks.
 	TableFormatPebblev4 // DELSIZED tombstones.
 	TableFormatPebblev5 // Columnar blocks.
-	TableFormatPebblev6 // Checksum footer + blob value handles + columnar metaindex/properties.
+	TableFormatPebblev6 // Checksum footer + blob value handles + columnar metaindex/properties + MinLZ compression support.
 	NumTableFormats
 
 	TableFormatMax = NumTableFormats - 1

sstable/options.go

@@ -330,9 +330,6 @@ func (o WriterOptions) ensureDefaults() WriterOptions {
 	if o.Comparer == nil {
 		o.Comparer = base.DefaultComparer
 	}
-	if o.Compression <= block.DefaultCompression || o.Compression >= block.NCompression {
-		o.Compression = block.SnappyCompression
-	}
 	if o.IndexBlockSize <= 0 {
 		o.IndexBlockSize = o.BlockSize
 	}
@@ -357,5 +354,9 @@ func (o WriterOptions) ensureDefaults() WriterOptions {
 		s := colblk.DefaultKeySchema(o.Comparer, 16 /* bundle size */)
 		o.KeySchema = &s
 	}
+	if o.Compression <= block.DefaultCompression || o.Compression >= block.NCompression ||
+		(o.Compression == block.MinLZCompression && o.TableFormat < TableFormatPebblev6) {
+		o.Compression = block.SnappyCompression
+	}
 	return o
 }

sstable/test_utils.go

@@ -21,6 +21,7 @@ import (
 	"github.com/cockroachdb/pebble/internal/testkeys"
 	"github.com/cockroachdb/pebble/objstorage"
 	"github.com/cockroachdb/pebble/sstable/blob"
+	"github.com/cockroachdb/pebble/sstable/block"
 )
 
 // ReadAll returns all point keys, range del spans, and range key spans from an
@@ -274,6 +275,16 @@ func ParseWriterOptions[StringOrStringer any](o *WriterOptions, args ...StringOr
 		case "format", "leveldb":
 			return errors.Errorf("%q is deprecated", key)
 
+		case "compression":
+			o.Compression, err = func() (block.Compression, error) {
+				for c := block.DefaultCompression; c < block.NCompression; c++ {
+					if strings.EqualFold(c.String(), value) {
+						return c, nil
+					}
+				}
+				return 0, errors.Errorf("unknown compression %q", value)
+			}()
+
 		default:
 			// TODO(radu): ignoring unknown keys is error-prone; we need to find an
 			// easy way for the upper layer to extract its own arguments.

sstable/testdata/writer_v5

@@ -342,3 +342,24 @@ close
 ----
 point:    [a@2#1,SET-b@2#1,SET]
 seqnums:  [1-1]
+
+build compression=zstd
+a.SET.1:thequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydog
+----
+point:    [a#1,SET-a#1,SET]
+seqnums:  [1-1]
+
+props rocksdb.compression
+----
+rocksdb.compression: ZSTD
+
+# MinLZ is not supported in v5; we should fall back to Snappy.
+build compression=minlz
+a.SET.1:thequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydog
+----
+point:    [a#1,SET-a#1,SET]
+seqnums:  [1-1]
+
+props rocksdb.compression
+----
+rocksdb.compression: Snappy

sstable/testdata/writer_v6

@@ -365,3 +365,24 @@ close
 ----
 point:    [a@2#1,SET-b@2#1,SET]
 seqnums:  [1-1]
+
+build compression=zstd
+a.SET.1:thequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydog
+----
+point:    [a#1,SET-a#1,SET]
+seqnums:  [1-1]
+
+props rocksdb.compression
+----
+rocksdb.compression: ZSTD
+
+# MinLZ is supported in v6.
+build compression=minlz
+a.SET.1:thequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydogthequickbrownfoxjumpsoverthelazydog
+----
+point:    [a#1,SET-a#1,SET]
+seqnums:  [1-1]
+
+props rocksdb.compression
+----
+rocksdb.compression: MinLZ

sstable/writer_test.go

@@ -17,6 +17,7 @@ import (
 	"testing"
 	"unsafe"
 
+	"github.com/cockroachdb/crlib/crstrings"
 	"github.com/cockroachdb/crlib/testutils/leaktest"
 	"github.com/cockroachdb/datadriven"
 	"github.com/cockroachdb/errors"
@@ -306,7 +307,22 @@ func runDataDriven(t *testing.T, file string, tableFormat TableFormat) {
 			return formatWriterMetadata(td, meta)
 
 		case "props":
-			return r.Properties.String()
+			var buf strings.Builder
+			for _, p := range crstrings.Lines(r.Properties.String()) {
+				if len(td.CmdArgs) > 0 {
+					ok := false
+					for i := range td.CmdArgs {
+						if strings.HasPrefix(p, td.CmdArgs[i].String()+":") {
+							ok = true
+						}
+					}
+					if !ok {
+						continue
+					}
+				}
+				fmt.Fprintf(&buf, "%s\n", p)
+			}
+			return buf.String()
 
 		default:
 			return fmt.Sprintf("unknown command: %s", td.Cmd)