[BREAKING]: Change how Badger handles WAL (dgraph-io#1555)

This PR significantly improves Badger's disk usage behavior.

Breaking: This PR increases the magic version from 7 to 8. So, no older Badger directories would work with this change.

With this PR, we no longer use value log as write-ahead log. Instead, each MemTable has its own WAL. Value logs now only write values which are greater than ValueThreshold, while MemTable WAL only writes smaller values and value pointers.

On a crash and restart, the MemTable WALs are replayed to apply updates to Skiplist. When MemTables are flushed to L0, the corresponding WALs are deleted.

This PR makes big changes to how value log GC works:
- Discard stats are now stored in a separate file, instead of within the LSM tree.
- GC only picks up value logs based off discard stats.
- GC no longer does sampling, it uses discard stats to inform when a value log needs to be GCed.
- Value log would now no longer grow indefinitely, because of the shift to MemTable WAL.
- Removed the `badger gc` tool.
- Value Log Head pointer tracking is removed.
- Only the last value log file is replayed on every start, and truncated as necessary.

This PR also makes a bunch of other changes:
- Removes ValueLogLoadingMode (always uses mmap now).
- Removes TableLoadingMode (always uses mmap now).
- Removes Truncate option.
- Removes KeepL0InMemory option.

backup_test.go

@@ -102,6 +102,7 @@ func TestBackupRestore1(t *testing.T) {
 			if err != nil {
 				return err
 			}
+			t.Logf("Got entry: %v\n", item.Version())
 			require.Equal(t, entries[count].key, item.Key())
 			require.Equal(t, entries[count].val, val)
 			require.Equal(t, entries[count].version, item.Version())
@@ -112,7 +113,7 @@ func TestBackupRestore1(t *testing.T) {
 		return nil
 	})
 	require.NoError(t, err)
-	require.Equal(t, db.orc.nextTs(), uint64(3))
+	require.Equal(t, 3, int(db.orc.nextTs()))
 }
 
 func TestBackupRestore2(t *testing.T) {

badger/cmd/backup.go

@@ -26,7 +26,6 @@ import (
 )
 
 var backupFile string
-var truncate bool
 
 // backupCmd represents the backup command
 var backupCmd = &cobra.Command{
@@ -45,16 +44,13 @@ func init() {
 	RootCmd.AddCommand(backupCmd)
 	backupCmd.Flags().StringVarP(&backupFile, "backup-file", "f",
 		"badger.bak", "File to backup to")
-	backupCmd.Flags().BoolVarP(&truncate, "truncate", "t",
-		false, "Allow value log truncation if required.")
 	backupCmd.Flags().IntVarP(&numVersions, "num-versions", "n",
 		0, "Number of versions to keep. A value <= 0 means keep all versions.")
 }
 
 func doBackup(cmd *cobra.Command, args []string) error {
 	opt := badger.DefaultOptions(sstDir).
 		WithValueDir(vlogDir).
-		WithTruncate(truncate).
 		WithNumVersionsToKeep(math.MaxInt32)
 
 	if numVersions > 0 {

badger/cmd/bank.go

@@ -364,11 +364,10 @@ func runTest(cmd *cobra.Command, args []string) error {
 		WithNumVersionsToKeep(int(math.MaxInt32)).
 		WithValueThreshold(1). // Make all values go to value log
 		WithCompression(options.ZSTD).
-		WithKeepL0InMemory(false).
 		WithBlockCacheSize(10 << 20)
 
-	if mmap {
-		opts = opts.WithTableLoadingMode(options.MemoryMap)
+	if verbose {
+		opts = opts.WithLoggingLevel(badger.DEBUG)
 	}
 
 	if encryptionKey != "" {

badger/cmd/flatten.go

@@ -52,7 +52,6 @@ func flatten(cmd *cobra.Command, args []string) error {
 	}
 	opt := badger.DefaultOptions(sstDir).
 		WithValueDir(vlogDir).
-		WithTruncate(truncate).
 		WithNumVersionsToKeep(numVersions).
 		WithNumCompactors(0)
 	fmt.Printf("Opening badger with options = %+v\n", opt)

badger/cmd/info.go

@@ -31,7 +31,6 @@ import (
 	"github.com/pkg/errors"
 
 	"github.com/dgraph-io/badger/v2"
-	"github.com/dgraph-io/badger/v2/options"
 	"github.com/dgraph-io/badger/v2/table"
 	"github.com/dgraph-io/badger/v2/y"
 	humanize "github.com/dustin/go-humanize"
@@ -93,18 +92,16 @@ to the Dgraph team.
 
 func handleInfo(cmd *cobra.Command, args []string) error {
 	if err := printInfo(sstDir, vlogDir); err != nil {
-		return errors.Wrap(err, "failed to print information in MANIFEST file")
+		return y.Wrap(err, "failed to print information in MANIFEST file")
 	}
 
 	// Open DB
 	db, err := badger.Open(badger.DefaultOptions(sstDir).
 		WithValueDir(vlogDir).
 		WithReadOnly(opt.readOnly).
-		WithTruncate(opt.truncate).
-		WithTableLoadingMode(options.MemoryMap).
 		WithEncryptionKey([]byte(opt.encryptionKey)))
 	if err != nil {
-		return errors.Wrap(err, "failed to open database")
+		return y.Wrap(err, "failed to open database")
 	}
 	defer db.Close()
 
@@ -114,7 +111,7 @@ func handleInfo(cmd *cobra.Command, args []string) error {
 
 	prefix, err := hex.DecodeString(opt.withPrefix)
 	if err != nil {
-		return errors.Wrapf(err, "failed to decode hex prefix: %s", opt.withPrefix)
+		return y.Wrapf(err, "failed to decode hex prefix: %s", opt.withPrefix)
 	}
 	if opt.showHistogram {
 		db.PrintHistogram(prefix)
@@ -128,7 +125,7 @@ func handleInfo(cmd *cobra.Command, args []string) error {
 
 	if len(opt.keyLookup) > 0 {
 		if err := lookup(db); err != nil {
-			return errors.Wrapf(err, "failed to perform lookup for the key: %x", opt.keyLookup)
+			return y.Wrapf(err, "failed to perform lookup for the key: %x", opt.keyLookup)
 		}
 	}
 	return nil
@@ -153,7 +150,7 @@ func showKeys(db *badger.DB, prefix []byte) error {
 	for it.Rewind(); it.Valid(); it.Next() {
 		item := it.Item()
 		if err := printKey(item, false); err != nil {
-			return errors.Wrapf(err, "failed to print information about key: %x(%d)",
+			return y.Wrapf(err, "failed to print information about key: %x(%d)",
 				item.Key(), item.Version())
 		}
 		totalKeys++
@@ -170,7 +167,7 @@ func lookup(db *badger.DB) error {
 
 	key, err := hex.DecodeString(opt.keyLookup)
 	if err != nil {
-		return errors.Wrapf(err, "failed to decode key: %q", opt.keyLookup)
+		return y.Wrapf(err, "failed to decode key: %q", opt.keyLookup)
 	}
 
 	iopts := badger.DefaultIteratorOptions
@@ -186,7 +183,7 @@ func lookup(db *badger.DB) error {
 	fmt.Println()
 	item := itr.Item()
 	if err := printKey(item, true); err != nil {
-		return errors.Wrapf(err, "failed to print information about key: %x(%d)",
+		return y.Wrapf(err, "failed to print information about key: %x(%d)",
 			item.Key(), item.Version())
 	}
 
@@ -201,7 +198,7 @@ func lookup(db *badger.DB) error {
 			break
 		}
 		if err := printKey(item, true); err != nil {
-			return errors.Wrapf(err, "failed to print information about key: %x(%d)",
+			return y.Wrapf(err, "failed to print information about key: %x(%d)",
 				item.Key(), item.Version())
 		}
 	}
@@ -223,7 +220,7 @@ func printKey(item *badger.Item, showValue bool) error {
 	if showValue {
 		val, err := item.ValueCopy(nil)
 		if err != nil {
-			return errors.Wrapf(err,
+			return y.Wrapf(err,
 				"failed to copy value of the key: %x(%d)", item.Key(), item.Version())
 		}
 		fmt.Fprintf(&buf, "\n\tvalue: %v", val)

badger/cmd/read_bench.go

@@ -21,15 +21,13 @@ import (
 	"fmt"
 	"math"
 	"math/rand"
-	"strings"
 	"sync/atomic"
 	"time"
 
 	humanize "github.com/dustin/go-humanize"
 	"github.com/spf13/cobra"
 
 	"github.com/dgraph-io/badger/v2"
-	"github.com/dgraph-io/badger/v2/options"
 	"github.com/dgraph-io/badger/v2/pb"
 	"github.com/dgraph-io/badger/v2/y"
 	"github.com/dgraph-io/ristretto/z"
@@ -107,12 +105,9 @@ func readBench(cmd *cobra.Command, args []string) error {
 		return y.Wrapf(err, "unable to parse duration")
 	}
 	y.AssertTrue(numGoroutines > 0)
-	mode := getLoadingMode(loadingMode)
 	opt := badger.DefaultOptions(sstDir).
 		WithValueDir(vlogDir).
 		WithReadOnly(readOnly).
-		WithTableLoadingMode(mode).
-		WithValueLogLoadingMode(mode).
 		WithBlockCacheSize(blockCacheSize << 20).
 		WithIndexCacheSize(indexCacheSize << 20)
 	fmt.Printf("Opening badger with options = %+v\n", opt)
@@ -239,18 +234,3 @@ func getSampleKeys(db *badger.DB) ([][]byte, error) {
 
 	return keys, nil
 }
-
-func getLoadingMode(m string) options.FileLoadingMode {
-	m = strings.ToLower(m)
-	var mode options.FileLoadingMode
-	switch m {
-	case "fileio":
-		mode = options.FileIO
-	case "mmap":
-		mode = options.MemoryMap
-	default:
-		panic("loading mode not supported")
-	}
-
-	return mode
-}

badger/cmd/rotate_test.go

@@ -31,10 +31,12 @@ func TestRotate(t *testing.T) {
 	dir, err := ioutil.TempDir("", "badger-test")
 	require.NoError(t, err)
 	defer os.RemoveAll(dir)
+
 	// Creating sample key.
 	key := make([]byte, 32)
 	_, err = rand.Read(key)
 	require.NoError(t, err)
+
 	fp, err := ioutil.TempFile("", "*.key")
 	require.NoError(t, err)
 	_, err = fp.Write(key)
@@ -44,6 +46,8 @@ func TestRotate(t *testing.T) {
 	// Opening DB with the encryption key.
 	opts := badger.DefaultOptions(dir)
 	opts.EncryptionKey = key
+	opts.BlockCacheSize = 1 << 20
+
 	db, err := badger.Open(opts)
 	require.NoError(t, err)
 	// Closing the db.
@@ -126,6 +130,7 @@ func TestRotatePlainTextToEncrypted(t *testing.T) {
 	require.Nil(t, doRotate(nil, []string{}))
 
 	// Try opening DB without the key.
+	opts.BlockCacheSize = 1 << 20
 	_, err = badger.Open(opts)
 	require.EqualError(t, err, badger.ErrEncryptionKeyMismatch.Error())
 

badger/cmd/stream.go

@@ -23,6 +23,7 @@ import (
 
 	"github.com/dgraph-io/badger/v2"
 	"github.com/dgraph-io/badger/v2/options"
+	"github.com/dgraph-io/badger/v2/y"
 	"github.com/pkg/errors"
 	"github.com/spf13/cobra"
 )
@@ -44,7 +45,6 @@ func init() {
 	RootCmd.AddCommand(streamCmd)
 	streamCmd.Flags().StringVarP(&outDir, "out", "o", "",
 		"Path to output DB. The directory should be empty.")
-	streamCmd.Flags().BoolVarP(&truncate, "truncate", "", false, "Option to truncate the DBs")
 	streamCmd.Flags().BoolVarP(&readOnly, "read_only", "", true,
 		"Option to open input DB in read-only mode")
 	streamCmd.Flags().IntVarP(&numVersions, "num_versions", "", 0,
@@ -76,7 +76,6 @@ func stream(cmd *cobra.Command, args []string) error {
 	}
 	inOpt := badger.DefaultOptions(sstDir).
 		WithReadOnly(readOnly).
-		WithTruncate(truncate).
 		WithValueThreshold(1 << 10 /* 1KB */).
 		WithNumVersionsToKeep(numVersions)
 
@@ -90,7 +89,7 @@ func stream(cmd *cobra.Command, args []string) error {
 
 	inDB, err := badger.OpenManaged(inOpt)
 	if err != nil {
-		return errors.Wrapf(err, "cannot open DB at %s", sstDir)
+		return y.Wrapf(err, "cannot open DB at %s", sstDir)
 	}
 	defer inDB.Close()
 	return inDB.StreamDB(outOpt)