[3.5] Adding optional revision bump and mark compacted to snapshot restore

etcdctl/ctlv3/command/snapshot_command.go

@@ -40,6 +40,8 @@ var (
 	restorePeerURLs     string
 	restoreName         string
 	skipHashCheck       bool
+	markCompacted       bool
+	revisionBump        uint64
 )
 
 // NewSnapshotCommand returns the cobra command for "snapshot".
@@ -89,6 +91,8 @@ func NewSnapshotRestoreCommand() *cobra.Command {
 	cmd.Flags().StringVar(&restorePeerURLs, "initial-advertise-peer-urls", defaultInitialAdvertisePeerURLs, "List of this member's peer URLs to advertise to the rest of the cluster")
 	cmd.Flags().StringVar(&restoreName, "name", defaultName, "Human-readable name for this member")
 	cmd.Flags().BoolVar(&skipHashCheck, "skip-hash-check", false, "Ignore snapshot integrity hash value (required if copied from data directory)")
+	cmd.Flags().Uint64Var(&revisionBump, "bump-revision", 0, "How much to increase the latest revision after restore")
+	cmd.Flags().BoolVar(&markCompacted, "mark-compacted", false, "Mark the latest revision after restore as the point of scheduled compaction (required if --bump-revision > 0, disallowed otherwise)")
 
 	return cmd
 }
@@ -127,7 +131,7 @@ func snapshotStatusCommandFunc(cmd *cobra.Command, args []string) {
 func snapshotRestoreCommandFunc(cmd *cobra.Command, args []string) {
 	fmt.Fprintf(os.Stderr, "Deprecated: Use `etcdutl snapshot restore` instead.\n\n")
 	etcdutl.SnapshotRestoreCommandFunc(restoreCluster, restoreClusterToken, restoreDataDir, restoreWalDir,
-		restorePeerURLs, restoreName, skipHashCheck, args)
+		restorePeerURLs, restoreName, skipHashCheck, revisionBump, markCompacted, args)
 }
 
 func initialClusterFromName(name string) string {

etcdutl/etcdutl/snapshot_command.go

@@ -38,6 +38,8 @@ var (
 	restorePeerURLs     string
 	restoreName         string
 	skipHashCheck       bool
+	markCompacted       bool
+	revisionBump        uint64
 )
 
 // NewSnapshotCommand returns the cobra command for "snapshot".
@@ -91,6 +93,8 @@ func NewSnapshotRestoreCommand() *cobra.Command {
 	cmd.Flags().StringVar(&restorePeerURLs, "initial-advertise-peer-urls", defaultInitialAdvertisePeerURLs, "List of this member's peer URLs to advertise to the rest of the cluster")
 	cmd.Flags().StringVar(&restoreName, "name", defaultName, "Human-readable name for this member")
 	cmd.Flags().BoolVar(&skipHashCheck, "skip-hash-check", false, "Ignore snapshot integrity hash value (required if copied from data directory)")
+	cmd.Flags().Uint64Var(&revisionBump, "bump-revision", 0, "How much to increase the latest revision after restore")
+	cmd.Flags().BoolVar(&markCompacted, "mark-compacted", false, "Mark the latest revision after restore as the point of scheduled compaction (required if --bump-revision > 0, disallowed otherwise)")
 
 	cmd.MarkFlagRequired("data-dir")
 
@@ -115,7 +119,7 @@ func SnapshotStatusCommandFunc(cmd *cobra.Command, args []string) {
 
 func snapshotRestoreCommandFunc(_ *cobra.Command, args []string) {
 	SnapshotRestoreCommandFunc(restoreCluster, restoreClusterToken, restoreDataDir, restoreWalDir,
-		restorePeerURLs, restoreName, skipHashCheck, args)
+		restorePeerURLs, restoreName, skipHashCheck, revisionBump, markCompacted, args)
 }
 
 func SnapshotRestoreCommandFunc(restoreCluster string,
@@ -125,12 +129,19 @@ func SnapshotRestoreCommandFunc(restoreCluster string,
 	restorePeerURLs string,
 	restoreName string,
 	skipHashCheck bool,
+	revisionBump uint64,
+	markCompacted bool,
 	args []string) {
 	if len(args) != 1 {
 		err := fmt.Errorf("snapshot restore requires exactly one argument")
 		cobrautl.ExitWithError(cobrautl.ExitBadArgs, err)
 	}
 
+	if (revisionBump == 0 && markCompacted) || (revisionBump > 0 && !markCompacted) {
+		err := fmt.Errorf("--mark-compacted required if --revision-bump > 0")
+		cobrautl.ExitWithError(cobrautl.ExitBadArgs, err)
+	}
+
 	dataDir := restoreDataDir
 	if dataDir == "" {
 		dataDir = restoreName + ".etcd"
@@ -153,6 +164,8 @@ func SnapshotRestoreCommandFunc(restoreCluster string,
 		InitialCluster:      restoreCluster,
 		InitialClusterToken: restoreClusterToken,
 		SkipHashCheck:       skipHashCheck,
+		RevisionBump:        revisionBump,
+		MarkCompacted:       markCompacted,
 	}); err != nil {
 		cobrautl.ExitWithError(cobrautl.ExitError, err)
 	}

etcdutl/snapshot/util.go

@@ -23,9 +23,31 @@ type revision struct {
 	sub  int64
 }
 
+// GreaterThan should be synced with function in server
+// https://github.com/etcd-io/etcd/blob/main/server/storage/mvcc/revision.go
+func (a revision) GreaterThan(b revision) bool {
+	if a.main > b.main {
+		return true
+	}
+	if a.main < b.main {
+		return false
+	}
+	return a.sub > b.sub
+}
+
+// bytesToRev should be synced with function in server
+// https://github.com/etcd-io/etcd/blob/main/server/storage/mvcc/revision.go
 func bytesToRev(bytes []byte) revision {
 	return revision{
 		main: int64(binary.BigEndian.Uint64(bytes[0:8])),
 		sub:  int64(binary.BigEndian.Uint64(bytes[9:])),
 	}
 }
+
+// revToBytes should be synced with function in server
+// https://github.com/etcd-io/etcd/blob/main/server/storage/mvcc/revision.go
+func revToBytes(bytes []byte, rev revision) {
+	binary.BigEndian.PutUint64(bytes[0:8], uint64(rev.main))
+	bytes[8] = '_'
+	binary.BigEndian.PutUint64(bytes[9:], uint64(rev.sub))
+}

etcdutl/snapshot/v3_snapshot.go

@@ -40,7 +40,9 @@ import (
 	"go.etcd.io/etcd/server/v3/etcdserver/api/snap"
 	"go.etcd.io/etcd/server/v3/etcdserver/api/v2store"
 	"go.etcd.io/etcd/server/v3/etcdserver/cindex"
+	"go.etcd.io/etcd/server/v3/mvcc"
 	"go.etcd.io/etcd/server/v3/mvcc/backend"
+	"go.etcd.io/etcd/server/v3/mvcc/buckets"
 	"go.etcd.io/etcd/server/v3/verify"
 	"go.etcd.io/etcd/server/v3/wal"
 	"go.etcd.io/etcd/server/v3/wal/walpb"
@@ -194,6 +196,16 @@ type RestoreConfig struct {
 	// SkipHashCheck is "true" to ignore snapshot integrity hash value
 	// (required if copied from data directory).
 	SkipHashCheck bool
+
+	// RevisionBump is the amount to increase the latest revision after restore,
+	// to allow administrators to trick clients into thinking that revision never decreased.
+	// If 0, revision bumping is skipped.
+	// (required if MarkCompacted == true)
+	RevisionBump uint64
+
+	// MarkCompacted is "true" to mark the latest revision as compacted.
+	// (required if RevisionBump > 0)
+	MarkCompacted bool
 }
 
 // Restore restores a new etcd data directory from given snapshot file.
@@ -257,6 +269,13 @@ func (s *v3Manager) Restore(cfg RestoreConfig) error {
 	if err = s.saveDB(); err != nil {
 		return err
 	}
+
+	if cfg.MarkCompacted && cfg.RevisionBump > 0 {
+		if err = s.modifyLatestRevision(cfg.RevisionBump); err != nil {
+			return err
+		}
+	}
+
 	hardstate, err := s.saveWALAndSnap()
 	if err != nil {
 		return err
@@ -303,6 +322,70 @@ func (s *v3Manager) saveDB() error {
 	return nil
 }
 
+// modifyLatestRevision can increase the latest revision by the given amount and sets the scheduled compaction
+// to that revision so that the server will consider this revision compacted.
+func (s *v3Manager) modifyLatestRevision(bumpAmount uint64) error {
+	be := backend.NewDefaultBackend(s.outDbPath())
+	defer func() {
+		be.ForceCommit()
+		be.Close()
+	}()
+
+	tx := be.BatchTx()
+	tx.LockOutsideApply()
+	defer tx.Unlock()
+
+	latest, err := s.unsafeGetLatestRevision(tx)
+	if err != nil {
+		return err
+	}
+
+	latest = s.unsafeBumpRevision(tx, latest, int64(bumpAmount))
+	s.unsafeMarkRevisionCompacted(tx, latest)
+
+	return nil
+}
+
+func (s *v3Manager) unsafeBumpRevision(tx backend.BatchTx, latest revision, amount int64) revision {
+	s.lg.Info(
+		"bumping latest revision",
+		zap.Int64("latest-revision", latest.main),
+		zap.Int64("bump-amount", amount),
+		zap.Int64("new-latest-revision", latest.main+amount),
+	)
+
+	latest.main += amount
+	latest.sub = 0
+	k := make([]byte, 17)
+	revToBytes(k, latest)
+	tx.UnsafePut(buckets.Key, k, []byte{})
+
+	return latest
+}
+
+func (s *v3Manager) unsafeMarkRevisionCompacted(tx backend.BatchTx, latest revision) {
+	s.lg.Info(
+		"marking revision compacted",
+		zap.Int64("revision", latest.main),
+	)
+
+	mvcc.UnsafeSetScheduledCompact(tx, latest.main)
+}
+
+func (s *v3Manager) unsafeGetLatestRevision(tx backend.BatchTx) (revision, error) {
+	var latest revision
+	err := tx.UnsafeForEach(buckets.Key, func(k, _ []byte) (err error) {
+		rev := bytesToRev(k)
+
+		if rev.GreaterThan(latest) {
+			latest = rev
+		}
+
+		return nil
+	})
+	return latest, err
+}
+
 func (s *v3Manager) copyAndVerifyDB() error {
 	srcf, ferr := os.Open(s.srcDbPath)
 	if ferr != nil {

server/mvcc/util.go

@@ -35,3 +35,9 @@ func WriteKV(be backend.Backend, kv mvccpb.KeyValue) {
 	be.BatchTx().UnsafePut(buckets.Key, ibytes, d)
 	be.BatchTx().Unlock()
 }
+
+func UnsafeSetScheduledCompact(tx backend.BatchTx, value int64) {
+	rbytes := newRevBytes()
+	revToBytes(revision{main: value}, rbytes)
+	tx.UnsafePut(buckets.Meta, scheduledCompactKeyName, rbytes)
+}