Merge pull request #16193 from wenjiaswe/cp

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

clientv3/snapshot/util.go

@@ -21,13 +21,35 @@ 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))
+}
+
 // initIndex implements ConsistentIndexGetter so the snapshot won't block
 // the new raft instance by waiting for a future raft index.
 type initIndex int

clientv3/snapshot/v3_snapshot.go

@@ -242,6 +242,16 @@ 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.
@@ -303,6 +313,13 @@ 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
+		}
+	}
+
 	if err = s.saveWALAndSnap(); err != nil {
 		return err
 	}
@@ -417,6 +434,71 @@ 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 {
+	dbpath := filepath.Join(s.snapDir, "db")
+	be := backend.NewDefaultBackend(dbpath)
+	defer func() {
+		be.ForceCommit()
+		be.Close()
+	}()
+
+	tx := be.BatchTx()
+	tx.Lock()
+	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([]byte("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([]byte("key"), func(k, _ []byte) (err error) {
+		rev := bytesToRev(k)
+
+		if rev.GreaterThan(latest) {
+			latest = rev
+		}
+
+		return nil
+	})
+	return latest, err
+}
+
 // saveWALAndSnap creates a WAL for the initial cluster
 func (s *v3Manager) saveWALAndSnap() error {
 	if err := fileutil.CreateDirAll(s.walDir); err != nil {

etcdctl/ctlv3/command/snapshot_command.go

@@ -39,6 +39,8 @@ var (
 	restorePeerURLs     string
 	restoreName         string
 	skipHashCheck       bool
+	markCompacted       bool
+	revisionBump        uint64
 )
 
 // NewSnapshotCommand returns the cobra command for "snapshot".
@@ -85,6 +87,8 @@ 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
 }
@@ -166,6 +170,8 @@ func snapshotRestoreCommandFunc(cmd *cobra.Command, args []string) {
 		InitialCluster:      restoreCluster,
 		InitialClusterToken: restoreClusterToken,
 		SkipHashCheck:       skipHashCheck,
+		RevisionBump:        revisionBump,
+		MarkCompacted:       markCompacted,
 	}); err != nil {
 		ExitWithError(ExitError, err)
 	}

mvcc/util.go

@@ -55,3 +55,9 @@ func WriteKV(be backend.Backend, kv mvccpb.KeyValue) {
 	be.BatchTx().UnsafePut(keyBucketName, ibytes, d)
 	be.BatchTx().Unlock()
 }
+
+func UnsafeSetScheduledCompact(tx backend.BatchTx, value int64) {
+	rbytes := newRevBytes()
+	revToBytes(revision{main: value}, rbytes)
+	tx.UnsafePut(metaBucketName, scheduledCompactKeyName, rbytes)
+}

tests/e2e/ctl_v3_snapshot_test.go

@@ -15,7 +15,9 @@
 package e2e
 
 import (
+	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
 	"io/ioutil"
@@ -25,7 +27,10 @@ import (
 	"testing"
 	"time"
 
+	"github.com/stretchr/testify/require"
+	"go.etcd.io/etcd/clientv3"
 	"go.etcd.io/etcd/clientv3/snapshot"
+	v3rpc "go.etcd.io/etcd/etcdserver/api/v3rpc/rpctypes"
 	"go.etcd.io/etcd/pkg/expect"
 	"go.etcd.io/etcd/pkg/testutil"
 )
@@ -267,3 +272,166 @@ func TestIssue6361(t *testing.T) {
 		t.Fatal(err)
 	}
 }
+
+func TestRestoreCompactionRevBump(t *testing.T) {
+	defer testutil.AfterTest(t)
+
+	epc, err := newEtcdProcessCluster(&etcdProcessClusterConfig{
+		clusterSize:  1,
+		initialToken: "new",
+		keepDataDir:  true,
+	})
+	if err != nil {
+		t.Fatalf("could not start etcd process cluster (%v)", err)
+	}
+	defer func() {
+		if errC := epc.Close(); errC != nil {
+			t.Fatalf("error closing etcd processes (%v)", errC)
+		}
+	}()
+
+	dialTimeout := 10 * time.Second
+	prefixArgs := []string{ctlBinPath, "--endpoints", strings.Join(epc.EndpointsV3(), ","), "--dial-timeout", dialTimeout.String()}
+
+	ctl := newClient(t, epc.EndpointsV3(), epc.cfg.clientTLS, epc.cfg.isClientAutoTLS)
+	watchCh := ctl.Watch(context.Background(), "foo", clientv3.WithPrefix())
+	// flake-fix: the watch can sometimes miss the first put below causing test failure
+	time.Sleep(100 * time.Millisecond)
+
+	kvs := []kv{{"foo1", "val1"}, {"foo2", "val2"}, {"foo3", "val3"}}
+	for i := range kvs {
+		if err = spawnWithExpect(append(prefixArgs, "put", kvs[i].key, kvs[i].val), "OK"); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	watchTimeout := 1 * time.Second
+	watchRes, err := keyValuesFromWatchChan(watchCh, len(kvs), watchTimeout)
+	require.NoErrorf(t, err, "failed to get key-values from watch channel %s", err)
+	require.Equal(t, kvs, watchRes)
+
+	// ensure we get the right revision back for each of the keys
+	currentRev := 4
+	baseRev := 2
+	hasKVs(t, ctl, kvs, currentRev, baseRev)
+
+	fpath := filepath.Join(t.TempDir(), "test.snapshot")
+
+	t.Log("etcdctl saving snapshot...")
+	require.NoError(t, spawnWithExpect(append(prefixArgs, "snapshot", "save", fpath), fmt.Sprintf("Snapshot saved at %s", fpath)))
+
+	// add some more kvs that are not in the snapshot that will be lost after restore
+	unsnappedKVs := []kv{{"unsnapped1", "one"}, {"unsnapped2", "two"}, {"unsnapped3", "three"}}
+	for i := range unsnappedKVs {
+		if err = spawnWithExpect(append(prefixArgs, "put", unsnappedKVs[i].key, unsnappedKVs[i].val), "OK"); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	t.Log("Stopping the original server...")
+	require.NoError(t, epc.Stop())
+
+	newDataDir := filepath.Join(t.TempDir(), "test.data")
+	t.Log("etcdctl restoring the snapshot...")
+	bumpAmount := 10000
+
+	err = spawnWithExpect([]string{
+		ctlBinPath,
+		"snapshot",
+		"restore", fpath,
+		"--name", epc.procs[0].Config().name,
+		"--initial-cluster", epc.procs[0].Config().initialCluster,
+		"--initial-cluster-token", epc.procs[0].Config().initialToken,
+		"--initial-advertise-peer-urls", epc.procs[0].Config().purl.String(),
+		"--bump-revision", fmt.Sprintf("%d", bumpAmount),
+		"--mark-compacted",
+		"--data-dir", newDataDir,
+	}, "added member")
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	t.Log("(Re)starting the etcd member using the restored snapshot...")
+	epc.procs[0].Config().dataDirPath = newDataDir
+	for i := range epc.procs[0].Config().args {
+		if epc.procs[0].Config().args[i] == "--data-dir" {
+			epc.procs[0].Config().args[i+1] = newDataDir
+		}
+	}
+
+	require.NoError(t, epc.Restart())
+
+	t.Log("Ensuring the restored member has the correct data...")
+	hasKVs(t, ctl, kvs, currentRev, baseRev)
+
+	for i := range unsnappedKVs {
+		v, err := ctl.Get(context.Background(), unsnappedKVs[i].key)
+		require.NoError(t, err)
+		require.Equal(t, int64(0), v.Count)
+	}
+
+	cancelResult, ok := <-watchCh
+	require.True(t, ok, "watchChannel should be open")
+	require.Equal(t, v3rpc.ErrCompacted, cancelResult.Err())
+	require.Truef(t, cancelResult.Canceled, "expected ongoing watch to be cancelled after restoring with --mark-compacted")
+	require.Equal(t, int64(bumpAmount+currentRev), cancelResult.CompactRevision)
+	_, ok = <-watchCh
+	require.False(t, ok, "watchChannel should be closed after restoring with --mark-compacted")
+
+	// clients might restart the watch at the old base revision, that should not yield any new data
+	// everything up until bumpAmount+currentRev should return "already compacted"
+	for i := bumpAmount - 2; i < bumpAmount+currentRev; i++ {
+		watchCh = ctl.Watch(context.Background(), "foo", clientv3.WithPrefix(), clientv3.WithRev(int64(i)))
+		cancelResult := <-watchCh
+		require.Equal(t, v3rpc.ErrCompacted, cancelResult.Err())
+		require.Truef(t, cancelResult.Canceled, "expected ongoing watch to be cancelled after restoring with --mark-compacted")
+		require.Equal(t, int64(bumpAmount+currentRev), cancelResult.CompactRevision)
+	}
+
+	// a watch after that revision should yield successful results when a new put arrives
+	ctx, cancel := context.WithTimeout(context.Background(), watchTimeout*5)
+	defer cancel()
+	watchCh = ctl.Watch(ctx, "foo", clientv3.WithPrefix(), clientv3.WithRev(int64(bumpAmount+currentRev+1)))
+	if err = spawnWithExpect(append(prefixArgs, "put", "foo4", "val4"), "OK"); err != nil {
+		t.Fatal(err)
+	}
+	watchRes, err = keyValuesFromWatchChan(watchCh, 1, watchTimeout)
+	require.NoErrorf(t, err, "failed to get key-values from watch channel %s", err)
+	require.Equal(t, []kv{{"foo4", "val4"}}, watchRes)
+
+}
+
+func hasKVs(t *testing.T, ctl *clientv3.Client, kvs []kv, currentRev int, baseRev int) {
+	for i := range kvs {
+		v, err := ctl.Get(context.Background(), kvs[i].key)
+		require.NoError(t, err)
+		require.Equal(t, int64(1), v.Count)
+		require.Equal(t, kvs[i].val, string(v.Kvs[0].Value))
+		require.Equal(t, int64(baseRev+i), v.Kvs[0].CreateRevision)
+		require.Equal(t, int64(baseRev+i), v.Kvs[0].ModRevision)
+		require.Equal(t, int64(1), v.Kvs[0].Version)
+	}
+}
+
+func keyValuesFromWatchResponse(resp clientv3.WatchResponse) (kvs []kv) {
+	for _, event := range resp.Events {
+		kvs = append(kvs, kv{string(event.Kv.Key), string(event.Kv.Value)})
+	}
+	return kvs
+}
+
+func keyValuesFromWatchChan(wch clientv3.WatchChan, wantedLen int, timeout time.Duration) (kvs []kv, err error) {
+	for {
+		select {
+		case watchResp, ok := <-wch:
+			if ok {
+				kvs = append(kvs, keyValuesFromWatchResponse(watchResp)...)
+				if len(kvs) == wantedLen {
+					return kvs, nil
+				}
+			}
+		case <-time.After(timeout):
+			return nil, errors.New("closed watcher channel should not block")
+		}
+	}
+}