Merge pull request #2842 from aawsome/rebuild-index-inmem

Rebuild index in prune by using in-memory index

changelog/unreleased/pull-2718

@@ -3,6 +3,11 @@ Enhancement: Improve pruning performance and make pruning more customizable
 The `prune` command is now much faster. This is especially the case for remote
 repositories or repositories with not much data to remove.
 Also the memory usage of the `prune` command is now reduced.
+Restic used to rebuild the index from scratch after pruning. This could lead
+to missing packs in the index in some cases for eventually consistent 
+backends, like e.g. AWS S3.
+This behavior is now changed and the index rebuilding uses the information
+already known by `prune`.
 
 By default, the `prune` command no longer removes all unused data. This
 behavior can be fine-tuned by new options, like the acceptable amount of unused space or
@@ -14,9 +19,11 @@ also shows what `prune` would do.
 
 Fixes several open issues, e.g.:
 https://github.com/restic/restic/issues/1140
+https://github.com/restic/restic/issues/1599
 https://github.com/restic/restic/issues/1985
 https://github.com/restic/restic/issues/2112
 https://github.com/restic/restic/issues/2227
 https://github.com/restic/restic/issues/2305
 
 https://github.com/restic/restic/pull/2718
+https://github.com/restic/restic/pull/2842

cmd/restic/cmd_prune.go

@@ -471,19 +471,31 @@ func prune(opts PruneOptions, gopts GlobalOptions, repo restic.Repository, usedB
 		DeleteFiles(gopts, repo, removePacksFirst, restic.PackFile)
 	}
 
+	packsAddedByRepack := 0
 	if len(repackPacks) != 0 {
+		// Remember the number of unique packs before repacking
+		packsBeforeRepacking := len(repo.Index().Packs())
+
 		Verbosef("repacking packs\n")
 		bar := newProgressMax(!gopts.Quiet, uint64(len(repackPacks)), "packs repacked")
 		_, err := repository.Repack(ctx, repo, repackPacks, keepBlobs, bar)
 		if err != nil {
 			return err
 		}
+
+		// Since repacking will only add new packs, we can calculate the number
+		// of packs like this:
+		packsAddedByRepack = len(repo.Index().Packs()) - packsBeforeRepacking
+
 		// Also remove repacked packs
 		removePacks.Merge(repackPacks)
 	}
 
 	if len(removePacks) != 0 {
-		if err = rebuildIndex(ctx, repo, removePacks); err != nil {
+		totalpacks := int(stats.packs.used+stats.packs.partlyUsed+stats.packs.unused) -
+			len(removePacks) + packsAddedByRepack
+		err = rebuildIndexFiles(gopts, repo, removePacks, uint64(totalpacks))
+		if err != nil {
 			return err
 		}
 
@@ -495,6 +507,20 @@ func prune(opts PruneOptions, gopts GlobalOptions, repo restic.Repository, usedB
 	return nil
 }
 
+func rebuildIndexFiles(gopts GlobalOptions, repo restic.Repository, removePacks restic.IDSet, packcount uint64) error {
+	Verbosef("rebuilding index\n")
+
+	bar := newProgressMax(!gopts.Quiet, packcount, "packs processed")
+	obsoleteIndexes, err := (repo.Index()).(*repository.MasterIndex).
+		Save(gopts.ctx, repo, removePacks, bar)
+	if err != nil {
+		return err
+	}
+
+	Verbosef("deleting obsolete index files\n")
+	return DeleteFilesChecked(gopts, repo, obsoleteIndexes, restic.IndexFile)
+}
+
 func getUsedBlobs(gopts GlobalOptions, repo restic.Repository, snapshots []*restic.Snapshot) (usedBlobs restic.BlobSet, err error) {
 	ctx := gopts.ctx
 

cmd/restic/integration_test.go

@@ -1559,6 +1559,62 @@ func testEdgeCaseRepo(t *testing.T, tarfile string, optionsCheck CheckOptions, o
 	}
 }
 
+// a listOnceBackend only allows listing once per filetype
+// listing filetypes more than once may cause problems with eventually consistent
+// backends (like e.g. AWS S3) as the second listing may be inconsistent to what
+// is expected by the first listing + some operations.
+type listOnceBackend struct {
+	restic.Backend
+	listedFileType map[restic.FileType]bool
+}
+
+func newListOnceBackend(be restic.Backend) *listOnceBackend {
+	return &listOnceBackend{
+		Backend:        be,
+		listedFileType: make(map[restic.FileType]bool),
+	}
+}
+
+func (be *listOnceBackend) List(ctx context.Context, t restic.FileType, fn func(restic.FileInfo) error) error {
+	if t != restic.LockFile && be.listedFileType[t] {
+		return errors.Errorf("tried listing type %v the second time", t)
+	}
+	be.listedFileType[t] = true
+	return be.Backend.List(ctx, t, fn)
+}
+
+func TestPruneListOnce(t *testing.T) {
+	env, cleanup := withTestEnvironment(t)
+	defer cleanup()
+
+	env.gopts.backendTestHook = func(r restic.Backend) (restic.Backend, error) {
+		return newListOnceBackend(r), nil
+	}
+
+	pruneOpts := PruneOptions{MaxUnused: "0"}
+	checkOpts := CheckOptions{ReadData: true, CheckUnused: true}
+
+	testSetupBackupData(t, env)
+	opts := BackupOptions{}
+
+	testRunBackup(t, "", []string{filepath.Join(env.testdata, "0", "0", "9")}, opts, env.gopts)
+	firstSnapshot := testRunList(t, "snapshots", env.gopts)
+	rtest.Assert(t, len(firstSnapshot) == 1,
+		"expected one snapshot, got %v", firstSnapshot)
+
+	testRunBackup(t, "", []string{filepath.Join(env.testdata, "0", "0", "9", "2")}, opts, env.gopts)
+	testRunBackup(t, "", []string{filepath.Join(env.testdata, "0", "0", "9", "3")}, opts, env.gopts)
+
+	snapshotIDs := testRunList(t, "snapshots", env.gopts)
+	rtest.Assert(t, len(snapshotIDs) == 3,
+		"expected 3 snapshot, got %v", snapshotIDs)
+
+	testRunForgetJSON(t, env.gopts)
+	testRunForget(t, env.gopts, firstSnapshot[0].String())
+	testRunPrune(t, env.gopts, pruneOpts)
+	rtest.OK(t, runCheck(checkOpts, env.gopts, nil))
+}
+
 func TestHardLink(t *testing.T) {
 	// this test assumes a test set with a single directory containing hard linked files
 	env, cleanup := withTestEnvironment(t)

doc/060_forget.rst

@@ -99,12 +99,10 @@ command must be run:
     
     repacking packs
     [0:00] 100.00%  2 / 2 packs repacked
-    counting files in repo
+    rebuilding index
-    [0:00] 100.00%  3 / 3 packs
+    [0:00] 100.00%  3 / 3 packs processed
-    finding old index files
+    deleting obsolete index files
-    saved new indexes as [59270b3a]
+    [0:00] 100.00%  3 / 3 files deleted
-    remove 4 old index files
-    [0:00] 100.00%  4 / 4 files deleted
     removing 3 old packs
     [0:00] 100.00%  3 / 3 files deleted
     done
@@ -147,12 +145,10 @@ to ``forget``:
     
     repacking packs
     [0:00] 100.00%  2 / 2 packs repacked
-    counting files in repo
+    rebuilding index
-    [0:00] 100.00%  3 / 3 packs
+    [0:00] 100.00%  3 / 3 packs processed
-    finding old index files
+    deleting obsolete index files
-    saved new indexes as [59270b3a]
+    [0:00] 100.00%  3 / 3 files deleted
-    remove 4 old index files
-    [0:00] 100.00%  4 / 4 files deleted
     removing 3 old packs
     [0:00] 100.00%  3 / 3 files deleted
     done

internal/repository/index.go

@@ -275,6 +275,55 @@ func (idx *Index) Each(ctx context.Context) <-chan restic.PackedBlob {
 	return ch
 }
 
+type EachByPackResult struct {
+	packID restic.ID
+	blobs  []restic.Blob
+}
+
+// EachByPack returns a channel that yields all blobs known to the index
+// grouped by packID but ignoring blobs with a packID in packPlacklist.
+// When the  context is cancelled, the background goroutine
+// terminates. This blocks any modification of the index.
+func (idx *Index) EachByPack(ctx context.Context, packBlacklist restic.IDSet) <-chan EachByPackResult {
+	idx.m.Lock()
+
+	ch := make(chan EachByPackResult)
+
+	go func() {
+		defer idx.m.Unlock()
+		defer func() {
+			close(ch)
+		}()
+
+		for typ := range idx.byType {
+			byPack := make(map[restic.ID][]*indexEntry)
+			m := &idx.byType[typ]
+			m.foreach(func(e *indexEntry) bool {
+				packID := idx.packs[e.packIndex]
+				if !packBlacklist.Has(packID) {
+					byPack[packID] = append(byPack[packID], e)
+				}
+				return true
+			})
+
+			for packID, pack := range byPack {
+				var result EachByPackResult
+				result.packID = packID
+				for _, e := range pack {
+					result.blobs = append(result.blobs, idx.toPackedBlob(e, restic.BlobType(typ)).Blob)
+				}
+				select {
+				case <-ctx.Done():
+					return
+				case ch <- result:
+				}
+			}
+		}
+	}()
+
+	return ch
+}
+
 // Packs returns all packs in this index
 func (idx *Index) Packs() restic.IDSet {
 	idx.m.Lock()

internal/repository/master_index.go

@@ -97,6 +97,19 @@ func (mi *MasterIndex) Has(id restic.ID, tpe restic.BlobType) bool {
 	return false
 }
 
+// Packs returns all packs that are covered by the index.
+func (mi *MasterIndex) Packs() restic.IDSet {
+	mi.idxMutex.RLock()
+	defer mi.idxMutex.RUnlock()
+
+	packs := restic.NewIDSet()
+	for _, idx := range mi.idx {
+		packs.Merge(idx.Packs())
+	}
+
+	return packs
+}
+
 // Count returns the number of blobs of type t in the index.
 func (mi *MasterIndex) Count(t restic.BlobType) (n uint) {
 	mi.idxMutex.RLock()
@@ -248,36 +261,34 @@ func (mi *MasterIndex) MergeFinalIndexes() {
 	mi.idx = newIdx
 }
 
-// RebuildIndex combines all known indexes to a new index, leaving out any
+// Save saves all known indexes to index files, leaving out any
 // packs whose ID is contained in packBlacklist. The new index contains the IDs
-// of all known indexes in the "supersedes" field.
+// of all known indexes in the "supersedes" field. The IDs are also returned in
-func (mi *MasterIndex) RebuildIndex(ctx context.Context, packBlacklist restic.IDSet) (*Index, error) {
+// the IDSet obsolete
+// After calling this function, you should remove the obsolete index files.
+func (mi *MasterIndex) Save(ctx context.Context, repo restic.Repository, packBlacklist restic.IDSet, p *restic.Progress) (obsolete restic.IDSet, err error) {
+	p.Start()
+	defer p.Done()
+
 	mi.idxMutex.Lock()
 	defer mi.idxMutex.Unlock()
 
 	debug.Log("start rebuilding index of %d indexes, pack blacklist: %v", len(mi.idx), packBlacklist)
 
 	newIndex := NewIndex()
+	obsolete = restic.NewIDSet()
 
-	ctx, cancel := context.WithCancel(ctx)
+	finalize := func() error {
-	defer cancel()
+		newIndex.Finalize()
+		if _, err := SaveIndex(ctx, repo, newIndex); err != nil {
+			return err
+		}
+		newIndex = NewIndex()
+		return nil
+	}
 
 	for i, idx := range mi.idx {
-		debug.Log("adding index %d", i)
+		if idx.Final() {
-
-		for pb := range idx.Each(ctx) {
-			if packBlacklist.Has(pb.PackID) {
-				continue
-			}
-
-			newIndex.Store(pb)
-		}
-
-		if !idx.Final() {
-			debug.Log("index %d isn't final, don't add to supersedes field", i)
-			continue
-		}
-
 			ids, err := idx.IDs()
 			if err != nil {
 				debug.Log("index %d does not have an ID: %v", err)
@@ -290,7 +301,26 @@ func (mi *MasterIndex) RebuildIndex(ctx context.Context, packBlacklist restic.ID
 			if err != nil {
 				return nil, err
 			}
+			obsolete.Merge(restic.NewIDSet(ids...))
+		} else {
+			debug.Log("index %d isn't final, don't add to supersedes field", i)
 		}
 
-	return newIndex, nil
+		debug.Log("adding index %d", i)
+
+		for pbs := range idx.EachByPack(ctx, packBlacklist) {
+			newIndex.StorePack(pbs.packID, pbs.blobs)
+			p.Report(restic.Stat{Blobs: 1})
+			if IndexFull(newIndex) {
+				if err := finalize(); err != nil {
+					return nil, err
+				}
+			}
+		}
+	}
+	if err := finalize(); err != nil {
+		return nil, err
+	}
+
+	return
 }

internal/repository/master_index_test.go

@@ -5,7 +5,9 @@ import (
 	"fmt"
 	"math/rand"
 	"testing"
+	"time"
 
+	"github.com/restic/restic/internal/checker"
 	"github.com/restic/restic/internal/repository"
 	"github.com/restic/restic/internal/restic"
 	rtest "github.com/restic/restic/internal/test"
@@ -322,3 +324,65 @@ func BenchmarkMasterIndexLookupBlobSize(b *testing.B) {
 		mIdx.LookupSize(lookupID, restic.DataBlob)
 	}
 }
+
+var (
+	snapshotTime = time.Unix(1470492820, 207401672)
+	depth        = 3
+)
+
+func createFilledRepo(t testing.TB, snapshots int, dup float32) (restic.Repository, func()) {
+	repo, cleanup := repository.TestRepository(t)
+
+	for i := 0; i < 3; i++ {
+		restic.TestCreateSnapshot(t, repo, snapshotTime.Add(time.Duration(i)*time.Second), depth, dup)
+	}
+
+	return repo, cleanup
+}
+
+func TestIndexSave(t *testing.T) {
+	repo, cleanup := createFilledRepo(t, 3, 0)
+	defer cleanup()
+
+	repo.LoadIndex(context.TODO())
+
+	obsoletes, err := repo.Index().(*repository.MasterIndex).Save(context.TODO(), repo, nil, nil)
+	if err != nil {
+		t.Fatalf("unable to save new index: %v", err)
+	}
+
+	for id := range obsoletes {
+		t.Logf("remove index %v", id.Str())
+		h := restic.Handle{Type: restic.IndexFile, Name: id.String()}
+		err = repo.Backend().Remove(context.TODO(), h)
+		if err != nil {
+			t.Errorf("error removing index %v: %v", id, err)
+		}
+	}
+
+	checker := checker.New(repo)
+	hints, errs := checker.LoadIndex(context.TODO())
+	for _, h := range hints {
+		t.Logf("hint: %v\n", h)
+	}
+
+	for _, err := range errs {
+		t.Errorf("checker found error: %v", err)
+	}
+
+	ctx, cancel := context.WithCancel(context.TODO())
+	defer cancel()
+
+	errCh := make(chan error)
+	go checker.Structure(ctx, errCh)
+	i := 0
+	for err := range errCh {
+		t.Errorf("checker returned error: %v", err)
+		i++
+		if i == 10 {
+			t.Errorf("more than 10 errors returned, skipping the rest")
+			cancel()
+			break
+		}
+	}
+}

internal/repository/repository.go

@@ -322,7 +322,10 @@ func (r *Repository) Flush(ctx context.Context) error {
 		return err
 	}
 
-	// Save index after flushing
+	// Save index after flushing only if noAutoIndexUpdate is not set
+	if r.noAutoIndexUpdate {
+		return nil
+	}
 	return r.SaveIndex(ctx)
 }
 

internal/restic/repository.go

@@ -62,6 +62,7 @@ type MasterIndex interface {
 	Has(ID, BlobType) bool
 	Lookup(ID, BlobType) []PackedBlob
 	Count(BlobType) uint
+	Packs() IDSet
 
 	// Each returns a channel that yields all blobs known to the index. When
 	// the context is cancelled, the background goroutine terminates. This