TheArchive/restic
1
0
Fork 0
mirror of https://github.com/restic/restic.git synced 2025-12-03 21:21:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

repository: split index into a separate package

Browse Source
This commit is contained in:
Michael Eischer
2022-06-12 14:43:43 +02:00
parent 5760ba6989
commit 2e3f1c08c5
20 changed files with 101 additions and 80 deletions
Download Patch File Download Diff File Expand all files Collapse all files

589
internal/index/index.go Normal file
View File

@@ -0,0 +1,589 @@
package index
import (
"context"
"encoding/json"
"io"
"sync"
"time"
"github.com/restic/restic/internal/crypto"
"github.com/restic/restic/internal/errors"
"github.com/restic/restic/internal/restic"
"github.com/restic/restic/internal/debug"
)
// In large repositories, millions of blobs are stored in the repository
// and restic needs to store an index entry for each blob in memory for
// most operations.
// Hence the index data structure defined here is one of the main contributions
// to the total memory requirements of restic.
//
// We store the index entries in indexMaps. In these maps, entries take 56
// bytes each, plus 8/4 = 2 bytes of unused pointers on average, not counting
// malloc and header struct overhead and ignoring duplicates (those are only
// present in edge cases and are also removed by prune runs).
//
// In the index entries, we need to reference the packID. As one pack may
// contain many blobs the packIDs are saved in a separate array and only the index
// within this array is saved in the indexEntry
//
// We assume on average a minimum of 8 blobs per pack; BP=8.
// (Note that for large files there should be 3 blobs per pack as the average chunk
// size is 1.5 MB and the minimum pack size is 4 MB)
//
// We have the following sizes:
// indexEntry: 56 bytes (on amd64)
// each packID: 32 bytes
//
// To save N index entries, we therefore need:
// N * (56 + 2) bytes + N * 32 bytes / BP = N * 62 bytes,
// i.e., fewer than 64 bytes per blob in an index.
// Index holds lookup tables for id -> pack.
type Index struct {
m sync.Mutex
byType [restic.NumBlobTypes]indexMap
packs restic.IDs
final bool // set to true for all indexes read from the backend ("finalized")
ids restic.IDs // set to the IDs of the contained finalized indexes
supersedes restic.IDs
created time.Time
}
// NewIndex returns a new index.
func NewIndex() *Index {
return &Index{
created: time.Now(),
}
}
// addToPacks saves the given pack ID and return the index.
// This procedere allows to use pack IDs which can be easily garbage collected after.
func (idx *Index) addToPacks(id restic.ID) int {
idx.packs = append(idx.packs, id)
return len(idx.packs) - 1
}
const maxuint32 = 1<<32 - 1
func (idx *Index) store(packIndex int, blob restic.Blob) {
// assert that offset and length fit into uint32!
if blob.Offset > maxuint32 || blob.Length > maxuint32 || blob.UncompressedLength > maxuint32 {
panic("offset or length does not fit in uint32. You have packs > 4GB!")
}
m := &idx.byType[blob.Type]
m.add(blob.ID, packIndex, uint32(blob.Offset), uint32(blob.Length), uint32(blob.UncompressedLength))
}
// Final returns true iff the index is already written to the repository, it is
// finalized.
func (idx *Index) Final() bool {
idx.m.Lock()
defer idx.m.Unlock()
return idx.final
}
const (
indexMaxBlobs = 50000
indexMaxBlobsCompressed = 3 * indexMaxBlobs
indexMaxAge = 10 * time.Minute
)
// IndexFull returns true iff the index is "full enough" to be saved as a preliminary index.
var IndexFull = func(idx *Index, compress bool) bool {
idx.m.Lock()
defer idx.m.Unlock()
debug.Log("checking whether index %p is full", idx)
var blobs uint
for typ := range idx.byType {
blobs += idx.byType[typ].len()
}
age := time.Since(idx.created)
var maxBlobs uint
if compress {
maxBlobs = indexMaxBlobsCompressed
} else {
maxBlobs = indexMaxBlobs
}
switch {
case age >= indexMaxAge:
debug.Log("index %p is old enough", idx, age)
return true
case blobs >= maxBlobs:
debug.Log("index %p has %d blobs", idx, blobs)
return true
}
debug.Log("index %p only has %d blobs and is too young (%v)", idx, blobs, age)
return false
}
// StorePack remembers the ids of all blobs of a given pack
// in the index
func (idx *Index) StorePack(id restic.ID, blobs []restic.Blob) {
idx.m.Lock()
defer idx.m.Unlock()
if idx.final {
panic("store new item in finalized index")
}
debug.Log("%v", blobs)
packIndex := idx.addToPacks(id)
for _, blob := range blobs {
idx.store(packIndex, blob)
}
}
func (idx *Index) toPackedBlob(e *indexEntry, t restic.BlobType) restic.PackedBlob {
return restic.PackedBlob{
Blob: restic.Blob{
BlobHandle: restic.BlobHandle{
ID: e.id,
Type: t},
Length: uint(e.length),
Offset: uint(e.offset),
UncompressedLength: uint(e.uncompressedLength),
},
PackID: idx.packs[e.packIndex],
}
}
// Lookup queries the index for the blob ID and returns all entries including
// duplicates. Adds found entries to blobs and returns the result.
func (idx *Index) Lookup(bh restic.BlobHandle, pbs []restic.PackedBlob) []restic.PackedBlob {
idx.m.Lock()
defer idx.m.Unlock()
idx.byType[bh.Type].foreachWithID(bh.ID, func(e *indexEntry) {
pbs = append(pbs, idx.toPackedBlob(e, bh.Type))
})
return pbs
}
// Has returns true iff the id is listed in the index.
func (idx *Index) Has(bh restic.BlobHandle) bool {
idx.m.Lock()
defer idx.m.Unlock()
return idx.byType[bh.Type].get(bh.ID) != nil
}
// LookupSize returns the length of the plaintext content of the blob with the
// given id.
func (idx *Index) LookupSize(bh restic.BlobHandle) (plaintextLength uint, found bool) {
idx.m.Lock()
defer idx.m.Unlock()
e := idx.byType[bh.Type].get(bh.ID)
if e == nil {
return 0, false
}
if e.uncompressedLength != 0 {
return uint(e.uncompressedLength), true
}
return uint(crypto.PlaintextLength(int(e.length))), true
}
// Supersedes returns the list of indexes this index supersedes, if any.
func (idx *Index) Supersedes() restic.IDs {
return idx.supersedes
}
// AddToSupersedes adds the ids to the list of indexes superseded by this
// index. If the index has already been finalized, an error is returned.
func (idx *Index) AddToSupersedes(ids ...restic.ID) error {
idx.m.Lock()
defer idx.m.Unlock()
if idx.final {
return errors.New("index already finalized")
}
idx.supersedes = append(idx.supersedes, ids...)
return nil
}
// Each passes all blobs known to the index to the callback fn. This blocks any
// modification of the index.
func (idx *Index) Each(ctx context.Context, fn func(restic.PackedBlob)) {
idx.m.Lock()
defer idx.m.Unlock()
for typ := range idx.byType {
m := &idx.byType[typ]
m.foreach(func(e *indexEntry) bool {
if ctx.Err() != nil {
return false
}
fn(idx.toPackedBlob(e, restic.BlobType(typ)))
return true
})
}
}
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 for
// finalized indexes.
// This filtering is used when rebuilding the index where we need to ignore packs
// from the finalized index which have been re-read into a non-finalized index.
// 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)
}()
byPack := make(map[restic.ID][][]*indexEntry)
for typ := range idx.byType {
m := &idx.byType[typ]
m.foreach(func(e *indexEntry) bool {
packID := idx.packs[e.packIndex]
if !idx.final || !packBlacklist.Has(packID) {
if _, ok := byPack[packID]; !ok {
byPack[packID] = make([][]*indexEntry, restic.NumBlobTypes)
}
byPack[packID][typ] = append(byPack[packID][typ], e)
}
return true
})
}
for packID, packByType := range byPack {
var result EachByPackResult
result.PackID = packID
for typ, pack := range packByType {
for _, e := range pack {
result.Blobs = append(result.Blobs, idx.toPackedBlob(e, restic.BlobType(typ)).Blob)
}
}
// allow GC once entry is no longer necessary
byPack[packID] = nil
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()
defer idx.m.Unlock()
packs := restic.NewIDSet()
for _, packID := range idx.packs {
packs.Insert(packID)
}
return packs
}
type packJSON struct {
ID restic.ID `json:"id"`
Blobs []blobJSON `json:"blobs"`
}
type blobJSON struct {
ID restic.ID `json:"id"`
Type restic.BlobType `json:"type"`
Offset uint `json:"offset"`
Length uint `json:"length"`
UncompressedLength uint `json:"uncompressed_length,omitempty"`
}
// generatePackList returns a list of packs.
func (idx *Index) generatePackList() ([]*packJSON, error) {
list := []*packJSON{}
packs := make(map[restic.ID]*packJSON)
for typ := range idx.byType {
m := &idx.byType[typ]
m.foreach(func(e *indexEntry) bool {
packID := idx.packs[e.packIndex]
if packID.IsNull() {
panic("null pack id")
}
debug.Log("handle blob %v", e.id)
// see if pack is already in map
p, ok := packs[packID]
if !ok {
// else create new pack
p = &packJSON{ID: packID}
// and append it to the list and map
list = append(list, p)
packs[p.ID] = p
}
// add blob
p.Blobs = append(p.Blobs, blobJSON{
ID: e.id,
Type: restic.BlobType(typ),
Offset: uint(e.offset),
Length: uint(e.length),
UncompressedLength: uint(e.uncompressedLength),
})
return true
})
}
debug.Log("done")
return list, nil
}
type jsonIndex struct {
Supersedes restic.IDs `json:"supersedes,omitempty"`
Packs []*packJSON `json:"packs"`
}
// Encode writes the JSON serialization of the index to the writer w.
func (idx *Index) Encode(w io.Writer) error {
debug.Log("encoding index")
idx.m.Lock()
defer idx.m.Unlock()
list, err := idx.generatePackList()
if err != nil {
return err
}
enc := json.NewEncoder(w)
idxJSON := jsonIndex{
Supersedes: idx.supersedes,
Packs: list,
}
return enc.Encode(idxJSON)
}
// Finalize sets the index to final.
func (idx *Index) Finalize() {
debug.Log("finalizing index")
idx.m.Lock()
defer idx.m.Unlock()
idx.final = true
}
// IDs returns the IDs of the index, if available. If the index is not yet
// finalized, an error is returned.
func (idx *Index) IDs() (restic.IDs, error) {
idx.m.Lock()
defer idx.m.Unlock()
if !idx.final {
return nil, errors.New("index not finalized")
}
return idx.ids, nil
}
// SetID sets the ID the index has been written to. This requires that
// Finalize() has been called before, otherwise an error is returned.
func (idx *Index) SetID(id restic.ID) error {
idx.m.Lock()
defer idx.m.Unlock()
if !idx.final {
return errors.New("index is not final")
}
if len(idx.ids) > 0 {
return errors.New("ID already set")
}
debug.Log("ID set to %v", id)
idx.ids = append(idx.ids, id)
return nil
}
// Dump writes the pretty-printed JSON representation of the index to w.
func (idx *Index) Dump(w io.Writer) error {
debug.Log("dumping index")
idx.m.Lock()
defer idx.m.Unlock()
list, err := idx.generatePackList()
if err != nil {
return err
}
outer := jsonIndex{
Supersedes: idx.Supersedes(),
Packs: list,
}
buf, err := json.MarshalIndent(outer, "", " ")
if err != nil {
return err
}
_, err = w.Write(append(buf, '\n'))
if err != nil {
return errors.Wrap(err, "Write")
}
debug.Log("done")
return nil
}
// merge() merges indexes, i.e. idx.merge(idx2) merges the contents of idx2 into idx.
// During merging exact duplicates are removed; idx2 is not changed by this method.
func (idx *Index) merge(idx2 *Index) error {
idx.m.Lock()
defer idx.m.Unlock()
idx2.m.Lock()
defer idx2.m.Unlock()
if !idx2.final {
return errors.New("index to merge is not final")
}
packlen := len(idx.packs)
// first append packs as they might be accessed when looking for duplicates below
idx.packs = append(idx.packs, idx2.packs...)
// copy all index entries of idx2 to idx
for typ := range idx2.byType {
m2 := &idx2.byType[typ]
m := &idx.byType[typ]
// helper func to test if identical entry is contained in idx
hasIdenticalEntry := func(e2 *indexEntry) (found bool) {
m.foreachWithID(e2.id, func(e *indexEntry) {
b := idx.toPackedBlob(e, restic.BlobType(typ))
b2 := idx2.toPackedBlob(e2, restic.BlobType(typ))
if b == b2 {
found = true
}
})
return found
}
m2.foreach(func(e2 *indexEntry) bool {
if !hasIdenticalEntry(e2) {
// packIndex needs to be changed as idx2.pack was appended to idx.pack, see above
m.add(e2.id, e2.packIndex+packlen, e2.offset, e2.length, e2.uncompressedLength)
}
return true
})
}
idx.ids = append(idx.ids, idx2.ids...)
idx.supersedes = append(idx.supersedes, idx2.supersedes...)
return nil
}
// isErrOldIndex returns true if the error may be caused by an old index
// format.
func isErrOldIndex(err error) bool {
e, ok := err.(*json.UnmarshalTypeError)
return ok && e.Value == "array"
}
// DecodeIndex unserializes an index from buf.
func DecodeIndex(buf []byte, id restic.ID) (idx *Index, oldFormat bool, err error) {
debug.Log("Start decoding index")
idxJSON := &jsonIndex{}
err = json.Unmarshal(buf, idxJSON)
if err != nil {
debug.Log("Error %v", err)
if isErrOldIndex(err) {
debug.Log("index is probably old format, trying that")
idx, err = decodeOldIndex(buf)
return idx, err == nil, err
}
return nil, false, errors.Wrap(err, "DecodeIndex")
}
idx = NewIndex()
for _, pack := range idxJSON.Packs {
packID := idx.addToPacks(pack.ID)
for _, blob := range pack.Blobs {
idx.store(packID, restic.Blob{
BlobHandle: restic.BlobHandle{
Type: blob.Type,
ID: blob.ID},
Offset: blob.Offset,
Length: blob.Length,
UncompressedLength: blob.UncompressedLength,
})
}
}
idx.supersedes = idxJSON.Supersedes
idx.ids = append(idx.ids, id)
idx.final = true
debug.Log("done")
return idx, false, nil
}
// DecodeOldIndex loads and unserializes an index in the old format from rd.
func decodeOldIndex(buf []byte) (idx *Index, err error) {
debug.Log("Start decoding old index")
list := []*packJSON{}
err = json.Unmarshal(buf, &list)
if err != nil {
debug.Log("Error %#v", err)
return nil, errors.Wrap(err, "Decode")
}
idx = NewIndex()
for _, pack := range list {
packID := idx.addToPacks(pack.ID)
for _, blob := range pack.Blobs {
idx.store(packID, restic.Blob{
BlobHandle: restic.BlobHandle{
Type: blob.Type,
ID: blob.ID},
Offset: blob.Offset,
Length: blob.Length,
// no compressed length in the old index format
})
}
}
idx.final = true
debug.Log("done")
return idx, nil
}

83
internal/index/index_parallel.go Normal file
View File

@@ -0,0 +1,83 @@
package index
import (
"context"
"runtime"
"sync"
"github.com/restic/restic/internal/debug"
"github.com/restic/restic/internal/restic"
"golang.org/x/sync/errgroup"
)
// ForAllIndexes loads all index files in parallel and calls the given callback.
// It is guaranteed that the function is not run concurrently. If the callback
// returns an error, this function is cancelled and also returns that error.
func ForAllIndexes(ctx context.Context, repo restic.Repository,
fn func(id restic.ID, index *Index, oldFormat bool, err error) error) error {
debug.Log("Start")
type FileInfo struct {
restic.ID
Size int64
}
var m sync.Mutex
// track spawned goroutines using wg, create a new context which is
// cancelled as soon as an error occurs.
wg, ctx := errgroup.WithContext(ctx)
ch := make(chan FileInfo)
// send list of index files through ch, which is closed afterwards
wg.Go(func() error {
defer close(ch)
return repo.List(ctx, restic.IndexFile, func(id restic.ID, size int64) error {
select {
case <-ctx.Done():
return ctx.Err()
case ch <- FileInfo{id, size}:
}
return nil
})
})
// a worker receives an index ID from ch, loads the index, and sends it to indexCh
worker := func() error {
var buf []byte
for fi := range ch {
debug.Log("worker got file %v", fi.ID.Str())
var err error
var idx *Index
oldFormat := false
if cap(buf) < int(fi.Size) {
// overallocate a bit
buf = make([]byte, fi.Size+128*1024)
}
buf, err = repo.LoadUnpacked(ctx, restic.IndexFile, fi.ID, buf[:0])
if err == nil {
idx, oldFormat, err = DecodeIndex(buf, fi.ID)
}
m.Lock()
err = fn(fi.ID, idx, oldFormat, err)
m.Unlock()
if err != nil {
return err
}
}
return nil
}
// decoding an index can take quite some time such that this can be both CPU- or IO-bound
// as the whole index is kept in memory anyways, a few workers too much don't matter
workerCount := int(repo.Connections()) + runtime.GOMAXPROCS(0)
// run workers on ch
for i := 0; i < workerCount; i++ {
wg.Go(worker)
}
return wg.Wait()
}

50
internal/index/index_parallel_test.go Normal file
View File

@@ -0,0 +1,50 @@
package index_test
import (
"context"
"path/filepath"
"testing"
"github.com/restic/restic/internal/errors"
"github.com/restic/restic/internal/index"
"github.com/restic/restic/internal/repository"
"github.com/restic/restic/internal/restic"
rtest "github.com/restic/restic/internal/test"
)
var repoFixture = filepath.Join("..", "repository", "testdata", "test-repo.tar.gz")
func TestRepositoryForAllIndexes(t *testing.T) {
repodir, cleanup := rtest.Env(t, repoFixture)
defer cleanup()
repo := repository.TestOpenLocal(t, repodir)
expectedIndexIDs := restic.NewIDSet()
rtest.OK(t, repo.List(context.TODO(), restic.IndexFile, func(id restic.ID, size int64) error {
expectedIndexIDs.Insert(id)
return nil
}))
// check that all expected indexes are loaded without errors
indexIDs := restic.NewIDSet()
var indexErr error
rtest.OK(t, index.ForAllIndexes(context.TODO(), repo, func(id restic.ID, index *index.Index, oldFormat bool, err error) error {
if err != nil {
indexErr = err
}
indexIDs.Insert(id)
return nil
}))
rtest.OK(t, indexErr)
rtest.Equals(t, expectedIndexIDs, indexIDs)
// must failed with the returned error
iterErr := errors.New("error to pass upwards")
err := index.ForAllIndexes(context.TODO(), repo, func(id restic.ID, index *index.Index, oldFormat bool, err error) error {
return iterErr
})
rtest.Equals(t, iterErr, err)
}

649
internal/index/index_test.go Normal file
View File

@@ -0,0 +1,649 @@
package index_test
import (
"bytes"
"context"
"math/rand"
"sync"
"testing"
"github.com/restic/restic/internal/index"
"github.com/restic/restic/internal/restic"
rtest "github.com/restic/restic/internal/test"
)
func TestIndexSerialize(t *testing.T) {
tests := []restic.PackedBlob{}
idx := index.NewIndex()
// create 50 packs with 20 blobs each
for i := 0; i < 50; i++ {
packID := restic.NewRandomID()
var blobs []restic.Blob
pos := uint(0)
for j := 0; j < 20; j++ {
length := uint(i*100 + j)
uncompressedLength := uint(0)
if i >= 25 {
// test a mix of compressed and uncompressed packs
uncompressedLength = 2 * length
}
pb := restic.PackedBlob{
Blob: restic.Blob{
BlobHandle: restic.NewRandomBlobHandle(),
Offset: pos,
Length: length,
UncompressedLength: uncompressedLength,
},
PackID: packID,
}
blobs = append(blobs, pb.Blob)
tests = append(tests, pb)
pos += length
}
idx.StorePack(packID, blobs)
}
wr := bytes.NewBuffer(nil)
err := idx.Encode(wr)
rtest.OK(t, err)
idx2ID := restic.NewRandomID()
idx2, oldFormat, err := index.DecodeIndex(wr.Bytes(), idx2ID)
rtest.OK(t, err)
rtest.Assert(t, idx2 != nil,
"nil returned for decoded index")
rtest.Assert(t, !oldFormat, "new index format recognized as old format")
indexID, err := idx2.IDs()
rtest.OK(t, err)
rtest.Equals(t, indexID, restic.IDs{idx2ID})
wr2 := bytes.NewBuffer(nil)
err = idx2.Encode(wr2)
rtest.OK(t, err)
for _, testBlob := range tests {
list := idx.Lookup(testBlob.BlobHandle, nil)
if len(list) != 1 {
t.Errorf("expected one result for blob %v, got %v: %v", testBlob.ID.Str(), len(list), list)
}
result := list[0]
rtest.Equals(t, testBlob, result)
list2 := idx2.Lookup(testBlob.BlobHandle, nil)
if len(list2) != 1 {
t.Errorf("expected one result for blob %v, got %v: %v", testBlob.ID.Str(), len(list2), list2)
}
result2 := list2[0]
rtest.Equals(t, testBlob, result2)
}
// add more blobs to idx
newtests := []restic.PackedBlob{}
for i := 0; i < 10; i++ {
packID := restic.NewRandomID()
var blobs []restic.Blob
pos := uint(0)
for j := 0; j < 10; j++ {
length := uint(i*100 + j)
pb := restic.PackedBlob{
Blob: restic.Blob{
BlobHandle: restic.NewRandomBlobHandle(),
Offset: pos,
Length: length,
},
PackID: packID,
}
blobs = append(blobs, pb.Blob)
newtests = append(newtests, pb)
pos += length
}
idx.StorePack(packID, blobs)
}
// finalize; serialize idx, unserialize to idx3
idx.Finalize()
wr3 := bytes.NewBuffer(nil)
err = idx.Encode(wr3)
rtest.OK(t, err)
rtest.Assert(t, idx.Final(),
"index not final after encoding")
id := restic.NewRandomID()
rtest.OK(t, idx.SetID(id))
ids, err := idx.IDs()
rtest.OK(t, err)
rtest.Equals(t, restic.IDs{id}, ids)
idx3, oldFormat, err := index.DecodeIndex(wr3.Bytes(), id)
rtest.OK(t, err)
rtest.Assert(t, idx3 != nil,
"nil returned for decoded index")
rtest.Assert(t, idx3.Final(),
"decoded index is not final")
rtest.Assert(t, !oldFormat, "new index format recognized as old format")
// all new blobs must be in the index
for _, testBlob := range newtests {
list := idx3.Lookup(testBlob.BlobHandle, nil)
if len(list) != 1 {
t.Errorf("expected one result for blob %v, got %v: %v", testBlob.ID.Str(), len(list), list)
}
blob := list[0]
rtest.Equals(t, testBlob, blob)
}
}
func TestIndexSize(t *testing.T) {
idx := index.NewIndex()
packs := 200
blobCount := 100
for i := 0; i < packs; i++ {
packID := restic.NewRandomID()
var blobs []restic.Blob
pos := uint(0)
for j := 0; j < blobCount; j++ {
length := uint(i*100 + j)
blobs = append(blobs, restic.Blob{
BlobHandle: restic.NewRandomBlobHandle(),
Offset: pos,
Length: length,
})
pos += length
}
idx.StorePack(packID, blobs)
}
wr := bytes.NewBuffer(nil)
err := idx.Encode(wr)
rtest.OK(t, err)
t.Logf("Index file size for %d blobs in %d packs is %d", blobCount*packs, packs, wr.Len())
}
// example index serialization from doc/Design.rst
var docExampleV1 = []byte(`
{
"supersedes": [
"ed54ae36197f4745ebc4b54d10e0f623eaaaedd03013eb7ae90df881b7781452"
],
"packs": [
{
"id": "73d04e6125cf3c28a299cc2f3cca3b78ceac396e4fcf9575e34536b26782413c",
"blobs": [
{
"id": "3ec79977ef0cf5de7b08cd12b874cd0f62bbaf7f07f3497a5b1bbcc8cb39b1ce",
"type": "data",
"offset": 0,
"length": 38
},{
"id": "9ccb846e60d90d4eb915848add7aa7ea1e4bbabfc60e573db9f7bfb2789afbae",
"type": "tree",
"offset": 38,
"length": 112
},
{
"id": "d3dc577b4ffd38cc4b32122cabf8655a0223ed22edfd93b353dc0c3f2b0fdf66",
"type": "data",
"offset": 150,
"length": 123
}
]
}
]
}
`)
var docExampleV2 = []byte(`
{
"supersedes": [
"ed54ae36197f4745ebc4b54d10e0f623eaaaedd03013eb7ae90df881b7781452"
],
"packs": [
{
"id": "73d04e6125cf3c28a299cc2f3cca3b78ceac396e4fcf9575e34536b26782413c",
"blobs": [
{
"id": "3ec79977ef0cf5de7b08cd12b874cd0f62bbaf7f07f3497a5b1bbcc8cb39b1ce",
"type": "data",
"offset": 0,
"length": 38
},
{
"id": "9ccb846e60d90d4eb915848add7aa7ea1e4bbabfc60e573db9f7bfb2789afbae",
"type": "tree",
"offset": 38,
"length": 112,
"uncompressed_length": 511
},
{
"id": "d3dc577b4ffd38cc4b32122cabf8655a0223ed22edfd93b353dc0c3f2b0fdf66",
"type": "data",
"offset": 150,
"length": 123,
"uncompressed_length": 234
}
]
}
]
}
`)
var docOldExample = []byte(`
[ {
"id": "73d04e6125cf3c28a299cc2f3cca3b78ceac396e4fcf9575e34536b26782413c",
"blobs": [
{
"id": "3ec79977ef0cf5de7b08cd12b874cd0f62bbaf7f07f3497a5b1bbcc8cb39b1ce",
"type": "data",
"offset": 0,
"length": 38
},{
"id": "9ccb846e60d90d4eb915848add7aa7ea1e4bbabfc60e573db9f7bfb2789afbae",
"type": "tree",
"offset": 38,
"length": 112
},
{
"id": "d3dc577b4ffd38cc4b32122cabf8655a0223ed22edfd93b353dc0c3f2b0fdf66",
"type": "data",
"offset": 150,
"length": 123
}
]
} ]
`)
var exampleTests = []struct {
id, packID restic.ID
tpe restic.BlobType
offset, length uint
uncompressedLength uint
}{
{
restic.TestParseID("3ec79977ef0cf5de7b08cd12b874cd0f62bbaf7f07f3497a5b1bbcc8cb39b1ce"),
restic.TestParseID("73d04e6125cf3c28a299cc2f3cca3b78ceac396e4fcf9575e34536b26782413c"),
restic.DataBlob, 0, 38, 0,
}, {
restic.TestParseID("9ccb846e60d90d4eb915848add7aa7ea1e4bbabfc60e573db9f7bfb2789afbae"),
restic.TestParseID("73d04e6125cf3c28a299cc2f3cca3b78ceac396e4fcf9575e34536b26782413c"),
restic.TreeBlob, 38, 112, 511,
}, {
restic.TestParseID("d3dc577b4ffd38cc4b32122cabf8655a0223ed22edfd93b353dc0c3f2b0fdf66"),
restic.TestParseID("73d04e6125cf3c28a299cc2f3cca3b78ceac396e4fcf9575e34536b26782413c"),
restic.DataBlob, 150, 123, 234,
},
}
var exampleLookupTest = struct {
packID restic.ID
blobs map[restic.ID]restic.BlobType
}{
restic.TestParseID("73d04e6125cf3c28a299cc2f3cca3b78ceac396e4fcf9575e34536b26782413c"),
map[restic.ID]restic.BlobType{
restic.TestParseID("3ec79977ef0cf5de7b08cd12b874cd0f62bbaf7f07f3497a5b1bbcc8cb39b1ce"): restic.DataBlob,
restic.TestParseID("9ccb846e60d90d4eb915848add7aa7ea1e4bbabfc60e573db9f7bfb2789afbae"): restic.TreeBlob,
restic.TestParseID("d3dc577b4ffd38cc4b32122cabf8655a0223ed22edfd93b353dc0c3f2b0fdf66"): restic.DataBlob,
},
}
func TestIndexUnserialize(t *testing.T) {
for _, task := range []struct {
idxBytes []byte
version int
}{
{docExampleV1, 1},
{docExampleV2, 2},
} {
oldIdx := restic.IDs{restic.TestParseID("ed54ae36197f4745ebc4b54d10e0f623eaaaedd03013eb7ae90df881b7781452")}
idx, oldFormat, err := index.DecodeIndex(task.idxBytes, restic.NewRandomID())
rtest.OK(t, err)
rtest.Assert(t, !oldFormat, "new index format recognized as old format")
for _, test := range exampleTests {
list := idx.Lookup(restic.BlobHandle{ID: test.id, Type: test.tpe}, nil)
if len(list) != 1 {
t.Errorf("expected one result for blob %v, got %v: %v", test.id.Str(), len(list), list)
}
blob := list[0]
t.Logf("looking for blob %v/%v, got %v", test.tpe, test.id.Str(), blob)
rtest.Equals(t, test.packID, blob.PackID)
rtest.Equals(t, test.tpe, blob.Type)
rtest.Equals(t, test.offset, blob.Offset)
rtest.Equals(t, test.length, blob.Length)
if task.version == 1 {
rtest.Equals(t, uint(0), blob.UncompressedLength)
} else if task.version == 2 {
rtest.Equals(t, test.uncompressedLength, blob.UncompressedLength)
} else {
t.Fatal("Invalid index version")
}
}
rtest.Equals(t, oldIdx, idx.Supersedes())
blobs := listPack(idx, exampleLookupTest.packID)
if len(blobs) != len(exampleLookupTest.blobs) {
t.Fatalf("expected %d blobs in pack, got %d", len(exampleLookupTest.blobs), len(blobs))
}
for _, blob := range blobs {
b, ok := exampleLookupTest.blobs[blob.ID]
if !ok {
t.Errorf("unexpected blob %v found", blob.ID.Str())
}
if blob.Type != b {
t.Errorf("unexpected type for blob %v: want %v, got %v", blob.ID.Str(), b, blob.Type)
}
}
}
}
func listPack(idx *index.Index, id restic.ID) (pbs []restic.PackedBlob) {
idx.Each(context.TODO(), func(pb restic.PackedBlob) {
if pb.PackID.Equal(id) {
pbs = append(pbs, pb)
}
})
return pbs
}
var (
benchmarkIndexJSON []byte
benchmarkIndexJSONOnce sync.Once
)
func initBenchmarkIndexJSON() {
idx, _ := createRandomIndex(rand.New(rand.NewSource(0)), 200000)
var buf bytes.Buffer
err := idx.Encode(&buf)
if err != nil {
panic(err)
}
benchmarkIndexJSON = buf.Bytes()
}
func BenchmarkDecodeIndex(b *testing.B) {
benchmarkIndexJSONOnce.Do(initBenchmarkIndexJSON)
id := restic.NewRandomID()
b.ResetTimer()
for i := 0; i < b.N; i++ {
_, _, err := index.DecodeIndex(benchmarkIndexJSON, id)
rtest.OK(b, err)
}
}
func BenchmarkDecodeIndexParallel(b *testing.B) {
benchmarkIndexJSONOnce.Do(initBenchmarkIndexJSON)
id := restic.NewRandomID()
b.ResetTimer()
b.RunParallel(func(pb *testing.PB) {
for pb.Next() {
_, _, err := index.DecodeIndex(benchmarkIndexJSON, id)
rtest.OK(b, err)
}
})
}
func TestIndexUnserializeOld(t *testing.T) {
idx, oldFormat, err := index.DecodeIndex(docOldExample, restic.NewRandomID())
rtest.OK(t, err)
rtest.Assert(t, oldFormat, "old index format recognized as new format")
for _, test := range exampleTests {
list := idx.Lookup(restic.BlobHandle{ID: test.id, Type: test.tpe}, nil)
if len(list) != 1 {
t.Errorf("expected one result for blob %v, got %v: %v", test.id.Str(), len(list), list)
}
blob := list[0]
rtest.Equals(t, test.packID, blob.PackID)
rtest.Equals(t, test.tpe, blob.Type)
rtest.Equals(t, test.offset, blob.Offset)
rtest.Equals(t, test.length, blob.Length)
}
rtest.Equals(t, 0, len(idx.Supersedes()))
}
func TestIndexPacks(t *testing.T) {
idx := index.NewIndex()
packs := restic.NewIDSet()
for i := 0; i < 20; i++ {
packID := restic.NewRandomID()
idx.StorePack(packID, []restic.Blob{
{
BlobHandle: restic.NewRandomBlobHandle(),
Offset: 0,
Length: 23,
},
})
packs.Insert(packID)
}
idxPacks := idx.Packs()
rtest.Assert(t, packs.Equals(idxPacks), "packs in index do not match packs added to index")
}
const maxPackSize = 16 * 1024 * 1024
// This function generates a (insecure) random ID, similar to NewRandomID
func NewRandomTestID(rng *rand.Rand) restic.ID {
id := restic.ID{}
rng.Read(id[:])
return id
}
func createRandomIndex(rng *rand.Rand, packfiles int) (idx *index.Index, lookupBh restic.BlobHandle) {
idx = index.NewIndex()
// create index with given number of pack files
for i := 0; i < packfiles; i++ {
packID := NewRandomTestID(rng)
var blobs []restic.Blob
offset := 0
for offset < maxPackSize {
size := 2000 + rng.Intn(4*1024*1024)
id := NewRandomTestID(rng)
blobs = append(blobs, restic.Blob{
BlobHandle: restic.BlobHandle{
Type: restic.DataBlob,
ID: id,
},
Length: uint(size),
UncompressedLength: uint(2 * size),
Offset: uint(offset),
})
offset += size
}
idx.StorePack(packID, blobs)
if i == 0 {
lookupBh = restic.BlobHandle{
Type: restic.DataBlob,
ID: blobs[rng.Intn(len(blobs))].ID,
}
}
}
return idx, lookupBh
}
func BenchmarkIndexHasUnknown(b *testing.B) {
idx, _ := createRandomIndex(rand.New(rand.NewSource(0)), 200000)
lookupBh := restic.NewRandomBlobHandle()
b.ResetTimer()
for i := 0; i < b.N; i++ {
idx.Has(lookupBh)
}
}
func BenchmarkIndexHasKnown(b *testing.B) {
idx, lookupBh := createRandomIndex(rand.New(rand.NewSource(0)), 200000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
idx.Has(lookupBh)
}
}
func BenchmarkIndexAlloc(b *testing.B) {
rng := rand.New(rand.NewSource(0))
b.ReportAllocs()
for i := 0; i < b.N; i++ {
createRandomIndex(rng, 200000)
}
}
func BenchmarkIndexAllocParallel(b *testing.B) {
b.ReportAllocs()
b.RunParallel(func(pb *testing.PB) {
rng := rand.New(rand.NewSource(0))
for pb.Next() {
createRandomIndex(rng, 200000)
}
})
}
func TestIndexHas(t *testing.T) {
tests := []restic.PackedBlob{}
idx := index.NewIndex()
// create 50 packs with 20 blobs each
for i := 0; i < 50; i++ {
packID := restic.NewRandomID()
var blobs []restic.Blob
pos := uint(0)
for j := 0; j < 20; j++ {
length := uint(i*100 + j)
uncompressedLength := uint(0)
if i >= 25 {
// test a mix of compressed and uncompressed packs
uncompressedLength = 2 * length
}
pb := restic.PackedBlob{
Blob: restic.Blob{
BlobHandle: restic.NewRandomBlobHandle(),
Offset: pos,
Length: length,
UncompressedLength: uncompressedLength,
},
PackID: packID,
}
blobs = append(blobs, pb.Blob)
tests = append(tests, pb)
pos += length
}
idx.StorePack(packID, blobs)
}
for _, testBlob := range tests {
rtest.Assert(t, idx.Has(testBlob.BlobHandle), "Index reports not having data blob added to it")
}
rtest.Assert(t, !idx.Has(restic.NewRandomBlobHandle()), "Index reports having a data blob not added to it")
rtest.Assert(t, !idx.Has(restic.BlobHandle{ID: tests[0].ID, Type: restic.TreeBlob}), "Index reports having a tree blob added to it with the same id as a data blob")
}
func TestMixedEachByPack(t *testing.T) {
idx := index.NewIndex()
expected := make(map[restic.ID]int)
// create 50 packs with 2 blobs each
for i := 0; i < 50; i++ {
packID := restic.NewRandomID()
expected[packID] = 1
blobs := []restic.Blob{
{
BlobHandle: restic.BlobHandle{Type: restic.DataBlob, ID: restic.NewRandomID()},
Offset: 0,
Length: 42,
},
{
BlobHandle: restic.BlobHandle{Type: restic.TreeBlob, ID: restic.NewRandomID()},
Offset: 42,
Length: 43,
},
}
idx.StorePack(packID, blobs)
}
reported := make(map[restic.ID]int)
for bp := range idx.EachByPack(context.TODO(), restic.NewIDSet()) {
reported[bp.PackID]++
rtest.Equals(t, 2, len(bp.Blobs)) // correct blob count
if bp.Blobs[0].Offset > bp.Blobs[1].Offset {
bp.Blobs[1], bp.Blobs[0] = bp.Blobs[0], bp.Blobs[1]
}
b0 := bp.Blobs[0]
rtest.Assert(t, b0.Type == restic.DataBlob && b0.Offset == 0 && b0.Length == 42, "wrong blob", b0)
b1 := bp.Blobs[1]
rtest.Assert(t, b1.Type == restic.TreeBlob && b1.Offset == 42 && b1.Length == 43, "wrong blob", b1)
}
rtest.Equals(t, expected, reported)
}
func TestEachByPackIgnoes(t *testing.T) {
idx := index.NewIndex()
ignores := restic.NewIDSet()
expected := make(map[restic.ID]int)
// create 50 packs with one blob each
for i := 0; i < 50; i++ {
packID := restic.NewRandomID()
if i < 3 {
ignores.Insert(packID)
} else {
expected[packID] = 1
}
blobs := []restic.Blob{
{
BlobHandle: restic.BlobHandle{Type: restic.DataBlob, ID: restic.NewRandomID()},
Offset: 0,
Length: 42,
},
}
idx.StorePack(packID, blobs)
}
idx.Finalize()
reported := make(map[restic.ID]int)
for bp := range idx.EachByPack(context.TODO(), ignores) {
reported[bp.PackID]++
rtest.Equals(t, 1, len(bp.Blobs)) // correct blob count
b0 := bp.Blobs[0]
rtest.Assert(t, b0.Type == restic.DataBlob && b0.Offset == 0 && b0.Length == 42, "wrong blob", b0)
}
rtest.Equals(t, expected, reported)
}

170
internal/index/indexmap.go Normal file
View File

@@ -0,0 +1,170 @@
package index
import (
"hash/maphash"
"github.com/restic/restic/internal/restic"
)
// An indexMap is a chained hash table that maps blob IDs to indexEntries.
// It allows storing multiple entries with the same key.
//
// IndexMap uses some optimizations that are not compatible with supporting
// deletions.
//
// The buckets in this hash table contain only pointers, rather than inlined
// key-value pairs like the standard Go map. This way, only a pointer array
// needs to be resized when the table grows, preventing memory usage spikes.
type indexMap struct {
// The number of buckets is always a power of two and never zero.
buckets []*indexEntry
numentries uint
mh maphash.Hash
free *indexEntry // Free list.
}
const (
growthFactor = 2 // Must be a power of 2.
maxLoad = 4 // Max. number of entries per bucket.
)
// add inserts an indexEntry for the given arguments into the map,
// using id as the key.
func (m *indexMap) add(id restic.ID, packIdx int, offset, length uint32, uncompressedLength uint32) {
switch {
case m.numentries == 0: // Lazy initialization.
m.init()
case m.numentries >= maxLoad*uint(len(m.buckets)):
m.grow()
}
h := m.hash(id)
e := m.newEntry()
e.id = id
e.next = m.buckets[h] // Prepend to existing chain.
e.packIndex = packIdx
e.offset = offset
e.length = length
e.uncompressedLength = uncompressedLength
m.buckets[h] = e
m.numentries++
}
// foreach calls fn for all entries in the map, until fn returns false.
func (m *indexMap) foreach(fn func(*indexEntry) bool) {
for _, e := range m.buckets {
for e != nil {
if !fn(e) {
return
}
e = e.next
}
}
}
// foreachWithID calls fn for all entries with the given id.
func (m *indexMap) foreachWithID(id restic.ID, fn func(*indexEntry)) {
if len(m.buckets) == 0 {
return
}
h := m.hash(id)
for e := m.buckets[h]; e != nil; e = e.next {
if e.id != id {
continue
}
fn(e)
}
}
// get returns the first entry for the given id.
func (m *indexMap) get(id restic.ID) *indexEntry {
if len(m.buckets) == 0 {
return nil
}
h := m.hash(id)
for e := m.buckets[h]; e != nil; e = e.next {
if e.id == id {
return e
}
}
return nil
}
func (m *indexMap) grow() {
old := m.buckets
m.buckets = make([]*indexEntry, growthFactor*len(m.buckets))
for _, e := range old {
for e != nil {
h := m.hash(e.id)
next := e.next
e.next = m.buckets[h]
m.buckets[h] = e
e = next
}
}
}
func (m *indexMap) hash(id restic.ID) uint {
// We use maphash to prevent backups of specially crafted inputs
// from degrading performance.
// While SHA-256 should be collision-resistant, for hash table indices
// we use only a few bits of it and finding collisions for those is
// much easier than breaking the whole algorithm.
m.mh.Reset()
_, _ = m.mh.Write(id[:])
h := uint(m.mh.Sum64())
return h & uint(len(m.buckets)-1)
}
func (m *indexMap) init() {
const initialBuckets = 64
m.buckets = make([]*indexEntry, initialBuckets)
}
func (m *indexMap) len() uint { return m.numentries }
func (m *indexMap) newEntry() *indexEntry {
// We keep a free list of objects to speed up allocation and GC.
// There's an obvious trade-off here: allocating in larger batches
// means we allocate faster and the GC has to keep fewer bits to track
// what we have in use, but it means we waste some space.
//
// Then again, allocating each indexEntry separately also wastes space
// on 32-bit platforms, because the Go malloc has no size class for
// exactly 52 bytes, so it puts the indexEntry in a 64-byte slot instead.
// See src/runtime/sizeclasses.go in the Go source repo.
//
// The batch size of 4 means we hit the size classes for 4×64=256 bytes
// (64-bit) and 4×52=208 bytes (32-bit), wasting nothing in malloc on
// 64-bit and relatively little on 32-bit.
const entryAllocBatch = 4
e := m.free
if e != nil {
m.free = e.next
} else {
free := new([entryAllocBatch]indexEntry)
e = &free[0]
for i := 1; i < len(free)-1; i++ {
free[i].next = &free[i+1]
}
m.free = &free[1]
}
return e
}
type indexEntry struct {
id restic.ID
next *indexEntry
packIndex int // Position in containing Index's packs field.
offset uint32
length uint32
uncompressedLength uint32
}

130
internal/index/indexmap_test.go Normal file
View File

@@ -0,0 +1,130 @@
package index
import (
"math/rand"
"testing"
"time"
"github.com/restic/restic/internal/restic"
rtest "github.com/restic/restic/internal/test"
)
func TestIndexMapBasic(t *testing.T) {
t.Parallel()
var (
id restic.ID
m indexMap
r = rand.New(rand.NewSource(98765))
)
for i := 1; i <= 400; i++ {
r.Read(id[:])
rtest.Assert(t, m.get(id) == nil, "%v retrieved but not added", id)
m.add(id, 0, 0, 0, 0)
rtest.Assert(t, m.get(id) != nil, "%v added but not retrieved", id)
rtest.Equals(t, uint(i), m.len())
}
}
func TestIndexMapForeach(t *testing.T) {
t.Parallel()
const N = 10
var m indexMap
// Don't crash on empty map.
m.foreach(func(*indexEntry) bool { return true })
for i := 0; i < N; i++ {
var id restic.ID
id[0] = byte(i)
m.add(id, i, uint32(i), uint32(i), uint32(i/2))
}
seen := make(map[int]struct{})
m.foreach(func(e *indexEntry) bool {
i := int(e.id[0])
rtest.Assert(t, i < N, "unknown id %v in indexMap", e.id)
rtest.Equals(t, i, e.packIndex)
rtest.Equals(t, i, int(e.length))
rtest.Equals(t, i, int(e.offset))
rtest.Equals(t, i/2, int(e.uncompressedLength))
seen[i] = struct{}{}
return true
})
rtest.Equals(t, N, len(seen))
ncalls := 0
m.foreach(func(*indexEntry) bool {
ncalls++
return false
})
rtest.Equals(t, 1, ncalls)
}
func TestIndexMapForeachWithID(t *testing.T) {
t.Parallel()
const ndups = 3
var (
id restic.ID
m indexMap
r = rand.New(rand.NewSource(1234321))
)
r.Read(id[:])
// No result (and no crash) for empty map.
n := 0
m.foreachWithID(id, func(*indexEntry) { n++ })
rtest.Equals(t, 0, n)
// Test insertion and retrieval of duplicates.
for i := 0; i < ndups; i++ {
m.add(id, i, 0, 0, 0)
}
for i := 0; i < 100; i++ {
var otherid restic.ID
r.Read(otherid[:])
m.add(otherid, -1, 0, 0, 0)
}
n = 0
var packs [ndups]bool
m.foreachWithID(id, func(e *indexEntry) {
packs[e.packIndex] = true
n++
})
rtest.Equals(t, ndups, n)
for i := range packs {
rtest.Assert(t, packs[i], "duplicate from pack %d not retrieved", i)
}
}
func BenchmarkIndexMapHash(b *testing.B) {
var m indexMap
m.add(restic.ID{}, 0, 0, 0, 0) // Trigger lazy initialization.
ids := make([]restic.ID, 128) // 4 KiB.
r := rand.New(rand.NewSource(time.Now().UnixNano()))
for i := range ids {
r.Read(ids[i][:])
}
b.ReportAllocs()
b.SetBytes(int64(len(restic.ID{}) * len(ids)))
b.ResetTimer()
for i := 0; i < b.N; i++ {
for _, id := range ids {
m.hash(id)
}
}
}

445
internal/index/master_index.go Normal file
View File

@@ -0,0 +1,445 @@
package index
import (
"bytes"
"context"
"fmt"
"runtime"
"sync"
"github.com/restic/restic/internal/debug"
"github.com/restic/restic/internal/restic"
"github.com/restic/restic/internal/ui/progress"
"golang.org/x/sync/errgroup"
)
// MasterIndex is a collection of indexes and IDs of chunks that are in the process of being saved.
type MasterIndex struct {
idx []*Index
pendingBlobs restic.BlobSet
idxMutex sync.RWMutex
compress bool
}
// NewMasterIndex creates a new master index.
func NewMasterIndex() *MasterIndex {
// Always add an empty final index, such that MergeFinalIndexes can merge into this.
// Note that removing this index could lead to a race condition in the rare
// sitation that only two indexes exist which are saved and merged concurrently.
idx := []*Index{NewIndex()}
idx[0].Finalize()
return &MasterIndex{idx: idx, pendingBlobs: restic.NewBlobSet()}
}
func (mi *MasterIndex) MarkCompressed() {
mi.compress = true
}
// Lookup queries all known Indexes for the ID and returns all matches.
func (mi *MasterIndex) Lookup(bh restic.BlobHandle) (pbs []restic.PackedBlob) {
mi.idxMutex.RLock()
defer mi.idxMutex.RUnlock()
for _, idx := range mi.idx {
pbs = idx.Lookup(bh, pbs)
}
return pbs
}
// LookupSize queries all known Indexes for the ID and returns the first match.
func (mi *MasterIndex) LookupSize(bh restic.BlobHandle) (uint, bool) {
mi.idxMutex.RLock()
defer mi.idxMutex.RUnlock()
for _, idx := range mi.idx {
if size, found := idx.LookupSize(bh); found {
return size, found
}
}
return 0, false
}
// AddPending adds a given blob to list of pending Blobs
// Before doing so it checks if this blob is already known.
// Returns true if adding was successful and false if the blob
// was already known
func (mi *MasterIndex) AddPending(bh restic.BlobHandle) bool {
mi.idxMutex.Lock()
defer mi.idxMutex.Unlock()
// Check if blob is pending or in index
if mi.pendingBlobs.Has(bh) {
return false
}
for _, idx := range mi.idx {
if idx.Has(bh) {
return false
}
}
// really not known -> insert
mi.pendingBlobs.Insert(bh)
return true
}
// Has queries all known Indexes for the ID and returns the first match.
// Also returns true if the ID is pending.
func (mi *MasterIndex) Has(bh restic.BlobHandle) bool {
mi.idxMutex.RLock()
defer mi.idxMutex.RUnlock()
// also return true if blob is pending
if mi.pendingBlobs.Has(bh) {
return true
}
for _, idx := range mi.idx {
if idx.Has(bh) {
return true
}
}
return false
}
// IDs returns the IDs of all indexes contained in the index.
func (mi *MasterIndex) IDs() restic.IDSet {
mi.idxMutex.RLock()
defer mi.idxMutex.RUnlock()
ids := restic.NewIDSet()
for _, idx := range mi.idx {
if !idx.Final() {
continue
}
indexIDs, err := idx.IDs()
if err != nil {
debug.Log("not using index, ID() returned error %v", err)
continue
}
for _, id := range indexIDs {
ids.Insert(id)
}
}
return ids
}
// Packs returns all packs that are covered by the index.
// If packBlacklist is given, those packs are only contained in the
// resulting IDSet if they are contained in a non-final (newly written) index.
func (mi *MasterIndex) Packs(packBlacklist restic.IDSet) restic.IDSet {
mi.idxMutex.RLock()
defer mi.idxMutex.RUnlock()
packs := restic.NewIDSet()
for _, idx := range mi.idx {
idxPacks := idx.Packs()
if idx.final && len(packBlacklist) > 0 {
idxPacks = idxPacks.Sub(packBlacklist)
}
packs.Merge(idxPacks)
}
return packs
}
// Insert adds a new index to the MasterIndex.
func (mi *MasterIndex) Insert(idx *Index) {
mi.idxMutex.Lock()
defer mi.idxMutex.Unlock()
mi.idx = append(mi.idx, idx)
}
// StorePack remembers the id and pack in the index.
func (mi *MasterIndex) StorePack(id restic.ID, blobs []restic.Blob) {
mi.idxMutex.Lock()
defer mi.idxMutex.Unlock()
// delete blobs from pending
for _, blob := range blobs {
mi.pendingBlobs.Delete(restic.BlobHandle{Type: blob.Type, ID: blob.ID})
}
for _, idx := range mi.idx {
if !idx.Final() {
idx.StorePack(id, blobs)
return
}
}
newIdx := NewIndex()
newIdx.StorePack(id, blobs)
mi.idx = append(mi.idx, newIdx)
}
// finalizeNotFinalIndexes finalizes all indexes that
// have not yet been saved and returns that list
func (mi *MasterIndex) finalizeNotFinalIndexes() []*Index {
mi.idxMutex.Lock()
defer mi.idxMutex.Unlock()
var list []*Index
for _, idx := range mi.idx {
if !idx.Final() {
idx.Finalize()
list = append(list, idx)
}
}
debug.Log("return %d indexes", len(list))
return list
}
// finalizeFullIndexes finalizes all indexes that are full and returns that list.
func (mi *MasterIndex) finalizeFullIndexes() []*Index {
mi.idxMutex.Lock()
defer mi.idxMutex.Unlock()
var list []*Index
debug.Log("checking %d indexes", len(mi.idx))
for _, idx := range mi.idx {
if idx.Final() {
continue
}
if IndexFull(idx, mi.compress) {
debug.Log("index %p is full", idx)
idx.Finalize()
list = append(list, idx)
} else {
debug.Log("index %p not full", idx)
}
}
debug.Log("return %d indexes", len(list))
return list
}
// Each runs fn on all blobs known to the index. When the context is cancelled,
// the index iteration return immediately. This blocks any modification of the index.
func (mi *MasterIndex) Each(ctx context.Context, fn func(restic.PackedBlob)) {
mi.idxMutex.RLock()
defer mi.idxMutex.RUnlock()
for _, idx := range mi.idx {
idx.Each(ctx, fn)
}
}
// MergeFinalIndexes merges all final indexes together.
// After calling, there will be only one big final index in MasterIndex
// containing all final index contents.
// Indexes that are not final are left untouched.
// This merging can only be called after all index files are loaded - as
// removing of superseded index contents is only possible for unmerged indexes.
func (mi *MasterIndex) MergeFinalIndexes() error {
mi.idxMutex.Lock()
defer mi.idxMutex.Unlock()
// The first index is always final and the one to merge into
newIdx := mi.idx[:1]
for i := 1; i < len(mi.idx); i++ {
idx := mi.idx[i]
// clear reference in masterindex as it may become stale
mi.idx[i] = nil
// do not merge indexes that have no id set
ids, _ := idx.IDs()
if !idx.Final() || len(ids) == 0 {
newIdx = append(newIdx, idx)
} else {
err := mi.idx[0].merge(idx)
if err != nil {
return fmt.Errorf("MergeFinalIndexes: %w", err)
}
}
}
mi.idx = newIdx
return nil
}
// Save saves all known indexes to index files, leaving out any
// packs whose ID is contained in packBlacklist from finalized indexes.
// The new index contains the IDs of all known indexes in the "supersedes"
// field. The IDs are also returned in the IDSet obsolete.
// After calling this function, you should remove the obsolete index files.
func (mi *MasterIndex) Save(ctx context.Context, repo restic.SaverUnpacked, packBlacklist restic.IDSet, extraObsolete restic.IDs, p *progress.Counter) (obsolete restic.IDSet, err error) {
p.SetMax(uint64(len(mi.Packs(packBlacklist))))
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()
// track spawned goroutines using wg, create a new context which is
// cancelled as soon as an error occurs.
wg, ctx := errgroup.WithContext(ctx)
ch := make(chan *Index)
wg.Go(func() error {
defer close(ch)
for i, idx := range mi.idx {
if idx.Final() {
ids, err := idx.IDs()
if err != nil {
debug.Log("index %d does not have an ID: %v", err)
return err
}
debug.Log("adding index ids %v to supersedes field", ids)
err = newIndex.AddToSupersedes(ids...)
if err != nil {
return err
}
obsolete.Merge(restic.NewIDSet(ids...))
} else {
debug.Log("index %d isn't final, don't add to supersedes field", i)
}
debug.Log("adding index %d", i)
for pbs := range idx.EachByPack(ctx, packBlacklist) {
newIndex.StorePack(pbs.PackID, pbs.Blobs)
p.Add(1)
if IndexFull(newIndex, mi.compress) {
select {
case ch <- newIndex:
case <-ctx.Done():
return ctx.Err()
}
newIndex = NewIndex()
}
}
}
err = newIndex.AddToSupersedes(extraObsolete...)
if err != nil {
return err
}
obsolete.Merge(restic.NewIDSet(extraObsolete...))
select {
case ch <- newIndex:
case <-ctx.Done():
}
return nil
})
// a worker receives an index from ch, and saves the index
worker := func() error {
for idx := range ch {
idx.Finalize()
if _, err := SaveIndex(ctx, repo, idx); err != nil {
return err
}
}
return nil
}
// encoding an index can take quite some time such that this can be both CPU- or IO-bound
workerCount := int(repo.Connections()) + runtime.GOMAXPROCS(0)
// run workers on ch
for i := 0; i < workerCount; i++ {
wg.Go(worker)
}
err = wg.Wait()
return obsolete, err
}
// SaveIndex saves an index in the repository.
func SaveIndex(ctx context.Context, repo restic.SaverUnpacked, index *Index) (restic.ID, error) {
buf := bytes.NewBuffer(nil)
err := index.Encode(buf)
if err != nil {
return restic.ID{}, err
}
id, err := repo.SaveUnpacked(ctx, restic.IndexFile, buf.Bytes())
ierr := index.SetID(id)
if ierr != nil {
// logic bug
panic(ierr)
}
return id, err
}
// saveIndex saves all indexes in the backend.
func (mi *MasterIndex) saveIndex(ctx context.Context, r restic.SaverUnpacked, indexes ...*Index) error {
for i, idx := range indexes {
debug.Log("Saving index %d", i)
sid, err := SaveIndex(ctx, r, idx)
if err != nil {
return err
}
debug.Log("Saved index %d as %v", i, sid)
}
return mi.MergeFinalIndexes()
}
// SaveIndex saves all new indexes in the backend.
func (mi *MasterIndex) SaveIndex(ctx context.Context, r restic.SaverUnpacked) error {
return mi.saveIndex(ctx, r, mi.finalizeNotFinalIndexes()...)
}
// SaveFullIndex saves all full indexes in the backend.
func (mi *MasterIndex) SaveFullIndex(ctx context.Context, r restic.SaverUnpacked) error {
return mi.saveIndex(ctx, r, mi.finalizeFullIndexes()...)
}
// ListPacks returns the blobs of the specified pack files grouped by pack file.
func (mi *MasterIndex) ListPacks(ctx context.Context, packs restic.IDSet) <-chan restic.PackBlobs {
out := make(chan restic.PackBlobs)
go func() {
defer close(out)
// only resort a part of the index to keep the memory overhead bounded
for i := byte(0); i < 16; i++ {
if ctx.Err() != nil {
return
}
packBlob := make(map[restic.ID][]restic.Blob)
for pack := range packs {
if pack[0]&0xf == i {
packBlob[pack] = nil
}
}
if len(packBlob) == 0 {
continue
}
mi.Each(ctx, func(pb restic.PackedBlob) {
if packs.Has(pb.PackID) && pb.PackID[0]&0xf == i {
packBlob[pb.PackID] = append(packBlob[pb.PackID], pb.Blob)
}
})
// pass on packs
for packID, pbs := range packBlob {
// allow GC
packBlob[packID] = nil
select {
case out <- restic.PackBlobs{PackID: packID, Blobs: pbs}:
case <-ctx.Done():
return
}
}
}
}()
return out
}

398
internal/index/master_index_test.go Normal file
View File

@@ -0,0 +1,398 @@
package index_test
import (
"context"
"fmt"
"math/rand"
"testing"
"time"
"github.com/restic/restic/internal/checker"
"github.com/restic/restic/internal/crypto"
"github.com/restic/restic/internal/index"
"github.com/restic/restic/internal/repository"
"github.com/restic/restic/internal/restic"
rtest "github.com/restic/restic/internal/test"
)
func TestMasterIndex(t *testing.T) {
bhInIdx1 := restic.NewRandomBlobHandle()
bhInIdx2 := restic.NewRandomBlobHandle()
bhInIdx12 := restic.BlobHandle{ID: restic.NewRandomID(), Type: restic.TreeBlob}
blob1 := restic.PackedBlob{
PackID: restic.NewRandomID(),
Blob: restic.Blob{
BlobHandle: bhInIdx1,
Length: uint(crypto.CiphertextLength(10)),
Offset: 0,
},
}
blob2 := restic.PackedBlob{
PackID: restic.NewRandomID(),
Blob: restic.Blob{
BlobHandle: bhInIdx2,
Length: uint(crypto.CiphertextLength(100)),
Offset: 10,
UncompressedLength: 200,
},
}
blob12a := restic.PackedBlob{
PackID: restic.NewRandomID(),
Blob: restic.Blob{
BlobHandle: bhInIdx12,
Length: uint(crypto.CiphertextLength(123)),
Offset: 110,
},
}
blob12b := restic.PackedBlob{
PackID: restic.NewRandomID(),
Blob: restic.Blob{
BlobHandle: bhInIdx12,
Length: uint(crypto.CiphertextLength(123)),
Offset: 50,
UncompressedLength: 80,
},
}
idx1 := index.NewIndex()
idx1.StorePack(blob1.PackID, []restic.Blob{blob1.Blob})
idx1.StorePack(blob12a.PackID, []restic.Blob{blob12a.Blob})
idx2 := index.NewIndex()
idx2.StorePack(blob2.PackID, []restic.Blob{blob2.Blob})
idx2.StorePack(blob12b.PackID, []restic.Blob{blob12b.Blob})
mIdx := index.NewMasterIndex()
mIdx.Insert(idx1)
mIdx.Insert(idx2)
// test idInIdx1
found := mIdx.Has(bhInIdx1)
rtest.Equals(t, true, found)
blobs := mIdx.Lookup(bhInIdx1)
rtest.Equals(t, []restic.PackedBlob{blob1}, blobs)
size, found := mIdx.LookupSize(bhInIdx1)
rtest.Equals(t, true, found)
rtest.Equals(t, uint(10), size)
// test idInIdx2
found = mIdx.Has(bhInIdx2)
rtest.Equals(t, true, found)
blobs = mIdx.Lookup(bhInIdx2)
rtest.Equals(t, []restic.PackedBlob{blob2}, blobs)
size, found = mIdx.LookupSize(bhInIdx2)
rtest.Equals(t, true, found)
rtest.Equals(t, uint(200), size)
// test idInIdx12
found = mIdx.Has(bhInIdx12)
rtest.Equals(t, true, found)
blobs = mIdx.Lookup(bhInIdx12)
rtest.Equals(t, 2, len(blobs))
// test Lookup result for blob12a
found = false
if blobs[0] == blob12a || blobs[1] == blob12a {
found = true
}
rtest.Assert(t, found, "blob12a not found in result")
// test Lookup result for blob12b
found = false
if blobs[0] == blob12b || blobs[1] == blob12b {
found = true
}
rtest.Assert(t, found, "blob12a not found in result")
size, found = mIdx.LookupSize(bhInIdx12)
rtest.Equals(t, true, found)
rtest.Equals(t, uint(123), size)
// test not in index
found = mIdx.Has(restic.BlobHandle{ID: restic.NewRandomID(), Type: restic.TreeBlob})
rtest.Assert(t, !found, "Expected no blobs when fetching with a random id")
blobs = mIdx.Lookup(restic.NewRandomBlobHandle())
rtest.Assert(t, blobs == nil, "Expected no blobs when fetching with a random id")
_, found = mIdx.LookupSize(restic.NewRandomBlobHandle())
rtest.Assert(t, !found, "Expected no blobs when fetching with a random id")
}
func TestMasterMergeFinalIndexes(t *testing.T) {
bhInIdx1 := restic.NewRandomBlobHandle()
bhInIdx2 := restic.NewRandomBlobHandle()
blob1 := restic.PackedBlob{
PackID: restic.NewRandomID(),
Blob: restic.Blob{
BlobHandle: bhInIdx1,
Length: 10,
Offset: 0,
},
}
blob2 := restic.PackedBlob{
PackID: restic.NewRandomID(),
Blob: restic.Blob{
BlobHandle: bhInIdx2,
Length: 100,
Offset: 10,
UncompressedLength: 200,
},
}
idx1 := index.NewIndex()
idx1.StorePack(blob1.PackID, []restic.Blob{blob1.Blob})
idx2 := index.NewIndex()
idx2.StorePack(blob2.PackID, []restic.Blob{blob2.Blob})
mIdx := index.NewMasterIndex()
mIdx.Insert(idx1)
mIdx.Insert(idx2)
finalIndexes, idxCount := index.TestMergeIndex(t, mIdx)
rtest.Equals(t, []*index.Index{idx1, idx2}, finalIndexes)
rtest.Equals(t, 1, idxCount)
blobCount := 0
mIdx.Each(context.TODO(), func(pb restic.PackedBlob) {
blobCount++
})
rtest.Equals(t, 2, blobCount)
blobs := mIdx.Lookup(bhInIdx1)
rtest.Equals(t, []restic.PackedBlob{blob1}, blobs)
blobs = mIdx.Lookup(bhInIdx2)
rtest.Equals(t, []restic.PackedBlob{blob2}, blobs)
blobs = mIdx.Lookup(restic.NewRandomBlobHandle())
rtest.Assert(t, blobs == nil, "Expected no blobs when fetching with a random id")
// merge another index containing identical blobs
idx3 := index.NewIndex()
idx3.StorePack(blob1.PackID, []restic.Blob{blob1.Blob})
idx3.StorePack(blob2.PackID, []restic.Blob{blob2.Blob})
mIdx.Insert(idx3)
finalIndexes, idxCount = index.TestMergeIndex(t, mIdx)
rtest.Equals(t, []*index.Index{idx3}, finalIndexes)
rtest.Equals(t, 1, idxCount)
// Index should have same entries as before!
blobs = mIdx.Lookup(bhInIdx1)
rtest.Equals(t, []restic.PackedBlob{blob1}, blobs)
blobs = mIdx.Lookup(bhInIdx2)
rtest.Equals(t, []restic.PackedBlob{blob2}, blobs)
blobCount = 0
mIdx.Each(context.TODO(), func(pb restic.PackedBlob) {
blobCount++
})
rtest.Equals(t, 2, blobCount)
}
func createRandomMasterIndex(t testing.TB, rng *rand.Rand, num, size int) (*index.MasterIndex, restic.BlobHandle) {
mIdx := index.NewMasterIndex()
for i := 0; i < num-1; i++ {
idx, _ := createRandomIndex(rng, size)
mIdx.Insert(idx)
}
idx1, lookupBh := createRandomIndex(rng, size)
mIdx.Insert(idx1)
index.TestMergeIndex(t, mIdx)
return mIdx, lookupBh
}
func BenchmarkMasterIndexAlloc(b *testing.B) {
rng := rand.New(rand.NewSource(0))
b.ReportAllocs()
for i := 0; i < b.N; i++ {
createRandomMasterIndex(b, rng, 10000, 5)
}
}
func BenchmarkMasterIndexLookupSingleIndex(b *testing.B) {
mIdx, lookupBh := createRandomMasterIndex(b, rand.New(rand.NewSource(0)), 1, 200000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
mIdx.Lookup(lookupBh)
}
}
func BenchmarkMasterIndexLookupMultipleIndex(b *testing.B) {
mIdx, lookupBh := createRandomMasterIndex(b, rand.New(rand.NewSource(0)), 100, 10000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
mIdx.Lookup(lookupBh)
}
}
func BenchmarkMasterIndexLookupSingleIndexUnknown(b *testing.B) {
lookupBh := restic.NewRandomBlobHandle()
mIdx, _ := createRandomMasterIndex(b, rand.New(rand.NewSource(0)), 1, 200000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
mIdx.Lookup(lookupBh)
}
}
func BenchmarkMasterIndexLookupMultipleIndexUnknown(b *testing.B) {
lookupBh := restic.NewRandomBlobHandle()
mIdx, _ := createRandomMasterIndex(b, rand.New(rand.NewSource(0)), 100, 10000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
mIdx.Lookup(lookupBh)
}
}
func BenchmarkMasterIndexLookupParallel(b *testing.B) {
for _, numindices := range []int{25, 50, 100} {
var lookupBh restic.BlobHandle
b.StopTimer()
rng := rand.New(rand.NewSource(0))
mIdx, lookupBh := createRandomMasterIndex(b, rng, numindices, 10000)
b.StartTimer()
name := fmt.Sprintf("known,indices=%d", numindices)
b.Run(name, func(b *testing.B) {
b.RunParallel(func(pb *testing.PB) {
for pb.Next() {
mIdx.Lookup(lookupBh)
}
})
})
lookupBh = restic.NewRandomBlobHandle()
name = fmt.Sprintf("unknown,indices=%d", numindices)
b.Run(name, func(b *testing.B) {
b.RunParallel(func(pb *testing.PB) {
for pb.Next() {
mIdx.Lookup(lookupBh)
}
})
})
}
}
func BenchmarkMasterIndexLookupBlobSize(b *testing.B) {
rng := rand.New(rand.NewSource(0))
mIdx, lookupBh := createRandomMasterIndex(b, rand.New(rng), 5, 200000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
mIdx.LookupSize(lookupBh)
}
}
func BenchmarkMasterIndexEach(b *testing.B) {
rng := rand.New(rand.NewSource(0))
mIdx, _ := createRandomMasterIndex(b, rand.New(rng), 5, 200000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
entries := 0
mIdx.Each(context.TODO(), func(pb restic.PackedBlob) {
entries++
})
}
}
var (
snapshotTime = time.Unix(1470492820, 207401672)
depth = 3
)
func createFilledRepo(t testing.TB, snapshots int, dup float32, version uint) (restic.Repository, func()) {
repo, cleanup := repository.TestRepositoryWithVersion(t, version)
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) {
repository.TestAllVersions(t, testIndexSave)
}
func testIndexSave(t *testing.T, version uint) {
repo, cleanup := createFilledRepo(t, 3, 0, version)
defer cleanup()
err := repo.LoadIndex(context.TODO())
if err != nil {
t.Fatal(err)
}
obsoletes, err := repo.Index().Save(context.TODO(), repo, nil, 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, false)
err = checker.LoadSnapshots(context.TODO())
if err != nil {
t.Error(err)
}
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, nil, 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
}
}
}

18
internal/index/testing.go Normal file
View File

@@ -0,0 +1,18 @@
package index
import (
"testing"
"github.com/restic/restic/internal/restic"
"github.com/restic/restic/internal/test"
)
func TestMergeIndex(t testing.TB, mi *MasterIndex) ([]*Index, int) {
finalIndexes := mi.finalizeNotFinalIndexes()
for _, idx := range finalIndexes {
test.OK(t, idx.SetID(restic.NewRandomID()))
}
test.OK(t, mi.MergeFinalIndexes())
return finalIndexes, len(mi.idx)
}