frostfs-s3-gw/api/layer/object.go

1261 lines
33 KiB
Go
Raw Normal View History

package layer
import (
"bytes"
"context"
"crypto/md5"
"crypto/sha256"
"encoding/base64"
"encoding/hex"
"encoding/json"
"errors"
"fmt"
"io"
"mime"
"path/filepath"
"runtime"
"sort"
"strconv"
"strings"
"sync"
"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api"
"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/auth"
"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/cache"
"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/data"
apiErrors "git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/errors"
"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/middleware"
"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/internal/logs"
"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/client"
cid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/container/id"
"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object"
oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id"
"github.com/minio/sio"
"github.com/panjf2000/ants/v2"
"go.uber.org/zap"
)
type (
getParams struct {
// payload range
off, ln uint64
objInfo *data.ObjectInfo
bktInfo *data.BucketInfo
}
getFrostFSParams struct {
// payload range
off, ln uint64
oid oid.ID
bktInfo *data.BucketInfo
}
// ListObjectsParamsCommon contains common parameters for ListObjectsV1 and ListObjectsV2.
ListObjectsParamsCommon struct {
BktInfo *data.BucketInfo
Delimiter string
Encode string
MaxKeys int
Prefix string
}
// ListObjectsParamsV1 contains params for ListObjectsV1.
ListObjectsParamsV1 struct {
ListObjectsParamsCommon
Marker string
}
// ListObjectsParamsV2 contains params for ListObjectsV2.
ListObjectsParamsV2 struct {
ListObjectsParamsCommon
ContinuationToken string
StartAfter string
FetchOwner bool
}
allObjectParams struct {
Bucket *data.BucketInfo
Delimiter string
Prefix string
MaxKeys int
Marker string
ContinuationToken string
}
DeleteMarkerError struct {
ErrorCode apiErrors.ErrorCode
}
)
func (e DeleteMarkerError) Error() string {
return "object is delete marker"
}
const (
continuationToken = "<continuation-token>"
)
func newAddress(cnr cid.ID, obj oid.ID) oid.Address {
var addr oid.Address
addr.SetContainer(cnr)
addr.SetObject(obj)
return addr
}
// objectHead returns all object's headers.
func (n *layer) objectHead(ctx context.Context, bktInfo *data.BucketInfo, idObj oid.ID) (*object.Object, error) {
prm := PrmObjectRead{
Container: bktInfo.CID,
Object: idObj,
WithHeader: true,
}
n.prepareAuthParameters(ctx, &prm.PrmAuth, bktInfo.Owner)
res, err := n.frostFS.ReadObject(ctx, prm)
if err != nil {
return nil, err
}
return res.Head, nil
}
func (n *layer) initObjectPayloadReader(ctx context.Context, p getParams) (io.Reader, error) {
if _, isCombined := p.objInfo.Headers[MultipartObjectSize]; !isCombined {
return n.initFrostFSObjectPayloadReader(ctx, getFrostFSParams{
off: p.off,
ln: p.ln,
oid: p.objInfo.ID,
bktInfo: p.bktInfo,
})
}
combinedObj, err := n.objectGet(ctx, p.bktInfo, p.objInfo.ID)
if err != nil {
return nil, fmt.Errorf("get combined object '%s': %w", p.objInfo.ID.EncodeToString(), err)
}
var parts []*data.PartInfo
if err = json.Unmarshal(combinedObj.Payload(), &parts); err != nil {
return nil, fmt.Errorf("unmarshal combined object parts: %w", err)
}
isEncrypted := FormEncryptionInfo(p.objInfo.Headers).Enabled
objParts := make([]partObj, len(parts))
for i, part := range parts {
size := part.Size
if isEncrypted {
if size, err = sio.EncryptedSize(part.Size); err != nil {
return nil, fmt.Errorf("compute encrypted size: %w", err)
}
}
objParts[i] = partObj{
OID: part.OID,
Size: size,
}
}
return newMultiObjectReader(ctx, multiObjectReaderConfig{
layer: n,
off: p.off,
ln: p.ln,
parts: objParts,
bktInfo: p.bktInfo,
})
}
// initializes payload reader of the FrostFS object.
// Zero range corresponds to full payload (panics if only offset is set).
func (n *layer) initFrostFSObjectPayloadReader(ctx context.Context, p getFrostFSParams) (io.Reader, error) {
prm := PrmObjectRead{
Container: p.bktInfo.CID,
Object: p.oid,
WithPayload: true,
PayloadRange: [2]uint64{p.off, p.ln},
}
n.prepareAuthParameters(ctx, &prm.PrmAuth, p.bktInfo.Owner)
res, err := n.frostFS.ReadObject(ctx, prm)
if err != nil {
return nil, err
}
return res.Payload, nil
}
// objectGet returns an object with payload in the object.
func (n *layer) objectGet(ctx context.Context, bktInfo *data.BucketInfo, objID oid.ID) (*object.Object, error) {
prm := PrmObjectRead{
Container: bktInfo.CID,
Object: objID,
WithHeader: true,
WithPayload: true,
}
n.prepareAuthParameters(ctx, &prm.PrmAuth, bktInfo.Owner)
res, err := n.frostFS.ReadObject(ctx, prm)
if err != nil {
return nil, err
}
return res.Head, nil
}
// MimeByFilePath detect mime type by file path extension.
func MimeByFilePath(path string) string {
ext := filepath.Ext(path)
if len(ext) == 0 {
return ""
}
return mime.TypeByExtension(ext)
}
func encryptionReader(r io.Reader, size uint64, key []byte) (io.Reader, uint64, error) {
encSize, err := sio.EncryptedSize(size)
if err != nil {
return nil, 0, fmt.Errorf("failed to compute enc size: %w", err)
}
r, err = sio.EncryptReader(r, sio.Config{MinVersion: sio.Version20, MaxVersion: sio.Version20, Key: key, CipherSuites: []byte{sio.AES_256_GCM}})
if err != nil {
return nil, 0, fmt.Errorf("couldn't create encrypter: %w", err)
}
return r, encSize, nil
}
func ParseCompletedPartHeader(hdr string) (*Part, error) {
// partInfo[0] -- part number, partInfo[1] -- part size, partInfo[2] -- checksum
partInfo := strings.Split(hdr, "-")
if len(partInfo) != 3 {
return nil, fmt.Errorf("invalid completed part header")
}
num, err := strconv.Atoi(partInfo[0])
if err != nil {
return nil, fmt.Errorf("invalid completed part number '%s': %w", partInfo[0], err)
}
size, err := strconv.ParseUint(partInfo[1], 10, 64)
if err != nil {
return nil, fmt.Errorf("invalid completed part size '%s': %w", partInfo[1], err)
}
return &Part{
ETag: partInfo[2],
PartNumber: num,
Size: size,
}, nil
}
// PutObject stores object into FrostFS, took payload from io.Reader.
func (n *layer) PutObject(ctx context.Context, p *PutObjectParams) (*data.ExtendedObjectInfo, error) {
bktSettings, err := n.GetBucketSettings(ctx, p.BktInfo)
if err != nil {
return nil, fmt.Errorf("couldn't get versioning settings object: %w", err)
}
r := p.Reader
if p.Encryption.Enabled() {
p.Header[AttributeDecryptedSize] = strconv.FormatUint(p.Size, 10)
if err = addEncryptionHeaders(p.Header, p.Encryption); err != nil {
return nil, fmt.Errorf("add encryption header: %w", err)
}
var encSize uint64
if r, encSize, err = encryptionReader(p.Reader, p.Size, p.Encryption.Key()); err != nil {
return nil, fmt.Errorf("create encrypter: %w", err)
}
p.Size = encSize
}
if r != nil {
if len(p.Header[api.ContentType]) == 0 {
if contentType := MimeByFilePath(p.Object); len(contentType) == 0 {
d := newDetector(r)
if contentType, err := d.Detect(); err == nil {
p.Header[api.ContentType] = contentType
}
r = d.MultiReader()
} else {
p.Header[api.ContentType] = contentType
}
}
}
prm := PrmObjectCreate{
Container: p.BktInfo.CID,
PayloadSize: p.Size,
Filepath: p.Object,
Payload: r,
CreationTime: TimeNow(ctx),
CopiesNumber: p.CopiesNumbers,
}
prm.Attributes = make([][2]string, 0, len(p.Header))
for k, v := range p.Header {
prm.Attributes = append(prm.Attributes, [2]string{k, v})
}
size, id, hash, md5Hash, err := n.objectPutAndHash(ctx, prm, p.BktInfo)
if err != nil {
return nil, err
}
if len(p.ContentMD5) > 0 {
headerMd5Hash, err := base64.StdEncoding.DecodeString(p.ContentMD5)
if err != nil {
return nil, apiErrors.GetAPIError(apiErrors.ErrInvalidDigest)
}
if !bytes.Equal(headerMd5Hash, md5Hash) {
err = n.objectDelete(ctx, p.BktInfo, id)
if err != nil {
n.reqLogger(ctx).Debug(logs.FailedToDeleteObject, zap.Stringer("cid", p.BktInfo.CID), zap.Stringer("oid", id))
}
return nil, apiErrors.GetAPIError(apiErrors.ErrInvalidDigest)
}
}
if !p.Encryption.Enabled() && len(p.ContentSHA256Hash) > 0 && !auth.IsStandardContentSHA256(p.ContentSHA256Hash) {
contentHashBytes, err := hex.DecodeString(p.ContentSHA256Hash)
if err != nil {
return nil, apiErrors.GetAPIError(apiErrors.ErrContentSHA256Mismatch)
}
if !bytes.Equal(contentHashBytes, hash) {
err = n.objectDelete(ctx, p.BktInfo, id)
if err != nil {
n.reqLogger(ctx).Debug(logs.FailedToDeleteObject, zap.Stringer("cid", p.BktInfo.CID), zap.Stringer("oid", id))
}
return nil, apiErrors.GetAPIError(apiErrors.ErrContentSHA256Mismatch)
}
}
n.reqLogger(ctx).Debug(logs.PutObject, zap.Stringer("cid", p.BktInfo.CID), zap.Stringer("oid", id))
newVersion := &data.NodeVersion{
BaseNodeVersion: data.BaseNodeVersion{
OID: id,
ETag: hex.EncodeToString(hash),
FilePath: p.Object,
Size: size,
},
IsUnversioned: !bktSettings.VersioningEnabled(),
IsCombined: p.Header[MultipartObjectSize] != "",
}
if len(p.CompleteMD5Hash) > 0 {
newVersion.MD5 = p.CompleteMD5Hash
} else {
newVersion.MD5 = hex.EncodeToString(md5Hash)
}
if newVersion.ID, err = n.treeService.AddVersion(ctx, p.BktInfo, newVersion); err != nil {
return nil, fmt.Errorf("couldn't add new verion to tree service: %w", err)
}
if p.Lock != nil && (p.Lock.Retention != nil || p.Lock.LegalHold != nil) {
putLockInfoPrms := &PutLockInfoParams{
ObjVersion: &ObjectVersion{
BktInfo: p.BktInfo,
ObjectName: p.Object,
VersionID: id.EncodeToString(),
},
NewLock: p.Lock,
CopiesNumbers: p.CopiesNumbers,
NodeVersion: newVersion, // provide new version to make one less tree service call in PutLockInfo
}
if err = n.PutLockInfo(ctx, putLockInfoPrms); err != nil {
return nil, err
}
}
n.cache.CleanListCacheEntriesContainingObject(p.Object, p.BktInfo.CID)
objInfo := &data.ObjectInfo{
ID: id,
CID: p.BktInfo.CID,
Owner: n.gateOwner,
Bucket: p.BktInfo.Name,
Name: p.Object,
Size: size,
Created: prm.CreationTime,
Headers: p.Header,
ContentType: p.Header[api.ContentType],
HashSum: newVersion.ETag,
MD5Sum: newVersion.MD5,
}
extendedObjInfo := &data.ExtendedObjectInfo{
ObjectInfo: objInfo,
NodeVersion: newVersion,
}
n.cache.PutObjectWithName(n.BearerOwner(ctx), extendedObjInfo)
return extendedObjInfo, nil
}
func (n *layer) headLastVersionIfNotDeleted(ctx context.Context, bkt *data.BucketInfo, objectName string) (*data.ExtendedObjectInfo, error) {
owner := n.BearerOwner(ctx)
if extObjInfo := n.cache.GetLastObject(owner, bkt.Name, objectName); extObjInfo != nil {
return extObjInfo, nil
}
node, err := n.treeService.GetLatestVersion(ctx, bkt, objectName)
if err != nil {
if errors.Is(err, ErrNodeNotFound) {
return nil, fmt.Errorf("%w: %s", apiErrors.GetAPIError(apiErrors.ErrNoSuchKey), err.Error())
}
return nil, err
}
if node.IsDeleteMarker() {
return nil, DeleteMarkerError{ErrorCode: apiErrors.ErrNoSuchKey}
}
meta, err := n.objectHead(ctx, bkt, node.OID)
if err != nil {
if client.IsErrObjectNotFound(err) {
return nil, fmt.Errorf("%w: %s", apiErrors.GetAPIError(apiErrors.ErrNoSuchKey), err.Error())
}
return nil, err
}
objInfo := objectInfoFromMeta(bkt, meta)
objInfo.MD5Sum = node.MD5
extObjInfo := &data.ExtendedObjectInfo{
ObjectInfo: objInfo,
NodeVersion: node,
}
n.cache.PutObjectWithName(owner, extObjInfo)
return extObjInfo, nil
}
func (n *layer) headVersion(ctx context.Context, bkt *data.BucketInfo, p *HeadObjectParams) (*data.ExtendedObjectInfo, error) {
var err error
var foundVersion *data.NodeVersion
if p.VersionID == data.UnversionedObjectVersionID {
foundVersion, err = n.treeService.GetUnversioned(ctx, bkt, p.Object)
if err != nil {
if errors.Is(err, ErrNodeNotFound) {
return nil, fmt.Errorf("%w: %s", apiErrors.GetAPIError(apiErrors.ErrNoSuchVersion), err.Error())
}
return nil, err
}
} else {
versions, err := n.treeService.GetVersions(ctx, bkt, p.Object)
if err != nil {
return nil, fmt.Errorf("couldn't get versions: %w", err)
}
for _, version := range versions {
if version.OID.EncodeToString() == p.VersionID {
foundVersion = version
break
}
}
if foundVersion == nil {
return nil, fmt.Errorf("%w: there isn't tree node with requested version id", apiErrors.GetAPIError(apiErrors.ErrNoSuchVersion))
}
}
owner := n.BearerOwner(ctx)
if extObjInfo := n.cache.GetObject(owner, newAddress(bkt.CID, foundVersion.OID)); extObjInfo != nil {
return extObjInfo, nil
}
if foundVersion.IsDeleteMarker() {
return nil, DeleteMarkerError{ErrorCode: apiErrors.ErrMethodNotAllowed}
}
meta, err := n.objectHead(ctx, bkt, foundVersion.OID)
if err != nil {
if client.IsErrObjectNotFound(err) {
return nil, fmt.Errorf("%w: %s", apiErrors.GetAPIError(apiErrors.ErrNoSuchVersion), err.Error())
}
return nil, err
}
objInfo := objectInfoFromMeta(bkt, meta)
objInfo.MD5Sum = foundVersion.MD5
extObjInfo := &data.ExtendedObjectInfo{
ObjectInfo: objInfo,
NodeVersion: foundVersion,
}
n.cache.PutObject(owner, extObjInfo)
return extObjInfo, nil
}
// objectDelete puts tombstone object into frostfs.
func (n *layer) objectDelete(ctx context.Context, bktInfo *data.BucketInfo, idObj oid.ID) error {
prm := PrmObjectDelete{
Container: bktInfo.CID,
Object: idObj,
}
n.prepareAuthParameters(ctx, &prm.PrmAuth, bktInfo.Owner)
n.cache.DeleteObject(newAddress(bktInfo.CID, idObj))
return n.frostFS.DeleteObject(ctx, prm)
}
// objectPutAndHash prepare auth parameters and invoke frostfs.CreateObject.
// Returns object ID and payload sha256 hash.
func (n *layer) objectPutAndHash(ctx context.Context, prm PrmObjectCreate, bktInfo *data.BucketInfo) (uint64, oid.ID, []byte, []byte, error) {
n.prepareAuthParameters(ctx, &prm.PrmAuth, bktInfo.Owner)
prm.ClientCut = n.features.ClientCut()
prm.BufferMaxSize = n.features.BufferMaxSizeForPut()
prm.WithoutHomomorphicHash = bktInfo.HomomorphicHashDisabled
var size uint64
hash := sha256.New()
md5Hash := md5.New()
prm.Payload = wrapReader(prm.Payload, 64*1024, func(buf []byte) {
size += uint64(len(buf))
hash.Write(buf)
md5Hash.Write(buf)
})
id, err := n.frostFS.CreateObject(ctx, prm)
if err != nil {
if _, errDiscard := io.Copy(io.Discard, prm.Payload); errDiscard != nil {
n.reqLogger(ctx).Warn(logs.FailedToDiscardPutPayloadProbablyGoroutineLeaks, zap.Error(errDiscard))
}
return 0, oid.ID{}, nil, nil, err
}
return size, id, hash.Sum(nil), md5Hash.Sum(nil), nil
}
// ListObjectsV1 returns objects in a bucket for requests of Version 1.
func (n *layer) ListObjectsV1(ctx context.Context, p *ListObjectsParamsV1) (*ListObjectsInfoV1, error) {
var result ListObjectsInfoV1
prm := allObjectParams{
Bucket: p.BktInfo,
Delimiter: p.Delimiter,
Prefix: p.Prefix,
MaxKeys: p.MaxKeys,
Marker: p.Marker,
}
objects, next, err := n.getLatestObjectsVersions(ctx, prm)
if err != nil {
return nil, err
}
if next != nil {
result.IsTruncated = true
result.NextMarker = objects[len(objects)-1].Name
}
result.Prefixes, result.Objects = triageObjects(objects)
return &result, nil
}
// ListObjectsV2 returns objects in a bucket for requests of Version 2.
func (n *layer) ListObjectsV2(ctx context.Context, p *ListObjectsParamsV2) (*ListObjectsInfoV2, error) {
var result ListObjectsInfoV2
prm := allObjectParams{
Bucket: p.BktInfo,
Delimiter: p.Delimiter,
Prefix: p.Prefix,
MaxKeys: p.MaxKeys,
Marker: p.StartAfter,
ContinuationToken: p.ContinuationToken,
}
objects, next, err := n.getLatestObjectsVersionsV2(ctx, prm)
if err != nil {
return nil, err
}
if next != nil {
result.IsTruncated = true
result.NextContinuationToken = next.ID.EncodeToString()
}
result.Prefixes, result.Objects = triageObjects(objects)
return &result, nil
}
type logWrapper struct {
log *zap.Logger
}
func (l *logWrapper) Printf(format string, args ...interface{}) {
l.log.Info(fmt.Sprintf(format, args...))
}
func PrintMemUsage() {
var m runtime.MemStats
runtime.ReadMemStats(&m)
// For info on each, see: https://golang.org/pkg/runtime/#MemStats
fmt.Printf("Alloc = %v MiB", bToMb(m.Alloc))
fmt.Printf("\tTotalAlloc = %v MiB", bToMb(m.TotalAlloc))
fmt.Printf("\tSys = %v MiB", bToMb(m.Sys))
fmt.Printf("\tNumGC = %v\n", m.NumGC)
}
func bToMb(b uint64) uint64 {
return b / 1024 / 1024
}
func (n *layer) getLatestObjectsVersions(ctx context.Context, p allObjectParams) (objects []*data.ObjectInfo, next *data.ObjectInfo, err error) {
if p.MaxKeys == 0 {
return nil, nil, nil
}
owner := n.BearerOwner(ctx)
cacheKey := cache.CreateObjectsListCacheKey(p.Bucket.CID, p.Prefix, true)
nodeVersions := n.cache.GetList(owner, cacheKey)
if nodeVersions == nil {
nodeVersions, err = n.treeService.GetLatestVersionsByPrefix(ctx, p.Bucket, p.Prefix)
if err != nil {
return nil, nil, err
}
n.cache.PutList(owner, cacheKey, nodeVersions)
}
if len(nodeVersions) == 0 {
return nil, nil, nil
}
sort.Slice(nodeVersions, func(i, j int) bool {
return nodeVersions[i].FilePath < nodeVersions[j].FilePath
})
poolCtx, cancel := context.WithCancel(ctx)
defer cancel()
objOutCh, err := n.initWorkerPool(poolCtx, 2, p, nodesGenerator(poolCtx, p, nodeVersions))
if err != nil {
return nil, nil, fmt.Errorf("failed to init worker pool: %w", err)
}
objects = make([]*data.ObjectInfo, 0, p.MaxKeys)
for obj := range objOutCh {
objects = append(objects, obj)
}
//for node := range nodesGenerator(poolCtx, p, nodeVersions) {
// objects = append(objects, &data.ObjectInfo{
// ID: node.OID,
// IsDir: false,
// IsDeleteMarker: node.IsDeleteMarker(),
// Name: node.FilePath,
// Size: node.Size,
// Created: time.Time{},
// HashSum: node.ETag,
// Owner: user.ID{},
// Headers: nil,
// })
//}
sort.Slice(objects, func(i, j int) bool {
return objects[i].Name < objects[j].Name
})
if len(objects) > p.MaxKeys {
next = objects[p.MaxKeys]
objects = objects[:p.MaxKeys]
}
return
}
func (n *layer) getLatestObjectsVersionsV2(ctx context.Context, p allObjectParams) (objects []*data.ObjectInfo, next *data.ObjectInfo, err error) {
if p.MaxKeys == 0 {
return nil, nil, nil
}
owner := n.BearerOwner(ctx)
cacheKey := cache.CreateListSessionCacheKey(p.Bucket.CID, p.Prefix, p.ContinuationToken)
session := n.cache.GetListSession(owner, cacheKey)
if session != nil {
// after reading next object from stream in session
// the current cache value already doesn't match with next token in cache key
n.cache.DeleteListSession(owner, cacheKey)
} else {
session = &data.ListSession{NamesMap: make(map[string]struct{})}
session.Context, session.Cancel = context.WithCancel(context.Background())
if bd, err := middleware.GetBoxData(ctx); err == nil {
session.Context = middleware.SetBoxData(session.Context, bd)
}
session.Stream, err = n.treeService.GetLatestVersionsByPrefixStream(session.Context, p.Bucket, p.Prefix)
if err != nil {
return nil, nil, err
}
}
poolCtx, cancel := context.WithCancel(ctx)
defer cancel()
generator, errorCh := nodesGeneratorStream(poolCtx, p, session)
objOutCh, err := n.initWorkerPoolStream(poolCtx, 2, p, generator)
if err != nil {
return nil, nil, fmt.Errorf("failed to init worker pool: %w", err)
}
objects = make([]*data.ObjectInfo, 0, p.MaxKeys+1)
if session.Next != nil {
objects = append(objects, session.Next)
}
for obj := range objOutCh {
objects = append(objects, obj)
}
//for node := range generator {
// objects = append(objects, &data.ObjectInfo{
// ID: node.OID,
// IsDir: false,
// IsDeleteMarker: node.IsDeleteMarker(),
// Name: node.FilePath,
// Size: node.Size,
// Created: time.Time{},
// HashSum: node.ETag,
// Owner: user.ID{},
// Headers: nil,
// })
//}
if err = <-errorCh; err != nil {
return nil, nil, fmt.Errorf("failed to get next object from stream: %w", err)
}
sort.Slice(objects, func(i, j int) bool {
return objects[i].Name < objects[j].Name
})
if len(objects) > p.MaxKeys {
next = objects[p.MaxKeys]
objects = objects[:p.MaxKeys]
}
if next != nil {
session.Next = next
n.cache.PutListSession(owner, cache.CreateListSessionCacheKey(p.Bucket.CID, p.Prefix, next.VersionID()), session)
}
return
}
func nodesGenerator(ctx context.Context, p allObjectParams, nodeVersions []*data.NodeVersion) <-chan *data.NodeVersion {
nodeCh := make(chan *data.NodeVersion)
existed := make(map[string]struct{}, len(nodeVersions)) // to squash the same directories
go func() {
var generated int
LOOP:
for _, node := range nodeVersions {
if shouldSkip(node, p, existed) {
continue
}
select {
case <-ctx.Done():
break LOOP
case nodeCh <- node:
generated++
if generated == p.MaxKeys+1 { // we use maxKeys+1 to be able to know nextMarker/nextContinuationToken
break LOOP
}
}
}
close(nodeCh)
}()
return nodeCh
}
func nodesGeneratorVersions(ctx context.Context, p allObjectParams, nodeVersions []*data.NodeVersion) <-chan *data.NodeVersion {
nodeCh := make(chan *data.NodeVersion)
existed := make(map[string]struct{}, len(nodeVersions)) // to squash the same directories
go func() {
var generated int
LOOP:
for _, node := range nodeVersions {
if shouldSkipVersions(node, p, existed) {
continue
}
select {
case <-ctx.Done():
break LOOP
case nodeCh <- node:
generated++
if generated == p.MaxKeys+1 { // we use maxKeys+1 to be able to know nextMarker/nextContinuationToken
break LOOP
}
}
}
close(nodeCh)
}()
return nodeCh
}
func nodesGeneratorStream(ctx context.Context, p allObjectParams, stream *data.ListSession) (<-chan *data.NodeVersion, <-chan error) {
nodeCh := make(chan *data.NodeVersion, 1000)
errCh := make(chan error, 1)
//existed := make(map[string]struct{}, p.MaxKeys) // to squash the same directories
existed := stream.NamesMap
if stream.Next != nil {
existed[continuationToken] = struct{}{}
}
limit := p.MaxKeys
if stream.Next == nil {
limit++
}
go func() {
var generated int
var err error
LOOP:
for err == nil {
node, err := stream.Stream.Next(ctx)
if err != nil {
if !errors.Is(err, io.EOF) {
fmt.Println(ctx.Err())
errCh <- fmt.Errorf("stream next: %w", err)
}
break LOOP
}
if shouldSkip(node, p, existed) {
continue
}
select {
case <-ctx.Done():
break LOOP
case nodeCh <- node:
generated++
if generated == limit { // we use maxKeys+1 to be able to know nextMarker/nextContinuationToken
break LOOP
}
}
}
close(nodeCh)
close(errCh)
}()
return nodeCh, errCh
}
func (n *layer) initWorkerPool(ctx context.Context, size int, p allObjectParams, input <-chan *data.NodeVersion) (<-chan *data.ObjectInfo, error) {
reqLog := n.reqLogger(ctx)
pool, err := ants.NewPool(size, ants.WithLogger(&logWrapper{reqLog}))
if err != nil {
return nil, fmt.Errorf("coudln't init go pool for listing: %w", err)
}
objCh := make(chan *data.ObjectInfo)
go func() {
var wg sync.WaitGroup
LOOP:
for node := range input {
select {
case <-ctx.Done():
break LOOP
default:
}
// We have to make a copy of pointer to data.NodeVersion
// to get correct value in submitted task function.
func(node *data.NodeVersion) {
wg.Add(1)
err = pool.Submit(func() {
defer wg.Done()
oi := n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node, p.Prefix, p.Delimiter)
if oi == nil {
// try to get object again
if oi = n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node, p.Prefix, p.Delimiter); oi == nil {
// do not process object which are definitely missing in object service
return
}
}
select {
case <-ctx.Done():
case objCh <- oi:
}
})
if err != nil {
wg.Done()
reqLog.Warn(logs.FailedToSubmitTaskToPool, zap.Error(err))
}
}(node)
}
wg.Wait()
close(objCh)
pool.Release()
}()
return objCh, nil
}
func (n *layer) initWorkerPoolVersions(ctx context.Context, size int, p allObjectParams, input <-chan *data.NodeVersion) (<-chan *data.ExtendedObjectInfo, error) {
reqLog := n.reqLogger(ctx)
pool, err := ants.NewPool(size, ants.WithLogger(&logWrapper{reqLog}))
if err != nil {
return nil, fmt.Errorf("coudln't init go pool for listing: %w", err)
}
objCh := make(chan *data.ExtendedObjectInfo)
go func() {
var wg sync.WaitGroup
LOOP:
for node := range input {
select {
case <-ctx.Done():
break LOOP
default:
}
// We have to make a copy of pointer to data.NodeVersion
// to get correct value in submitted task function.
func(node *data.NodeVersion) {
wg.Add(1)
err = pool.Submit(func() {
defer wg.Done()
oi := &data.ObjectInfo{}
if node.IsDeleteMarker() { // delete marker does not match any object in FrostFS
oi.ID = node.OID
oi.Name = node.FilePath
oi.Owner = node.DeleteMarker.Owner
oi.Created = node.DeleteMarker.Created
oi.IsDeleteMarker = true
} else {
oi = n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node, p.Prefix, p.Delimiter)
if oi == nil {
// try to get object again
if oi = n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node, p.Prefix, p.Delimiter); oi == nil {
// do not process object which are definitely missing in object service
return
}
}
}
eoi := &data.ExtendedObjectInfo{
ObjectInfo: oi,
NodeVersion: node,
}
select {
case <-ctx.Done():
case objCh <- eoi:
}
})
if err != nil {
wg.Done()
reqLog.Warn(logs.FailedToSubmitTaskToPool, zap.Error(err))
}
}(node)
}
wg.Wait()
close(objCh)
pool.Release()
}()
return objCh, nil
}
func (n *layer) initWorkerPoolStream(ctx context.Context, size int, p allObjectParams, input <-chan *data.NodeVersion) (<-chan *data.ObjectInfo, error) {
reqLog := n.reqLogger(ctx)
pool, err := ants.NewPool(size, ants.WithLogger(&logWrapper{reqLog}))
if err != nil {
return nil, fmt.Errorf("coudln't init go pool for listing: %w", err)
}
objCh := make(chan *data.ObjectInfo)
go func() {
var wg sync.WaitGroup
LOOP:
for node := range input {
select {
case <-ctx.Done():
break LOOP
default:
}
// We have to make a copy of pointer to data.NodeVersion
// to get correct value in submitted task function.
func(node *data.NodeVersion) {
wg.Add(1)
err = pool.Submit(func() {
defer wg.Done()
oi := n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node, p.Prefix, p.Delimiter)
if oi == nil {
// try to get object again
if oi = n.objectInfoFromObjectsCacheOrFrostFS(ctx, p.Bucket, node, p.Prefix, p.Delimiter); oi == nil {
// do not process object which are definitely missing in object service
return
}
}
select {
case <-ctx.Done():
case objCh <- oi:
}
})
if err != nil {
wg.Done()
reqLog.Warn(logs.FailedToSubmitTaskToPool, zap.Error(err))
}
}(node)
}
wg.Wait()
close(objCh)
pool.Release()
}()
return objCh, nil
}
func (n *layer) bucketNodeVersions(ctx context.Context, bkt *data.BucketInfo, prefix string) ([]*data.NodeVersion, error) {
var err error
owner := n.BearerOwner(ctx)
cacheKey := cache.CreateObjectsListCacheKey(bkt.CID, prefix, false)
nodeVersions := n.cache.GetList(owner, cacheKey)
if nodeVersions == nil {
nodeVersions, err = n.treeService.GetAllVersionsByPrefix(ctx, bkt, prefix)
if err != nil {
return nil, fmt.Errorf("get all versions from tree service: %w", err)
}
n.cache.PutList(owner, cacheKey, nodeVersions)
}
return nodeVersions, nil
}
func (n *layer) getAllObjectsVersions(ctx context.Context, p *ListObjectVersionsParams) (map[string][]*data.ExtendedObjectInfo, error) {
nodeVersions, err := n.bucketNodeVersions(ctx, p.BktInfo, p.Prefix)
if err != nil {
return nil, err
}
versions := make(map[string][]*data.ExtendedObjectInfo, len(nodeVersions))
sort.Slice(nodeVersions, func(i, j int) bool {
return nodeVersions[i].FilePath < nodeVersions[j].FilePath
})
poolCtx, cancel := context.WithCancel(ctx)
defer cancel()
pp := allObjectParams{
Bucket: p.BktInfo,
Delimiter: p.Delimiter,
Prefix: p.Prefix,
Marker: p.KeyMarker,
ContinuationToken: p.VersionIDMarker,
MaxKeys: p.MaxKeys,
}
objOutCh, err := n.initWorkerPoolVersions(poolCtx, 2, pp, nodesGeneratorVersions(poolCtx, pp, nodeVersions))
if err != nil {
return nil, err
}
for eoi := range objOutCh {
objVersions, ok := versions[eoi.ObjectInfo.Name]
if !ok {
objVersions = []*data.ExtendedObjectInfo{eoi}
} else if !eoi.ObjectInfo.IsDir {
objVersions = append(objVersions, eoi)
}
versions[eoi.ObjectInfo.Name] = objVersions
}
return versions, nil
}
func IsSystemHeader(key string) bool {
_, ok := api.SystemMetadata[key]
return ok || strings.HasPrefix(key, api.FrostFSSystemMetadataPrefix)
}
func shouldSkip(node *data.NodeVersion, p allObjectParams, existed map[string]struct{}) bool {
if node.IsDeleteMarker() {
return true
}
filePath := node.FilePath
if dirName := tryDirectoryName(node, p.Prefix, p.Delimiter); len(dirName) != 0 {
filePath = dirName
}
if _, ok := existed[filePath]; ok {
return true
}
if filePath <= p.Marker {
return true
}
if p.ContinuationToken != "" {
if _, ok := existed[continuationToken]; !ok {
if p.ContinuationToken != node.OID.EncodeToString() {
return true
}
existed[continuationToken] = struct{}{}
}
}
existed[filePath] = struct{}{}
return false
}
func shouldSkipVersions(node *data.NodeVersion, p allObjectParams, existed map[string]struct{}) bool {
filePath := node.FilePath
if dirName := tryDirectoryName(node, p.Prefix, p.Delimiter); len(dirName) != 0 {
filePath = dirName
if _, ok := existed[filePath]; ok {
return true
}
}
if filePath < p.Marker {
return true
}
if p.ContinuationToken != "" {
if _, ok := existed[continuationToken]; !ok {
if p.ContinuationToken != node.OID.EncodeToString() {
return true
}
existed[continuationToken] = struct{}{}
}
}
existed[filePath] = struct{}{}
return false
}
func triageObjects(allObjects []*data.ObjectInfo) (prefixes []string, objects []*data.ObjectInfo) {
for _, ov := range allObjects {
if ov.IsDir {
prefixes = append(prefixes, ov.Name)
} else {
objects = append(objects, ov)
}
}
return
}
func triageExtendedObjects(allObjects []*data.ExtendedObjectInfo) (prefixes []string, objects []*data.ExtendedObjectInfo) {
for _, ov := range allObjects {
if ov.ObjectInfo.IsDir {
prefixes = append(prefixes, ov.ObjectInfo.Name)
} else {
objects = append(objects, ov)
}
}
return
}
func (n *layer) objectInfoFromObjectsCacheOrFrostFS(ctx context.Context, bktInfo *data.BucketInfo, node *data.NodeVersion, prefix, delimiter string) (oi *data.ObjectInfo) {
if oiDir := tryDirectory(bktInfo, node, prefix, delimiter); oiDir != nil {
return oiDir
}
owner := n.BearerOwner(ctx)
if extInfo := n.cache.GetObject(owner, newAddress(bktInfo.CID, node.OID)); extInfo != nil {
return extInfo.ObjectInfo
}
meta, err := n.objectHead(ctx, bktInfo, node.OID)
if err != nil {
n.reqLogger(ctx).Warn(logs.CouldNotFetchObjectMeta, zap.Error(err))
return nil
}
oi = objectInfoFromMeta(bktInfo, meta)
oi.MD5Sum = node.MD5
n.cache.PutObject(owner, &data.ExtendedObjectInfo{ObjectInfo: oi, NodeVersion: node})
return oi
}
func tryDirectory(bktInfo *data.BucketInfo, node *data.NodeVersion, prefix, delimiter string) *data.ObjectInfo {
dirName := tryDirectoryName(node, prefix, delimiter)
if len(dirName) == 0 {
return nil
}
return &data.ObjectInfo{
ID: node.OID, // to use it as continuation token
CID: bktInfo.CID,
IsDir: true,
IsDeleteMarker: node.IsDeleteMarker(),
Bucket: bktInfo.Name,
Name: dirName,
}
}
// tryDirectoryName forms directory name by prefix and delimiter.
// If node isn't a directory empty string is returned.
// This function doesn't check if node has a prefix. It must do a caller.
func tryDirectoryName(node *data.NodeVersion, prefix, delimiter string) string {
if len(delimiter) == 0 {
return ""
}
tail := strings.TrimPrefix(node.FilePath, prefix)
index := strings.Index(tail, delimiter)
if index >= 0 {
return prefix + tail[:index+1]
}
return ""
}
func wrapReader(input io.Reader, bufSize int, f func(buf []byte)) io.Reader {
if input == nil {
return nil
}
r, w := io.Pipe()
go func() {
var buf = make([]byte, bufSize)
for {
n, err := input.Read(buf)
if n > 0 {
f(buf[:n])
_, _ = w.Write(buf[:n]) // ignore error, input is not ReadCloser
}
if err != nil {
_ = w.CloseWithError(err)
break
}
}
}()
return r
}