Files
seaweedfs/weed/s3api/s3api_bucket_owner_index.go
T
Chris Lu 7bb0a1c127 s3: replay a delete whose reply the transport dropped (#11022)
* s3: stop retrying a delete the filer refused for a non-empty folder

The filer looked and the children are there, so the answer will not change.
retryFilerOp spent six attempts and up to 3.1s of backoff on it before the
caller could act on the condition it was already holding.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD

* s3: thread the request context through the unversioned delete path

doDeleteEntry issued every DeleteEntry on context.Background(), so an S3
client that hung up left the gateway working on its behalf, out of reach of
both cancellation and the per-request retry allowance that
DeleteMultipleObjectsHandler installs.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD

* s3: treat a cancelled filer RPC as terminal, not transient

isRetryableFilerErr matched context.Canceled and DeadlineExceeded by
sentinel, which only holds while the error is still local. Once it has
crossed gRPC it is a status, so an abandoned request was retried six times
on behalf of a caller that had already gone.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD

* s3: replay a delete whose reply the transport dropped

A delete is idempotent at the filer, which answers an entry that is already
gone with an empty resp.Error, so a reply lost in transit can be reissued
rather than surfaced. Surfaced, it becomes a 500 on the bucket delete, which
boto3 resends and is then answered NoSuchBucket, or a per-key InternalError
inside the 200 of a multi-object delete, which no SDK retries at all.

The replay runs through retryFilerOp, so it draws on the allowance the
request already installs rather than paying a backoff per key, and stops for
a caller that has gone. rm and rmObject re-enter WithFilerClient per attempt,
so each one walks the failover list again on a connection the failed attempt
had invalidated; the multi-object loop holds one client for the batch, so
there the replay reuses it.

Classification stays structural. The filer reports its own refusals in
resp.Error, which carries no status and has the deleted path - and, for a
recursive delete, the children it stopped on - formatted into it, so no key
name can steer the decision either way.

rm and rmObject now take the caller's context. Cleanup and rollback paths
pass context.Background() deliberately: they have to run whether or not the
caller is still waiting.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD

* s3: share one retry allowance across multipart completion cleanup

The unused-entry loop deletes once per entry, and each delete now retries,
so a filer that stays unavailable held the response for 3.1s per entry after
the object was already committed.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD
2026-08-28 14:30:21 -07:00

318 lines
10 KiB
Go

package s3api
import (
"context"
"net/http"
"net/url"
"slices"
"strings"
"time"
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/pb/filer_pb"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3_constants"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3err"
"github.com/seaweedfs/seaweedfs/weed/util"
)
// The owner index maps bucket owner -> owned buckets so ListBuckets can serve
// an identity's buckets without scanning the global /buckets directory:
//
// <BucketsPath>/.system/owners/<escaped owner>/<bucket>
//
// Index entries are zero-length files whose Crtime mirrors the bucket's
// creation time. The bucket handlers write the index synchronously, the
// /buckets metadata subscription reconciles changes made elsewhere (weed
// shell, another gateway, direct filer operations), and a startup backfill
// indexes pre-existing buckets. The .complete marker is written once the
// backfill has finished; until it exists readers keep scanning /buckets.
const (
s3SystemFolder = ".system"
bucketOwnerIndexFolder = "owners"
bucketOwnerIndexReadyMarker = ".complete"
)
func (s3a *S3ApiServer) bucketOwnerIndexPath() string {
return s3a.option.BucketsPath + "/" + s3SystemFolder + "/" + bucketOwnerIndexFolder
}
func (s3a *S3ApiServer) bucketOwnerDir(owner string) string {
return s3a.bucketOwnerIndexPath() + "/" + escapeOwnerName(owner)
}
// escapeOwnerName makes an owner id safe to use as a single directory name.
// url.PathEscape escapes "/", so no owner name can traverse out of the index.
// It leaves dots alone, so a leading dot is encoded by hand: "." and ".."
// must not resolve as path segments, and other dot-names would collide with
// index-internal entries like the ready marker. PathEscape never emits a
// literal "%2E" itself, so the encoding stays collision-free.
func escapeOwnerName(owner string) string {
escaped := url.PathEscape(owner)
if strings.HasPrefix(escaped, ".") {
escaped = "%2E" + escaped[1:]
}
return escaped
}
func bucketEntryOwner(entry *filer_pb.Entry) string {
if entry == nil || entry.Extended == nil {
return ""
}
return string(entry.Extended[s3_constants.AmzIdentityId])
}
func (s3a *S3ApiServer) addBucketToOwnerIndex(owner, bucket string, crtime int64) error {
return s3a.mkFile(s3a.bucketOwnerDir(owner), bucket, nil, func(entry *filer_pb.Entry) {
entry.Attributes.Crtime = crtime
})
}
func (s3a *S3ApiServer) removeBucketFromOwnerIndex(owner, bucket string) error {
return s3a.rm(context.Background(), s3a.bucketOwnerDir(owner), bucket, false, false)
}
// maintainBucketOwnerIndex applies a /buckets metadata event to the owner
// index, covering bucket changes made outside this gateway's handlers. Every
// operation is idempotent, so overlap with the handlers' synchronous writes
// and with other gateways is harmless.
func (s3a *S3ApiServer) maintainBucketOwnerIndex(oldEntry, newEntry *filer_pb.Entry) {
var oldOwner, newOwner string
if oldEntry != nil && (!oldEntry.IsDirectory || strings.HasPrefix(oldEntry.Name, ".")) {
oldEntry = nil
}
if newEntry != nil && (!newEntry.IsDirectory || strings.HasPrefix(newEntry.Name, ".")) {
newEntry = nil
}
if oldEntry != nil {
oldOwner = bucketEntryOwner(oldEntry)
}
if newEntry != nil {
newOwner = bucketEntryOwner(newEntry)
}
unchanged := oldEntry != nil && newEntry != nil && oldEntry.Name == newEntry.Name && oldOwner == newOwner
if unchanged {
return
}
if oldOwner != "" {
if err := s3a.removeBucketFromOwnerIndex(oldOwner, oldEntry.Name); err != nil {
glog.V(1).Infof("owner index: remove %s/%s: %v", oldOwner, oldEntry.Name, err)
}
}
if newOwner != "" {
if err := s3a.addBucketToOwnerIndex(newOwner, newEntry.Name, newEntry.Attributes.Crtime); err != nil {
glog.V(1).Infof("owner index: add %s/%s: %v", newOwner, newEntry.Name, err)
}
}
}
// bucketOwnerIndexReady reports whether the backfill marker exists, caching a
// positive answer for the life of the process.
func (s3a *S3ApiServer) bucketOwnerIndexReady() bool {
if s3a.ownerIndexReady.Load() {
return true
}
if exists, err := s3a.exists(s3a.bucketOwnerIndexPath(), bucketOwnerIndexReadyMarker, false); err == nil && exists {
s3a.ownerIndexReady.Store(true)
return true
}
return false
}
// startBucketOwnerIndexBackfill brings the owner index up to date with the
// buckets that existed before it was introduced, then writes the ready
// marker. Multiple gateways may race the backfill; every step is idempotent.
func (s3a *S3ApiServer) startBucketOwnerIndexBackfill() {
util.RetryUntil("bucketOwnerIndexBackfill", func() error {
if s3a.bucketOwnerIndexReady() {
return nil
}
return s3a.backfillBucketOwnerIndex()
}, func(err error) bool {
glog.V(1).Infof("bucket owner index backfill: %v", err)
return true
})
}
// listBucketsFromOwnerIndex serves ListBuckets from the owner index: the
// identity's owned buckets merged with the buckets its legacy actions name
// explicitly. Cost is proportional to the identity's own buckets, not to the
// global bucket count.
func (s3a *S3ApiServer) listBucketsFromOwnerIndex(r *http.Request, identity *Identity, grantedNames []string, prefix, startAfter string, maxBuckets int) ([]ListAllMyBucketsEntry, string, error) {
granted := s3a.resolveGrantedBuckets(r, identity, grantedNames, prefix, startAfter)
var listOwned bucketPageLister
if identity.Name != "" {
ownerDir := s3a.bucketOwnerDir(identity.Name)
listOwned = func(startFrom string) ([]*filer_pb.Entry, bool, error) {
return s3a.list(ownerDir, prefix, startFrom, false, bucketScanPageSize)
}
}
return listMergedBuckets(listOwned, granted, startAfter, maxBuckets)
}
// resolveGrantedBuckets maps action-granted bucket names to listing entries,
// dropping duplicates, names outside the requested page, buckets that no
// longer exist, and names the identity cannot actually list.
func (s3a *S3ApiServer) resolveGrantedBuckets(r *http.Request, identity *Identity, names []string, prefix, startAfter string) (granted []ListAllMyBucketsEntry) {
slices.Sort(names)
names = slices.Compact(names)
for _, name := range names {
if name <= startAfter || !strings.HasPrefix(name, prefix) || strings.HasPrefix(name, ".") {
continue
}
config, errCode := s3a.getBucketConfig(name)
if errCode != s3err.ErrNone {
if errCode != s3err.ErrNoSuchBucket {
glog.V(1).Infof("owner index: resolve granted bucket %s: %v", name, errCode)
}
continue
}
if !s3a.bucketVisibleToIdentity(r, name, config.IdentityId, identity) {
continue
}
granted = append(granted, ListAllMyBucketsEntry{
Name: name,
CreationDate: time.Unix(config.Crtime, 0).UTC(),
})
}
return granted
}
type bucketPageLister func(startFrom string) (entries []*filer_pb.Entry, isLast bool, err error)
// listMergedBuckets merges the sorted owned-bucket index stream with the
// pre-filtered, sorted granted entries, returning up to maxBuckets and a
// continuation token when more remain.
func listMergedBuckets(listOwned bucketPageLister, granted []ListAllMyBucketsEntry, startAfter string, maxBuckets int) (buckets []ListAllMyBucketsEntry, nextToken string, err error) {
gi := 0
startFrom := startAfter
if listOwned != nil {
owned:
for len(buckets) <= maxBuckets {
entries, isLast, listErr := listOwned(startFrom)
if listErr != nil {
return nil, "", listErr
}
if len(entries) == 0 {
break
}
for _, entry := range entries {
startFrom = entry.Name
if entry.IsDirectory || strings.HasPrefix(entry.Name, ".") {
continue
}
for gi < len(granted) && granted[gi].Name < entry.Name {
buckets = append(buckets, granted[gi])
gi++
if len(buckets) > maxBuckets {
break owned
}
}
if gi < len(granted) && granted[gi].Name == entry.Name {
gi++ // also owned; keep the index entry
}
buckets = append(buckets, ListAllMyBucketsEntry{
Name: entry.Name,
CreationDate: time.Unix(entry.Attributes.Crtime, 0).UTC(),
})
if len(buckets) > maxBuckets {
break owned
}
}
if isLast {
break
}
}
}
for gi < len(granted) && len(buckets) <= maxBuckets {
buckets = append(buckets, granted[gi])
gi++
}
if len(buckets) > maxBuckets {
buckets = buckets[:maxBuckets]
nextToken = encodeContinuationToken(buckets[maxBuckets-1].Name)
}
return buckets, nextToken, nil
}
func (s3a *S3ApiServer) backfillBucketOwnerIndex() error {
startFrom := ""
for {
pageStart := startFrom
entries, isLast, err := s3a.list(s3a.option.BucketsPath, "", startFrom, false, bucketScanPageSize)
if err != nil {
return err
}
if len(entries) == 0 {
break
}
var indexed []*filer_pb.Entry
for _, entry := range entries {
startFrom = entry.Name
if !entry.IsDirectory || strings.HasPrefix(entry.Name, ".") {
continue
}
if owner := bucketEntryOwner(entry); owner != "" {
if err := s3a.addBucketToOwnerIndex(owner, entry.Name, entry.Attributes.Crtime); err != nil {
return err
}
indexed = append(indexed, entry)
}
}
// A bucket deleted while this page was in flight would leave a permanent
// phantom record: the index write lands after the delete already tried to
// clean up. Re-list the page and drop records for buckets that vanished;
// a delete after this point sees the record and removes it itself.
if err := s3a.dropVanishedFromOwnerIndex(pageStart, indexed); err != nil {
return err
}
if isLast {
break
}
}
if err := s3a.mkFile(s3a.bucketOwnerIndexPath(), bucketOwnerIndexReadyMarker, nil, nil); err != nil {
return err
}
s3a.ownerIndexReady.Store(true)
glog.V(0).Infof("bucket owner index ready")
return nil
}
func (s3a *S3ApiServer) dropVanishedFromOwnerIndex(startFrom string, indexed []*filer_pb.Entry) error {
if len(indexed) == 0 {
return nil
}
last := indexed[len(indexed)-1].Name
present := make(map[string]bool, len(indexed))
from := startFrom
scan:
for {
entries, isLast, err := s3a.list(s3a.option.BucketsPath, "", from, false, bucketScanPageSize)
if err != nil {
return err
}
if len(entries) == 0 {
break
}
for _, entry := range entries {
if entry.Name > last {
break scan
}
from = entry.Name
present[entry.Name] = true
}
if isLast {
break
}
}
for _, entry := range indexed {
if present[entry.Name] {
continue
}
if err := s3a.removeBucketFromOwnerIndex(bucketEntryOwner(entry), entry.Name); err != nil {
return err
}
}
return nil
}