astelm/minio
1
0
Fork 0
mirror of https://github.com/pgsty/minio.git synced 2026-11-10 09:35:31 +03:00
Code Issues Packages Projects Releases Wiki Activity

fix(restore): stage native restores outside the object lock

Browse Source 
The unreleased native-restore fixes after RELEASE.2026-10-26T00-00-00Z
let the lock decide who owns a restore. Since fa42a0982 a single-pool
worker holds the object lock for the whole tier download, and since
7f38bf628 a busy lock means "a restore runs". 7f38bf628 also keeps the
restore's data in a marked multipart upload. Two regressions followed:

- While an old version restored, every GET, HEAD and write of the same
  key waited for the download. On a single pool this is new; with
  several pools the released code already held the lock that long.
- An ordinary upload whose completion failed after staging its metadata
  became InvalidUploadID for every caller, and could not be aborted for
  an hour. The marker that hid restore uploads could not tell them apart
  once a completion had rewritten the upload's xl.meta.

A restore now runs in two phases:

- The worker erasure-codes the tier copy into the tmp volume of the
  version's set, without the object lock, as heal does. A drive that
  fails any part, or misses a part when CheckParts runs before the
  commit, takes no part in the commit.
- A restore lock per version tells whether a restore runs. POST takes it
  before the object lock and hands it to the worker. It is a liveness
  hint: only the checks under the object lock decide what is written.
- The commit takes the object lock briefly, on one pool or several. It
  reads the version again and checks the attempt's ownership, the
  version identity, the tier copy, the part layout and the erasure
  layout. It then publishes with RenameData drive by drive, never
  through erasure renameData's undo, and needs the write quorum. The
  replaced data directory goes only when every published drive reports
  the same one.
- A publication that reached some drives but not the write quorum
  converges under the same lock hold, in bounded retries after it: back
  to the unrestored version on every drive that holds it, or to a
  committed restore when the copy turns out to be on a write quorum.
  Each pass first checks that every drive holds what the version held
  before the commit, the commit wrote, or a convergence wrote, and stops
  on anything else. It deletes no directory.

Multipart goes back to main: erasure-multipart.go and
bucket-object-lock.go are identical to 69fcf50cd. The commit lock, the
marker and the restore exceptions of multipart completion and of the
null-version Object Lock check are gone, and single-pool completion
again takes the upload lock first and the object lock right before it
commits. With several pools, abandoned-data cleanup now takes the pools
lock that writes take, as heal does, so it cannot remove a directory
that a commit moved in place and has not yet referenced. The restore
capability becomes protocol 2, and a queued worker checks it again.

Recovery after a coordinator crash during publication, and storage RPCs
that complete after their caller returned, stay excluded. Main has the
same exposure, and they are tracked separately.

Tests:
- New regression tests cover both problems. They fail on f85d4a5fa,
  and the multipart one also passes on main.
- Fault-injection tests cover a publication below the write quorum,
  including seven of twelve drives on 8+4; lost answers; convergence
  retries and stops; the old data directory; a stale inline marker; a
  recreated tmp volume; a published revision read across its expiry; a
  second POST; a target deleted during the download; a tier 404; and
  pools-locked abandoned cleanup.
- Thirteen tests of the upload marker from 7f38bf628 are dropped, and
  so is TestNativeRestoreConcurrentCompletion from fa42a0982, which
  completed the restore's upload from a client. Four tests are reframed
  for the new semantics, including the lock-domain restore test, which
  now gates the commit.
- Three self-recursive WriteMetadata test doubles are fixed, and the
  helpers of the dropped tests are removed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Feng Ruohang <rh@vonng.com>
This commit is contained in:
Feng Ruohang
2026-11-06 11:30:29 +08:00
parent f85d4a5fa1
commit 45b1e4bc00
17 changed files with 1956 additions and 2026 deletions
Download Patch File Download Diff File Expand all files Collapse all files
cmd/restore-native_test.go
+25 -124
View File
@@ -373,10 +373,9 @@ func TestNativeRestorePOSTVersionIdentity(t *testing.T) {
}
}
// A restore request is internal context, not an exception for normal S3
// multipart overwrites or for a version that has never completed transition.
// Only the upload of a native restore attempt, whose marker names the very
// version being completed, may write that version back.
// A restore request is internal context, not an exception for multipart
// overwrites: a native restore never completes an upload, so no completion,
// whatever restore context it carries, writes an existing version back.
func TestNativeRestoreMultipartOverwriteBoundary(t *testing.T) {
for _, mode := range []string{"matching-restore", "ordinary-complete", "restore-without-attempt", "other-attempt", "not-tiered", "changed-time", "changed-size", "changed-etag"} {
t.Run(mode, func(t *testing.T) {
@@ -446,24 +445,7 @@ func TestNativeRestoreMultipartOverwriteBoundary(t *testing.T) {
}
before := nativeRestoreSnapshot(t, r)
_, err = pool.CompleteMultipartUpload(t.Context(), r.bucket, r.object, upload.UploadID, complete, opts)
if mode == "matching-restore" {
if err != nil {
t.Fatalf("matching restore rejected: %v", err)
}
_, got, err := r.read(oi.VersionID)
if err != nil || !bytes.Equal(got, body) {
t.Fatalf("matching restore contents: %v", err)
}
if after := r.fileInfo(oi.VersionID); after.Metadata[restoreUploadKey] != "" || after.Metadata[restoreAttemptKey] != "" {
t.Error("restore committed its private upload or attempt marker")
}
return
}
if mode == "other-attempt" {
if !errors.As(err, &InvalidUploadID{}) {
t.Fatalf("another attempt completed a private upload: %v", err)
}
} else if !errors.Is(err, errAuthentication) {
if !errors.Is(err, errAuthentication) {
t.Fatalf("existing version overwrite: %v", err)
}
if !reflect.DeepEqual(before, nativeRestoreSnapshot(t, r)) {
@@ -525,7 +507,10 @@ func nativeRestoreLockRequest(ctx context.Context, t *testing.T, r *restoreTest,
return rec
}
func TestNativeRestoreSerializesObjectLock(t *testing.T) {
// A native restore downloads without the object lock. An Object Lock
// update made meanwhile commits at once, and the restore, whether it
// completes or fails, keeps it.
func TestNativeRestoreKeepsObjectLockUpdate(t *testing.T) {
for _, kind := range []string{"legal-hold", "retention"} {
for _, fail := range []bool{false, true} {
t.Run(fmt.Sprintf("%s/failure=%v", kind, fail), func(t *testing.T) {
@@ -567,22 +552,19 @@ func TestNativeRestoreSerializesObjectLock(t *testing.T) {
case <-ctx.Done():
t.Fatal("restore never started download")
}
updateCtx, updateCancel := context.WithTimeout(t.Context(), 200*time.Millisecond)
// The download holds no object lock: an Object Lock update
// commits at once, and the restore's commit keeps it.
updateCtx, updateCancel := context.WithTimeout(t.Context(), 5*time.Second)
rec := nativeRestoreLockRequest(updateCtx, t, r, oi.VersionID, kind, updated)
updateCancel()
if rec.Code == http.StatusOK {
t.Error("Object Lock update committed during restore's metadata snapshot")
// On the unfixed path this update commits, and the restore
// must not subsequently erase the acknowledged protection.
readOpts := r.opts(oi.VersionID)
readOpts.NoLock = true
before, err = r.z.GetObjectInfo(t.Context(), r.bucket, r.object, readOpts)
if err != nil {
t.Fatal(err)
}
if rec.Code != http.StatusOK {
t.Fatalf("Object Lock update during the restore download: %d %s", rec.Code, rec.Body.String())
}
if rec.Code != http.StatusOK && (rec.Code != http.StatusServiceUnavailable || !strings.Contains(rec.Body.String(), "<Code>RequestTimeout</Code>")) {
t.Errorf("update failed for an unrelated reason: %d %s", rec.Code, rec.Body.String())
readOpts := r.opts(oi.VersionID)
readOpts.NoLock = true
during, err := r.z.GetObjectInfo(t.Context(), r.bucket, r.object, readOpts)
if err != nil {
t.Fatal(err)
}
unblock()
select {
@@ -598,26 +580,19 @@ func TestNativeRestoreSerializesObjectLock(t *testing.T) {
t.Fatal(err)
}
for _, key := range keys {
if before.UserDefined[key] == "" || after.UserDefined[key] != before.UserDefined[key] {
t.Errorf("restore changed protection %s: %q -> %q", key, before.UserDefined[key], after.UserDefined[key])
if during.UserDefined[key] == "" || after.UserDefined[key] != during.UserDefined[key] {
t.Errorf("restore changed protection %s: %q -> %q", key, during.UserDefined[key], after.UserDefined[key])
}
}
if rec := nativeRestoreLockRequest(t.Context(), t, r, oi.VersionID, kind, updated); rec.Code != http.StatusOK {
t.Fatalf("protection update after restore: %d %s", rec.Code, rec.Body.String())
}
final, err := r.z.GetObjectInfo(t.Context(), r.bucket, r.object, r.opts(oi.VersionID))
if err != nil {
t.Fatal(err)
}
if !final.ModTime.Equal(before.ModTime) || final.ETag != before.ETag || final.Size != before.Size {
t.Error("Object Lock update changed content identity")
}
changed := keys[0]
if kind == "retention" {
changed = keys[1]
}
if final.UserDefined[changed] == before.UserDefined[changed] || final.UserDefined[keys[len(keys)-1]] == before.UserDefined[keys[len(keys)-1]] {
t.Error("new protection or its timestamp was not stored")
if after.UserDefined[changed] == before.UserDefined[changed] {
t.Error("the update made during the download was not kept")
}
if !after.ModTime.Equal(before.ModTime) || after.ETag != before.ETag || after.Size != before.Size {
t.Error("Object Lock update changed content identity")
}
})
}
@@ -704,80 +679,6 @@ func TestNativeRestoreQueuedBehindDownload(t *testing.T) {
}
}
func TestNativeRestoreConcurrentCompletion(t *testing.T) {
for _, separateSets := range []bool{false, true} {
t.Run(fmt.Sprintf("separate-set-locks=%v", separateSets), func(t *testing.T) {
r := newRestoreSafetyTest(t, 1, versioning.Enabled)
if separateSets {
// Local production sets share one map. Independent maps model the
// distinct locker groups of distributed sets while reusing test disks.
pool := r.z.serverPools[0]
sets, count := pool.sets, pool.setCount
t.Cleanup(func() { pool.sets, pool.setCount = sets, count })
first, second := *sets[0], *sets[0]
first.nsMutex, second.nsMutex = newNSLock(false), newNSLock(false)
second.setIndex = 1
pool.sets, pool.setCount = []*erasureObjects{&first, &second}, 2
for i := 0; ; i++ {
r.object = fmt.Sprintf("second-set-%d", i)
if pool.getHashedSetIndex(r.object) == 1 {
break
}
}
}
oi, body := nativeRestoreSeed(t, r, 0, false, true)
blocked := &nativeRestoreBlockedTier{WarmBackend: globalTierConfigMgr.drivercache["WARM"], entered: make(chan struct{}), release: make(chan struct{}), afterFirstPart: true}
globalTierConfigMgr.drivercache["WARM"] = blocked
var once sync.Once
unblock := func() { once.Do(func() { close(blocked.release) }) }
defer unblock()
ctx, cancel := context.WithTimeout(t.Context(), 3*time.Second)
defer cancel()
restored := make(chan error, 1)
go func() {
restored <- r.z.RestoreTransitionedObject(ctx, r.bucket, r.object, nativeRestoreOptions(r, oi.VersionID))
}()
select {
case <-blocked.entered:
case <-ctx.Done():
t.Fatal("restore never started")
}
// A client that learns the internal upload ID, which listings no
// longer show, can try to complete the first part while native
// restore owns the object.
id, part := restoreIsolationUpload(t, r, 0)
completed := make(chan struct{})
var completeErr error
go func() {
_, completeErr = r.z.CompleteMultipartUpload(ctx, r.bucket, r.object, id, []CompletePart{{PartNumber: 1, ETag: part.ETag}}, r.opts(oi.VersionID))
close(completed)
}()
waitForListingReadLock(t, r.z.serverPools[0].getHashedSet(r.object).nsMutex, r.bucket, r.object, completed)
unblock()
select {
case err := <-restored:
if err != nil {
t.Errorf("restore blocked by concurrent completion: %v", err)
}
case <-time.After(5 * time.Second):
t.Fatal("restore did not finish")
}
select {
case <-completed:
case <-time.After(5 * time.Second):
t.Fatal("completion did not finish")
}
if !errors.As(completeErr, &InvalidUploadID{}) {
t.Errorf("completion after native restore: %v, want InvalidUploadID", completeErr)
}
_, got, err := r.read(oi.VersionID)
if err != nil || !bytes.Equal(got, body) {
t.Fatalf("restore contents after concurrent completion: %v", err)
}
})
}
}
type nativeRestorePartGateReader struct {
io.ReadCloser
gate *nativeRestoreBlockedTier