astelm/minio
1
0
Fork 0
mirror of https://github.com/pgsty/minio.git synced 2026-11-10 17:45:32 +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/erasure-object.go
+2 -177
View File
@@ -45,7 +45,6 @@ import (
"github.com/minio/minio/internal/crypto"
"github.com/minio/minio/internal/event"
"github.com/minio/minio/internal/grid"
"github.com/minio/minio/internal/hash"
xhttp "github.com/minio/minio/internal/http"
xioutil "github.com/minio/minio/internal/ioutil"
"github.com/minio/minio/internal/logger"
@@ -156,7 +155,7 @@ func (er erasureObjects) CopyObject(ctx context.Context, srcBucket, srcObject, d
if dstOpts.replicaObjectInfo != nil {
getObjectInfo = dstOpts.replicaObjectInfo
}
if err := checkObjectLockedForOverwrite(ctx, dstBucket, dstObject, getObjectInfo, dstOpts.Transition); err != nil {
if err := checkObjectLockedForOverwrite(ctx, dstBucket, dstObject, getObjectInfo); err != nil {
return oi, toObjectErr(err, dstBucket, dstObject)
}
}
@@ -1673,7 +1672,7 @@ func (er erasureObjects) putObject(ctx context.Context, bucket string, object st
if opts.replicaObjectInfo != nil {
getObjectInfo = opts.replicaObjectInfo
}
if err := checkObjectLockedForOverwrite(ctx, bucket, object, getObjectInfo, opts.Transition); err != nil {
if err := checkObjectLockedForOverwrite(ctx, bucket, object, getObjectInfo); err != nil {
return ObjectInfo{}, toObjectErr(err, bucket, object)
}
}
@@ -2665,180 +2664,6 @@ func (er erasureObjects) TransitionObject(ctx context.Context, bucket, object st
return err
}
// RestoreTransitionedObject - restore transitioned object content locally on this cluster.
// This is similar to PostObjectRestore from AWS GLACIER
// storage class. When PostObjectRestore API is called, a temporary copy of the object
// is restored locally to the bucket on source cluster until the restore expiry date.
// The copy that was transitioned continues to reside in the transitioned tier.
func (er erasureObjects) RestoreTransitionedObject(ctx context.Context, bucket, object string, opts ObjectOptions) error {
return er.restoreTransitionedObject(ctx, bucket, object, opts)
}
// getRestoreLock waits for a restore's object lock for as long as ctx lives.
// The POST handler has already marked the version ongoing, and only the
// restore clears that marker. A restore queued behind another version's
// download must therefore keep waiting rather than give up after one timeout.
func getRestoreLock(ctx context.Context, lk RWLocker) (LockContext, error) {
for {
lkctx, err := lk.GetLock(ctx, globalRestoreLockTimeout)
if !errors.As(err, &OperationTimedOut{}) || ctx.Err() != nil {
return lkctx, err
}
}
}
// restoreTransitionedObject for multipart object chunks the file stream from remote tier into the same number of parts
// as in the xl.meta for this version and rehydrates the part.n into the fi.DataDir for this version as in the xl.meta
//
// A POST restore publishes its attempt on the version's ongoing marker. The
// restore rehydrates the version only while that attempt owns the marker, and
// on failure clears the marker only while it still does. A direct caller
// publishes no attempt; its attempt only names the private upload.
func (er erasureObjects) restoreTransitionedObject(ctx context.Context, bucket string, object string, opts ObjectOptions) error {
if !opts.NoLock {
// Single-pool and direct callers must protect the metadata snapshot
// through download, rehydration and failure cleanup, just like pools.
lk := er.NewNSLock(bucket, object)
lkctx, err := getRestoreLock(ctx, lk)
if err != nil {
return restoreResetError{err}
}
ctx = lkctx.Context()
defer lk.Unlock(lkctx)
opts.NoLock = true
}
published := opts.Transition.attempt
// Restore the authoritative generation of the version, never a revision
// that only some drives hold.
gen, err := er.readRestoreGeneration(ctx, bucket, object, opts.VersionID)
if err != nil {
err = toObjectErr(err, bucket, object, opts.VersionID)
if isErrObjectNotFound(err) || isErrVersionNotFound(err) {
// No version was read, so there is no restore marker to clear.
return err
}
return restoreResetError{err}
}
actualfi := gen.fi
if actualfi.Deleted {
return toObjectErr(errMethodNotAllowed, bucket, object)
}
oi := actualfi.ToObjectInfo(bucket, object, opts.Versioned || opts.VersionSuspended)
attempt := published
if attempt == nil {
attempt = newRestoreAttempt(oi)
} else if !attempt.owns(oi) {
// A later POST took the version over, or it was replaced.
return errRestoreSuperseded
}
opts.Transition.attempt = attempt
versionID := restoreVersionID(actualfi.VersionID)
fail := func(cause error) error {
return er.restoreFailed(ctx, bucket, object, versionID, published, cause)
}
ropts := putRestoreOpts(opts.Transition.RestoreRequest, oi, opts)
ropts.NoLock = opts.NoLock
ropts.Transition = opts.Transition
// Rehydration copies encoded bytes without changing their content identity.
ropts.PreserveETag = oi.ETag
// Even a one-part multipart upload must retain its original part number:
// encrypted multipart data uses that number to derive its decryption key.
if len(oi.Parts) == 1 && !oi.isMultipart() {
actualSize, err := oi.GetActualSize()
if err != nil {
return fail(err)
}
ropts.IndexCB = func() []byte { return oi.Parts[0].Index }
var rs *HTTPRangeSpec
gr, err := getTransitionedObjectReader(ctx, bucket, object, rs, http.Header{}, oi, opts)
if err != nil {
return fail(toObjectErr(err, bucket, object))
}
defer gr.Close()
hashReader, err := hash.NewReader(ctx, gr, gr.ObjInfo.Size, "", "", actualSize)
if err != nil {
return fail(toObjectErr(err, bucket, object))
}
pReader := NewPutObjReader(hashReader)
if _, err = er.PutObject(ctx, bucket, object, pReader, ropts); err != nil {
return fail(toObjectErr(err, bucket, object))
}
return nil
}
// Every server must keep the upload private, also a queued restore's.
if err := nativeRestoreAvailable(ctx); err != nil {
return fail(err)
}
// The upload records when it starts, which the stale-upload cleanup ages;
// completion still commits the version's own ModTime.
uopts := ropts
uopts.MTime = time.Time{}
res, err := er.NewMultipartUpload(ctx, bucket, object, uopts)
if err != nil {
return fail(err)
}
completed := false
defer func() {
if !completed {
er.abortRestoreUpload(bucket, object, res.UploadID, attempt)
}
}()
var uploadedParts []CompletePart
var rs *HTTPRangeSpec
// get reader from the warm backend - note that even in the case of encrypted objects, this stream is still encrypted.
gr, err := getTransitionedObjectReader(ctx, bucket, object, rs, http.Header{}, oi, opts)
if err != nil {
return fail(err)
}
defer gr.Close()
// rehydrate the parts back on disk as per the original xl.meta prior to transition
for _, partInfo := range oi.Parts {
// Old uncompressed parts may not record ActualSize. Encrypted parts
// derive it from the written ciphertext in PutObjectPart; compressed
// parts need their recorded logical length, including valid zero sizes.
actualSize := partInfo.Size
if oi.IsCompressed() {
actualSize = partInfo.ActualSize
}
hr, err := hash.NewReader(ctx, io.LimitReader(gr, partInfo.Size), partInfo.Size, "", "", actualSize)
if err != nil {
return fail(err)
}
pInfo, err := er.PutObjectPart(ctx, bucket, object, res.UploadID, partInfo.Number, NewPutObjReader(hr), ObjectOptions{
IndexCB: func() []byte { return partInfo.Index },
Transition: TransitionOptions{attempt: attempt},
})
if err != nil {
return fail(err)
}
if pInfo.Size != partInfo.Size {
return fail(InvalidObjectState{Bucket: bucket, Object: object})
}
uploadedParts = append(uploadedParts, CompletePart{
PartNumber: pInfo.PartNumber,
ETag: pInfo.ETag,
})
}
_, err = er.CompleteMultipartUpload(ctx, bucket, object, res.UploadID, uploadedParts, ObjectOptions{
MTime: oi.ModTime,
NoLock: opts.NoLock,
Versioned: opts.Versioned,
VersionSuspended: opts.VersionSuspended,
Transition: opts.Transition,
})
if err != nil {
return fail(err)
}
completed = true
return nil
}
// DecomTieredObject - moves tiered object to another pool during decommissioning.
func (er erasureObjects) DecomTieredObject(ctx context.Context, bucket, object string, fi FileInfo, opts ObjectOptions) error {
if opts.UserDefined == nil {