add DeleteBulk support, add sufficient deadlines per rename() (#20185)

deadlines per moveToTrash() allows for a more granular timeout
approach for syscalls, instead of an aggregate timeout.

This PR also enhances multipart state cleanup to be optimal by
removing 100's of multipart network rename() calls into single
network call.
This commit is contained in:
Harshavardhana
2024-07-29 18:56:40 -07:00
committed by GitHub
parent 673df6d517
commit 80ff907d08
12 changed files with 909 additions and 251 deletions

View File

@@ -103,10 +103,9 @@ func (er erasureObjects) checkUploadIDExists(ctx context.Context, bucket, object
return fi, partsMetadata, err
}
// Removes part.meta given by partName belonging to a multipart upload from minioMetaBucket
func (er erasureObjects) removePartMeta(bucket, object, uploadID, dataDir string, partNumber int) {
uploadIDPath := er.getUploadIDDir(bucket, object, uploadID)
curpartPath := pathJoin(uploadIDPath, dataDir, fmt.Sprintf("part.%d", partNumber))
// cleanMultipartPath removes all extraneous files and parts from the multipart folder, this is used per CompleteMultipart.
// do not use this function outside of completeMultipartUpload()
func (er erasureObjects) cleanupMultipartPath(ctx context.Context, paths ...string) {
storageDisks := er.getDisks()
g := errgroup.WithNErrs(len(storageDisks))
@@ -116,42 +115,7 @@ func (er erasureObjects) removePartMeta(bucket, object, uploadID, dataDir string
}
index := index
g.Go(func() error {
_ = storageDisks[index].Delete(context.TODO(), minioMetaMultipartBucket, curpartPath+".meta", DeleteOptions{
Recursive: false,
Immediate: false,
})
return nil
}, index)
}
g.Wait()
}
// Removes part given by partName belonging to a multipart upload from minioMetaBucket
func (er erasureObjects) removeObjectPart(bucket, object, uploadID, dataDir string, partNumber int) {
uploadIDPath := er.getUploadIDDir(bucket, object, uploadID)
curpartPath := pathJoin(uploadIDPath, dataDir, fmt.Sprintf("part.%d", partNumber))
storageDisks := er.getDisks()
g := errgroup.WithNErrs(len(storageDisks))
for index, disk := range storageDisks {
if disk == nil {
continue
}
index := index
g.Go(func() error {
// Ignoring failure to remove parts that weren't present in CompleteMultipartUpload
// requests. xl.meta is the authoritative source of truth on which parts constitute
// the object. The presence of parts that don't belong in the object doesn't affect correctness.
_ = storageDisks[index].Delete(context.TODO(), minioMetaMultipartBucket, curpartPath, DeleteOptions{
Recursive: false,
Immediate: false,
})
_ = storageDisks[index].Delete(context.TODO(), minioMetaMultipartBucket, curpartPath+".meta", DeleteOptions{
Recursive: false,
Immediate: false,
})
_ = storageDisks[index].DeleteBulk(ctx, minioMetaMultipartBucket, paths...)
return nil
}, index)
}
@@ -1359,10 +1323,10 @@ func (er erasureObjects) CompleteMultipartUpload(ctx context.Context, bucket str
}
}
paths := make([]string, 0, len(currentFI.Parts))
// Remove parts that weren't present in CompleteMultipartUpload request.
for _, curpart := range currentFI.Parts {
// Remove part.meta which is not needed anymore.
er.removePartMeta(bucket, object, uploadID, currentFI.DataDir, curpart.Number)
paths = append(paths, pathJoin(uploadIDPath, currentFI.DataDir, fmt.Sprintf("part.%d.meta", curpart.Number)))
if objectPartIndex(fi.Parts, curpart.Number) == -1 {
// Delete the missing part files. e.g,
@@ -1371,10 +1335,12 @@ func (er erasureObjects) CompleteMultipartUpload(ctx context.Context, bucket str
// Request 3: PutObjectPart 2
// Request 4: CompleteMultipartUpload --part 2
// N.B. 1st part is not present. This part should be removed from the storage.
er.removeObjectPart(bucket, object, uploadID, currentFI.DataDir, curpart.Number)
paths = append(paths, pathJoin(uploadIDPath, currentFI.DataDir, fmt.Sprintf("part.%d", curpart.Number)))
}
}
er.cleanupMultipartPath(ctx, paths...) // cleanup all part.N.meta, and skipped part.N's before final rename().
defer func() {
if err == nil {
er.deleteAll(context.Background(), minioMetaMultipartBucket, uploadIDPath)