1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
23 "git.arvados.org/arvados.git/sdk/go/arvados"
24 "github.com/aws/aws-sdk-go-v2/aws"
25 "github.com/aws/aws-sdk-go-v2/aws/awserr"
26 "github.com/aws/aws-sdk-go-v2/aws/defaults"
27 "github.com/aws/aws-sdk-go-v2/aws/ec2metadata"
28 "github.com/aws/aws-sdk-go-v2/aws/ec2rolecreds"
29 "github.com/aws/aws-sdk-go-v2/aws/endpoints"
30 "github.com/aws/aws-sdk-go-v2/service/s3"
31 "github.com/aws/aws-sdk-go-v2/service/s3/s3manager"
32 "github.com/prometheus/client_golang/prometheus"
33 "github.com/sirupsen/logrus"
37 driver["S3"] = news3Volume
41 s3DefaultReadTimeout = arvados.Duration(10 * time.Minute)
42 s3DefaultConnectTimeout = arvados.Duration(time.Minute)
43 maxClockSkew = 600 * time.Second
44 nearlyRFC1123 = "Mon, 2 Jan 2006 15:04:05 GMT"
45 s3downloaderPartSize = 6 * 1024 * 1024
46 s3downloaderReadConcurrency = 11
47 s3uploaderPartSize = 5 * 1024 * 1024
48 s3uploaderWriteConcurrency = 5
52 errS3TrashDisabled = fmt.Errorf("trash function is disabled because Collections.BlobTrashLifetime=0 and DriverParameters.UnsafeDelete=false")
53 s3AWSKeepBlockRegexp = regexp.MustCompile(`^[0-9a-f]{32}$`)
54 s3AWSZeroTime time.Time
57 // s3Volume implements Volume using an S3 bucket.
58 type s3Volume struct {
59 arvados.S3VolumeDriverParameters
60 AuthToken string // populated automatically when IAMRole is used
61 AuthExpiration time.Time // populated automatically when IAMRole is used
63 cluster *arvados.Cluster
65 logger logrus.FieldLogger
66 metrics *volumeMetricsVecs
67 bufferPool *bufferPool
73 // s3bucket wraps s3.bucket and counts I/O and API usage stats. The
74 // wrapped bucket can be replaced atomically with SetBucket in order
75 // to update credentials.
76 type s3Bucket struct {
79 stats s3awsbucketStats
83 func (v *s3Volume) isKeepBlock(s string) (string, bool) {
84 if v.PrefixLength > 0 && len(s) == v.PrefixLength+33 && s[:v.PrefixLength] == s[v.PrefixLength+1:v.PrefixLength*2+1] {
85 s = s[v.PrefixLength+1:]
87 return s, s3AWSKeepBlockRegexp.MatchString(s)
90 // Return the key used for a given loc. If PrefixLength==0 then
91 // key("abcdef0123") is "abcdef0123", if PrefixLength==3 then key is
92 // "abc/abcdef0123", etc.
93 func (v *s3Volume) key(loc string) string {
94 if v.PrefixLength > 0 && v.PrefixLength < len(loc)-1 {
95 return loc[:v.PrefixLength] + "/" + loc
101 func news3Volume(params newVolumeParams) (volume, error) {
103 cluster: params.Cluster,
104 volume: params.ConfigVolume,
105 metrics: params.MetricsVecs,
106 bufferPool: params.BufferPool,
108 err := json.Unmarshal(params.ConfigVolume.DriverParameters, v)
112 v.logger = params.Logger.WithField("Volume", v.DeviceID())
113 return v, v.check("")
116 func (v *s3Volume) translateError(err error) error {
117 if _, ok := err.(*aws.RequestCanceledError); ok {
118 return context.Canceled
119 } else if aerr, ok := err.(awserr.Error); ok {
120 if aerr.Code() == "NotFound" {
121 return os.ErrNotExist
122 } else if aerr.Code() == "NoSuchKey" {
123 return os.ErrNotExist
129 // safeCopy calls CopyObjectRequest, and checks the response to make
130 // sure the copy succeeded and updated the timestamp on the
131 // destination object
133 // (If something goes wrong during the copy, the error will be
134 // embedded in the 200 OK response)
135 func (v *s3Volume) safeCopy(dst, src string) error {
136 input := &s3.CopyObjectInput{
137 Bucket: aws.String(v.bucket.bucket),
138 ContentType: aws.String("application/octet-stream"),
139 CopySource: aws.String(v.bucket.bucket + "/" + src),
140 Key: aws.String(dst),
143 req := v.bucket.svc.CopyObjectRequest(input)
144 resp, err := req.Send(context.Background())
146 err = v.translateError(err)
147 if os.IsNotExist(err) {
149 } else if err != nil {
150 return fmt.Errorf("PutCopy(%q ← %q): %s", dst, v.bucket.bucket+"/"+src, err)
153 if resp.CopyObjectResult.LastModified == nil {
154 return fmt.Errorf("PutCopy succeeded but did not return a timestamp: %q: %s", resp.CopyObjectResult.LastModified, err)
155 } else if time.Now().Sub(*resp.CopyObjectResult.LastModified) > maxClockSkew {
156 return fmt.Errorf("PutCopy succeeded but returned an old timestamp: %q: %s", resp.CopyObjectResult.LastModified, resp.CopyObjectResult.LastModified)
161 func (v *s3Volume) check(ec2metadataHostname string) error {
163 return errors.New("DriverParameters: Bucket must be provided")
165 if v.IndexPageSize == 0 {
166 v.IndexPageSize = 1000
168 if v.RaceWindow < 0 {
169 return errors.New("DriverParameters: RaceWindow must not be negative")
173 return errors.New("DriverParameters: V2Signature is not supported")
176 defaultResolver := endpoints.NewDefaultResolver()
178 cfg := defaults.Config()
180 if v.Endpoint == "" && v.Region == "" {
181 return fmt.Errorf("AWS region or endpoint must be specified")
182 } else if v.Endpoint != "" || ec2metadataHostname != "" {
183 myCustomResolver := func(service, region string) (aws.Endpoint, error) {
184 if v.Endpoint != "" && service == "s3" {
187 SigningRegion: region,
189 } else if service == "ec2metadata" && ec2metadataHostname != "" {
191 URL: ec2metadataHostname,
194 return defaultResolver.ResolveEndpoint(service, region)
197 cfg.EndpointResolver = aws.EndpointResolverFunc(myCustomResolver)
200 // Endpoint is already specified (otherwise we would
201 // have errored out above), but Region is also
202 // required by the aws sdk, in order to determine
203 // SignatureVersions.
204 v.Region = "us-east-1"
206 cfg.Region = v.Region
208 // Zero timeouts mean "wait forever", which is a bad
209 // default. Default to long timeouts instead.
210 if v.ConnectTimeout == 0 {
211 v.ConnectTimeout = s3DefaultConnectTimeout
213 if v.ReadTimeout == 0 {
214 v.ReadTimeout = s3DefaultReadTimeout
217 creds := aws.NewChainProvider(
218 []aws.CredentialsProvider{
219 aws.NewStaticCredentialsProvider(v.AccessKeyID, v.SecretAccessKey, v.AuthToken),
220 ec2rolecreds.New(ec2metadata.New(cfg), func(opts *ec2rolecreds.ProviderOptions) {
221 // (from aws-sdk-go-v2 comments)
222 // "allow the credentials to trigger
223 // refreshing prior to the credentials
224 // actually expiring. This is
225 // beneficial so race conditions with
226 // expiring credentials do not cause
227 // request to fail unexpectedly due to
228 // ExpiredTokenException exceptions."
231 // https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/iam-roles-for-amazon-ec2.html)
232 // "We make new credentials available
233 // at least five minutes before the
234 // expiration of the old credentials."
235 opts.ExpiryWindow = 5 * time.Minute
239 cfg.Credentials = creds
241 v.bucket = &s3Bucket{
246 // Set up prometheus metrics
247 lbls := prometheus.Labels{"device_id": v.DeviceID()}
248 v.bucket.stats.opsCounters, v.bucket.stats.errCounters, v.bucket.stats.ioBytes = v.metrics.getCounterVecsFor(lbls)
253 // DeviceID returns a globally unique ID for the storage bucket.
254 func (v *s3Volume) DeviceID() string {
255 return "s3://" + v.Endpoint + "/" + v.Bucket
258 // EmptyTrash looks for trashed blocks that exceeded BlobTrashLifetime
259 // and deletes them from the volume.
260 func (v *s3Volume) EmptyTrash() {
261 var bytesInTrash, blocksInTrash, bytesDeleted, blocksDeleted int64
263 // Define "ready to delete" as "...when EmptyTrash started".
266 emptyOneKey := func(trash *s3.Object) {
267 key := strings.TrimPrefix(*trash.Key, "trash/")
268 loc, isblk := v.isKeepBlock(key)
272 atomic.AddInt64(&bytesInTrash, *trash.Size)
273 atomic.AddInt64(&blocksInTrash, 1)
275 trashT := *trash.LastModified
276 recent, err := v.head("recent/" + key)
277 if err != nil && os.IsNotExist(v.translateError(err)) {
278 v.logger.Warnf("EmptyTrash: found trash marker %q but no %q (%s); calling Untrash", *trash.Key, "recent/"+key, err)
279 err = v.BlockUntrash(loc)
281 v.logger.WithError(err).Errorf("EmptyTrash: Untrash(%q) failed", loc)
284 } else if err != nil {
285 v.logger.WithError(err).Warnf("EmptyTrash: HEAD %q failed", "recent/"+key)
288 if trashT.Sub(*recent.LastModified) < v.cluster.Collections.BlobSigningTTL.Duration() {
289 if age := startT.Sub(*recent.LastModified); age >= v.cluster.Collections.BlobSigningTTL.Duration()-time.Duration(v.RaceWindow) {
290 // recent/key is too old to protect
291 // loc from being Trashed again during
292 // the raceWindow that starts if we
293 // delete trash/X now.
295 // Note this means (TrashSweepInterval
296 // < BlobSigningTTL - raceWindow) is
297 // necessary to avoid starvation.
298 v.logger.Infof("EmptyTrash: detected old race for %q, calling fixRace + Touch", loc)
303 _, err := v.head(key)
304 if os.IsNotExist(err) {
305 v.logger.Infof("EmptyTrash: detected recent race for %q, calling fixRace", loc)
308 } else if err != nil {
309 v.logger.WithError(err).Warnf("EmptyTrash: HEAD %q failed", loc)
313 if startT.Sub(trashT) < v.cluster.Collections.BlobTrashLifetime.Duration() {
316 err = v.bucket.Del(*trash.Key)
318 v.logger.WithError(err).Errorf("EmptyTrash: error deleting %q", *trash.Key)
321 atomic.AddInt64(&bytesDeleted, *trash.Size)
322 atomic.AddInt64(&blocksDeleted, 1)
324 _, err = v.head(*trash.Key)
326 v.logger.Warnf("EmptyTrash: HEAD %q succeeded immediately after deleting %q", loc, loc)
329 if !os.IsNotExist(v.translateError(err)) {
330 v.logger.WithError(err).Warnf("EmptyTrash: HEAD %q failed", key)
333 err = v.bucket.Del("recent/" + key)
335 v.logger.WithError(err).Warnf("EmptyTrash: error deleting %q", "recent/"+key)
339 var wg sync.WaitGroup
340 todo := make(chan *s3.Object, v.cluster.Collections.BlobDeleteConcurrency)
341 for i := 0; i < v.cluster.Collections.BlobDeleteConcurrency; i++ {
345 for key := range todo {
351 trashL := s3awsLister{
355 PageSize: v.IndexPageSize,
356 Stats: &v.bucket.stats,
358 for trash := trashL.First(); trash != nil; trash = trashL.Next() {
364 if err := trashL.Error(); err != nil {
365 v.logger.WithError(err).Error("EmptyTrash: lister failed")
367 v.logger.Infof("EmptyTrash: stats for %v: Deleted %v bytes in %v blocks. Remaining in trash: %v bytes in %v blocks.", v.DeviceID(), bytesDeleted, blocksDeleted, bytesInTrash-bytesDeleted, blocksInTrash-blocksDeleted)
370 // fixRace(X) is called when "recent/X" exists but "X" doesn't
371 // exist. If the timestamps on "recent/X" and "trash/X" indicate there
372 // was a race between Put and Trash, fixRace recovers from the race by
373 // Untrashing the block.
374 func (v *s3Volume) fixRace(key string) bool {
375 trash, err := v.head("trash/" + key)
377 if !os.IsNotExist(v.translateError(err)) {
378 v.logger.WithError(err).Errorf("fixRace: HEAD %q failed", "trash/"+key)
383 recent, err := v.head("recent/" + key)
385 v.logger.WithError(err).Errorf("fixRace: HEAD %q failed", "recent/"+key)
389 recentTime := *recent.LastModified
390 trashTime := *trash.LastModified
391 ageWhenTrashed := trashTime.Sub(recentTime)
392 if ageWhenTrashed >= v.cluster.Collections.BlobSigningTTL.Duration() {
393 // No evidence of a race: block hasn't been written
394 // since it became eligible for Trash. No fix needed.
398 v.logger.Infof("fixRace: %q: trashed at %s but touched at %s (age when trashed = %s < %s)", key, trashTime, recentTime, ageWhenTrashed, v.cluster.Collections.BlobSigningTTL)
399 v.logger.Infof("fixRace: copying %q to %q to recover from race between Put/Touch and Trash", "recent/"+key, key)
400 err = v.safeCopy(key, "trash/"+key)
402 v.logger.WithError(err).Error("fixRace: copy failed")
408 func (v *s3Volume) head(key string) (result *s3.HeadObjectOutput, err error) {
409 input := &s3.HeadObjectInput{
410 Bucket: aws.String(v.bucket.bucket),
411 Key: aws.String(key),
414 req := v.bucket.svc.HeadObjectRequest(input)
415 res, err := req.Send(context.TODO())
417 v.bucket.stats.TickOps("head")
418 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.HeadOps)
419 v.bucket.stats.TickErr(err)
422 return nil, v.translateError(err)
424 result = res.HeadObjectOutput
428 // BlockRead reads a Keep block that has been stored as a block blob
430 func (v *s3Volume) BlockRead(ctx context.Context, hash string, w io.WriterAt) error {
432 err := v.readWorker(ctx, key, w)
434 err = v.translateError(err)
435 if !os.IsNotExist(err) {
439 _, err = v.head("recent/" + key)
440 err = v.translateError(err)
442 // If we can't read recent/X, there's no point in
443 // trying fixRace. Give up.
451 err = v.readWorker(ctx, key, w)
453 v.logger.Warnf("reading %s after successful fixRace: %s", hash, err)
454 err = v.translateError(err)
461 func (v *s3Volume) readWorker(ctx context.Context, key string, dst io.WriterAt) error {
462 downloader := s3manager.NewDownloaderWithClient(v.bucket.svc, func(u *s3manager.Downloader) {
463 u.PartSize = s3downloaderPartSize
464 u.Concurrency = s3downloaderReadConcurrency
466 count, err := downloader.DownloadWithContext(ctx, dst, &s3.GetObjectInput{
467 Bucket: aws.String(v.bucket.bucket),
468 Key: aws.String(key),
470 v.bucket.stats.TickOps("get")
471 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.GetOps)
472 v.bucket.stats.TickErr(err)
473 v.bucket.stats.TickInBytes(uint64(count))
474 return v.translateError(err)
477 func (v *s3Volume) writeObject(ctx context.Context, key string, r io.Reader) error {
479 // r == nil leads to a memory violation in func readFillBuf in
480 // aws-sdk-go-v2@v0.23.0/service/s3/s3manager/upload.go
481 r = bytes.NewReader(nil)
484 uploadInput := s3manager.UploadInput{
485 Bucket: aws.String(v.bucket.bucket),
486 Key: aws.String(key),
490 if loc, ok := v.isKeepBlock(key); ok {
491 var contentMD5 string
492 md5, err := hex.DecodeString(loc)
494 return v.translateError(err)
496 contentMD5 = base64.StdEncoding.EncodeToString(md5)
497 uploadInput.ContentMD5 = &contentMD5
500 // Experimentation indicated that using concurrency 5 yields the best
501 // throughput, better than higher concurrency (10 or 13) by ~5%.
502 // Defining u.BufferProvider = s3manager.NewBufferedReadSeekerWriteToPool(64 * 1024 * 1024)
503 // is detrimental to throughput (minus ~15%).
504 uploader := s3manager.NewUploaderWithClient(v.bucket.svc, func(u *s3manager.Uploader) {
505 u.PartSize = s3uploaderPartSize
506 u.Concurrency = s3uploaderWriteConcurrency
509 // Unlike the goamz S3 driver, we don't need to precompute ContentSHA256:
510 // the aws-sdk-go v2 SDK uses a ReadSeeker to avoid having to copy the
511 // block, so there is no extra memory use to be concerned about. See
512 // makeSha256Reader in aws/signer/v4/v4.go. In fact, we explicitly disable
513 // calculating the Sha-256 because we don't need it; we already use md5sum
514 // hashes that match the name of the block.
515 _, err := uploader.UploadWithContext(ctx, &uploadInput, s3manager.WithUploaderRequestOptions(func(r *aws.Request) {
516 r.HTTPRequest.Header.Set("X-Amz-Content-Sha256", "UNSIGNED-PAYLOAD")
519 v.bucket.stats.TickOps("put")
520 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.PutOps)
521 v.bucket.stats.TickErr(err)
523 return v.translateError(err)
526 // Put writes a block.
527 func (v *s3Volume) BlockWrite(ctx context.Context, hash string, data []byte) error {
528 // Do not use putWithPipe here; we want to pass an io.ReadSeeker to the S3
529 // sdk to avoid memory allocation there. See #17339 for more information.
530 rdr := bytes.NewReader(data)
531 r := newCountingReaderAtSeeker(rdr, v.bucket.stats.TickOutBytes)
533 err := v.writeObject(ctx, key, r)
537 return v.writeObject(ctx, "recent/"+key, nil)
540 type s3awsLister struct {
541 Logger logrus.FieldLogger
545 Stats *s3awsbucketStats
546 ContinuationToken string
551 // First fetches the first page and returns the first item. It returns
552 // nil if the response is the empty set or an error occurs.
553 func (lister *s3awsLister) First() *s3.Object {
558 // Next returns the next item, fetching the next page if necessary. It
559 // returns nil if the last available item has already been fetched, or
561 func (lister *s3awsLister) Next() *s3.Object {
562 if len(lister.buf) == 0 && lister.ContinuationToken != "" {
568 // Return the most recent error encountered by First or Next.
569 func (lister *s3awsLister) Error() error {
573 func (lister *s3awsLister) getPage() {
574 lister.Stats.TickOps("list")
575 lister.Stats.Tick(&lister.Stats.Ops, &lister.Stats.ListOps)
577 var input *s3.ListObjectsV2Input
578 if lister.ContinuationToken == "" {
579 input = &s3.ListObjectsV2Input{
580 Bucket: aws.String(lister.Bucket.bucket),
581 MaxKeys: aws.Int64(int64(lister.PageSize)),
582 Prefix: aws.String(lister.Prefix),
585 input = &s3.ListObjectsV2Input{
586 Bucket: aws.String(lister.Bucket.bucket),
587 MaxKeys: aws.Int64(int64(lister.PageSize)),
588 Prefix: aws.String(lister.Prefix),
589 ContinuationToken: &lister.ContinuationToken,
593 req := lister.Bucket.svc.ListObjectsV2Request(input)
594 resp, err := req.Send(context.Background())
596 if aerr, ok := err.(awserr.Error); ok {
604 if *resp.IsTruncated {
605 lister.ContinuationToken = *resp.NextContinuationToken
607 lister.ContinuationToken = ""
609 lister.buf = make([]s3.Object, 0, len(resp.Contents))
610 for _, key := range resp.Contents {
611 if !strings.HasPrefix(*key.Key, lister.Prefix) {
612 lister.Logger.Warnf("s3awsLister: S3 Bucket.List(prefix=%q) returned key %q", lister.Prefix, *key.Key)
615 lister.buf = append(lister.buf, key)
619 func (lister *s3awsLister) pop() (k *s3.Object) {
620 if len(lister.buf) > 0 {
622 lister.buf = lister.buf[1:]
627 // Index writes a complete list of locators with the given prefix
628 // for which Get() can retrieve data.
629 func (v *s3Volume) Index(ctx context.Context, prefix string, writer io.Writer) error {
630 prefix = v.key(prefix)
631 // Use a merge sort to find matching sets of X and recent/X.
632 dataL := s3awsLister{
636 PageSize: v.IndexPageSize,
637 Stats: &v.bucket.stats,
639 recentL := s3awsLister{
642 Prefix: "recent/" + prefix,
643 PageSize: v.IndexPageSize,
644 Stats: &v.bucket.stats,
646 for data, recent := dataL.First(), recentL.First(); data != nil && dataL.Error() == nil; data = dataL.Next() {
647 if ctx.Err() != nil {
650 if *data.Key >= "g" {
651 // Conveniently, "recent/*" and "trash/*" are
652 // lexically greater than all hex-encoded data
653 // hashes, so stopping here avoids iterating
654 // over all of them needlessly with dataL.
657 loc, isblk := v.isKeepBlock(*data.Key)
662 // stamp is the list entry we should use to report the
663 // last-modified time for this data block: it will be
664 // the recent/X entry if one exists, otherwise the
665 // entry for the data block itself.
668 // Advance to the corresponding recent/X marker, if any
669 for recent != nil && recentL.Error() == nil {
670 if cmp := strings.Compare((*recent.Key)[7:], *data.Key); cmp < 0 {
671 recent = recentL.Next()
675 recent = recentL.Next()
678 // recent/X marker is missing: we'll
679 // use the timestamp on the data
684 if err := recentL.Error(); err != nil {
687 // We truncate sub-second precision here. Otherwise
688 // timestamps will never match the RFC1123-formatted
689 // Last-Modified values parsed by Mtime().
690 fmt.Fprintf(writer, "%s+%d %d\n", loc, *data.Size, stamp.LastModified.Unix()*1000000000)
695 // Mtime returns the stored timestamp for the given locator.
696 func (v *s3Volume) Mtime(loc string) (time.Time, error) {
698 _, err := v.head(key)
700 return s3AWSZeroTime, v.translateError(err)
702 resp, err := v.head("recent/" + key)
703 err = v.translateError(err)
704 if os.IsNotExist(err) {
705 // The data object X exists, but recent/X is missing.
706 err = v.writeObject(context.Background(), "recent/"+key, nil)
708 v.logger.WithError(err).Errorf("error creating %q", "recent/"+key)
709 return s3AWSZeroTime, v.translateError(err)
711 v.logger.Infof("Mtime: created %q to migrate existing block to new storage scheme", "recent/"+key)
712 resp, err = v.head("recent/" + key)
714 v.logger.WithError(err).Errorf("HEAD failed after creating %q", "recent/"+key)
715 return s3AWSZeroTime, v.translateError(err)
717 } else if err != nil {
718 // HEAD recent/X failed for some other reason.
719 return s3AWSZeroTime, err
721 return *resp.LastModified, err
724 // InternalStats returns bucket I/O and API call counters.
725 func (v *s3Volume) InternalStats() interface{} {
726 return &v.bucket.stats
729 // BlockTouch sets the timestamp for the given locator to the current time.
730 func (v *s3Volume) BlockTouch(hash string) error {
732 _, err := v.head(key)
733 err = v.translateError(err)
734 if os.IsNotExist(err) && v.fixRace(key) {
735 // The data object got trashed in a race, but fixRace
737 } else if err != nil {
740 err = v.writeObject(context.Background(), "recent/"+key, nil)
741 return v.translateError(err)
744 // checkRaceWindow returns a non-nil error if trash/key is, or might
745 // be, in the race window (i.e., it's not safe to trash key).
746 func (v *s3Volume) checkRaceWindow(key string) error {
747 resp, err := v.head("trash/" + key)
748 err = v.translateError(err)
749 if os.IsNotExist(err) {
750 // OK, trash/X doesn't exist so we're not in the race
753 } else if err != nil {
754 // Error looking up trash/X. We don't know whether
755 // we're in the race window
758 t := resp.LastModified
759 safeWindow := t.Add(v.cluster.Collections.BlobTrashLifetime.Duration()).Sub(time.Now().Add(time.Duration(v.RaceWindow)))
761 // We can't count on "touch trash/X" to prolong
762 // trash/X's lifetime. The new timestamp might not
763 // become visible until now+raceWindow, and EmptyTrash
764 // is allowed to delete trash/X before then.
765 return fmt.Errorf("%s: same block is already in trash, and safe window ended %s ago", key, -safeWindow)
767 // trash/X exists, but it won't be eligible for deletion until
768 // after now+raceWindow, so it's safe to overwrite it.
772 func (b *s3Bucket) Del(path string) error {
773 input := &s3.DeleteObjectInput{
774 Bucket: aws.String(b.bucket),
775 Key: aws.String(path),
777 req := b.svc.DeleteObjectRequest(input)
778 _, err := req.Send(context.Background())
779 b.stats.TickOps("delete")
780 b.stats.Tick(&b.stats.Ops, &b.stats.DelOps)
785 // Trash a Keep block.
786 func (v *s3Volume) BlockTrash(loc string) error {
787 if t, err := v.Mtime(loc); err != nil {
789 } else if time.Since(t) < v.cluster.Collections.BlobSigningTTL.Duration() {
793 if v.cluster.Collections.BlobTrashLifetime == 0 {
795 return errS3TrashDisabled
797 return v.translateError(v.bucket.Del(key))
799 err := v.checkRaceWindow(key)
803 err = v.safeCopy("trash/"+key, key)
807 return v.translateError(v.bucket.Del(key))
810 // BlockUntrash moves block from trash back into store
811 func (v *s3Volume) BlockUntrash(hash string) error {
813 err := v.safeCopy(key, "trash/"+key)
817 err = v.writeObject(context.Background(), "recent/"+key, nil)
818 return v.translateError(err)
821 type s3awsbucketStats struct {
831 func (s *s3awsbucketStats) TickErr(err error) {
835 errType := fmt.Sprintf("%T", err)
836 if aerr, ok := err.(awserr.Error); ok {
837 if reqErr, ok := err.(awserr.RequestFailure); ok {
838 // A service error occurred
839 errType = errType + fmt.Sprintf(" %d %s", reqErr.StatusCode(), aerr.Code())
841 errType = errType + fmt.Sprintf(" 000 %s", aerr.Code())
844 s.statsTicker.TickErr(err, errType)