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"
36 // S3AWSVolume implements Volume using an S3 bucket.
37 type S3AWSVolume struct {
38 arvados.S3VolumeDriverParameters
39 AuthToken string // populated automatically when IAMRole is used
40 AuthExpiration time.Time // populated automatically when IAMRole is used
42 cluster *arvados.Cluster
44 logger logrus.FieldLogger
45 metrics *volumeMetricsVecs
51 // s3bucket wraps s3.bucket and counts I/O and API usage stats. The
52 // wrapped bucket can be replaced atomically with SetBucket in order
53 // to update credentials.
54 type s3AWSbucket struct {
57 stats s3awsbucketStats
61 // chooseS3VolumeDriver distinguishes between the old goamz driver and
62 // aws-sdk-go based on the UseAWSS3v2Driver feature flag
63 func chooseS3VolumeDriver(cluster *arvados.Cluster, volume arvados.Volume, logger logrus.FieldLogger, metrics *volumeMetricsVecs) (Volume, error) {
64 v := &S3Volume{cluster: cluster, volume: volume, metrics: metrics}
65 // Default value will be overriden if it happens to be defined in the config
66 v.S3VolumeDriverParameters.UseAWSS3v2Driver = true
67 err := json.Unmarshal(volume.DriverParameters, v)
71 if v.UseAWSS3v2Driver {
72 logger.Debugln("Using AWS S3 v2 driver")
73 return newS3AWSVolume(cluster, volume, logger, metrics)
75 logger.Debugln("Using goamz S3 driver")
76 return newS3Volume(cluster, volume, logger, metrics)
80 PartSize = 5 * 1024 * 1024
85 var s3AWSKeepBlockRegexp = regexp.MustCompile(`^[0-9a-f]{32}$`)
86 var s3AWSZeroTime time.Time
88 func (v *S3AWSVolume) isKeepBlock(s string) (string, bool) {
89 if v.PrefixLength > 0 && len(s) == v.PrefixLength+33 && s[:v.PrefixLength] == s[v.PrefixLength+1:v.PrefixLength*2+1] {
90 s = s[v.PrefixLength+1:]
92 return s, s3AWSKeepBlockRegexp.MatchString(s)
95 // Return the key used for a given loc. If PrefixLength==0 then
96 // key("abcdef0123") is "abcdef0123", if PrefixLength==3 then key is
97 // "abc/abcdef0123", etc.
98 func (v *S3AWSVolume) key(loc string) string {
99 if v.PrefixLength > 0 && v.PrefixLength < len(loc)-1 {
100 return loc[:v.PrefixLength] + "/" + loc
106 func newS3AWSVolume(cluster *arvados.Cluster, volume arvados.Volume, logger logrus.FieldLogger, metrics *volumeMetricsVecs) (Volume, error) {
107 v := &S3AWSVolume{cluster: cluster, volume: volume, metrics: metrics}
108 err := json.Unmarshal(volume.DriverParameters, v)
112 v.logger = logger.WithField("Volume", v.String())
113 return v, v.check("")
116 func (v *S3AWSVolume) 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 *S3AWSVolume) 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 *S3AWSVolume) 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)),
223 cfg.Credentials = creds
225 v.bucket = &s3AWSbucket{
230 // Set up prometheus metrics
231 lbls := prometheus.Labels{"device_id": v.GetDeviceID()}
232 v.bucket.stats.opsCounters, v.bucket.stats.errCounters, v.bucket.stats.ioBytes = v.metrics.getCounterVecsFor(lbls)
237 // String implements fmt.Stringer.
238 func (v *S3AWSVolume) String() string {
239 return fmt.Sprintf("s3-bucket:%+q", v.Bucket)
242 // GetDeviceID returns a globally unique ID for the storage bucket.
243 func (v *S3AWSVolume) GetDeviceID() string {
244 return "s3://" + v.Endpoint + "/" + v.Bucket
247 // Compare the given data with the stored data.
248 func (v *S3AWSVolume) Compare(ctx context.Context, loc string, expect []byte) error {
250 errChan := make(chan error, 1)
252 _, err := v.head("recent/" + key)
259 case err = <-errChan:
262 // Checking for the key itself here would interfere
263 // with future GET requests.
265 // On AWS, if X doesn't exist, a HEAD or GET request
266 // for X causes X's non-existence to be cached. Thus,
267 // if we test for X, then create X and return a
268 // signature to our client, the client might still get
269 // 404 from all keepstores when trying to read it.
271 // To avoid this, we avoid doing HEAD X or GET X until
272 // we know X has been written.
274 // Note that X might exist even though recent/X
275 // doesn't: for example, the response to HEAD recent/X
276 // might itself come from a stale cache. In such
277 // cases, we will return a false negative and
278 // PutHandler might needlessly create another replica
279 // on a different volume. That's not ideal, but it's
280 // better than passing the eventually-consistent
281 // problem on to our clients.
282 return v.translateError(err)
285 input := &s3.GetObjectInput{
286 Bucket: aws.String(v.bucket.bucket),
287 Key: aws.String(key),
290 req := v.bucket.svc.GetObjectRequest(input)
291 result, err := req.Send(ctx)
293 return v.translateError(err)
295 return v.translateError(compareReaderWithBuf(ctx, result.Body, expect, loc[:32]))
298 // EmptyTrash looks for trashed blocks that exceeded BlobTrashLifetime
299 // and deletes them from the volume.
300 func (v *S3AWSVolume) EmptyTrash() {
301 if v.cluster.Collections.BlobDeleteConcurrency < 1 {
305 var bytesInTrash, blocksInTrash, bytesDeleted, blocksDeleted int64
307 // Define "ready to delete" as "...when EmptyTrash started".
310 emptyOneKey := func(trash *s3.Object) {
311 key := strings.TrimPrefix(*trash.Key, "trash/")
312 loc, isblk := v.isKeepBlock(key)
316 atomic.AddInt64(&bytesInTrash, *trash.Size)
317 atomic.AddInt64(&blocksInTrash, 1)
319 trashT := *trash.LastModified
320 recent, err := v.head("recent/" + key)
321 if err != nil && os.IsNotExist(v.translateError(err)) {
322 v.logger.Warnf("EmptyTrash: found trash marker %q but no %q (%s); calling Untrash", *trash.Key, "recent/"+key, err)
325 v.logger.WithError(err).Errorf("EmptyTrash: Untrash(%q) failed", loc)
328 } else if err != nil {
329 v.logger.WithError(err).Warnf("EmptyTrash: HEAD %q failed", "recent/"+key)
332 if trashT.Sub(*recent.LastModified) < v.cluster.Collections.BlobSigningTTL.Duration() {
333 if age := startT.Sub(*recent.LastModified); age >= v.cluster.Collections.BlobSigningTTL.Duration()-time.Duration(v.RaceWindow) {
334 // recent/key is too old to protect
335 // loc from being Trashed again during
336 // the raceWindow that starts if we
337 // delete trash/X now.
339 // Note this means (TrashSweepInterval
340 // < BlobSigningTTL - raceWindow) is
341 // necessary to avoid starvation.
342 v.logger.Infof("EmptyTrash: detected old race for %q, calling fixRace + Touch", loc)
347 _, err := v.head(key)
348 if os.IsNotExist(err) {
349 v.logger.Infof("EmptyTrash: detected recent race for %q, calling fixRace", loc)
352 } else if err != nil {
353 v.logger.WithError(err).Warnf("EmptyTrash: HEAD %q failed", loc)
357 if startT.Sub(trashT) < v.cluster.Collections.BlobTrashLifetime.Duration() {
360 err = v.bucket.Del(*trash.Key)
362 v.logger.WithError(err).Errorf("EmptyTrash: error deleting %q", *trash.Key)
365 atomic.AddInt64(&bytesDeleted, *trash.Size)
366 atomic.AddInt64(&blocksDeleted, 1)
368 _, err = v.head(*trash.Key)
370 v.logger.Warnf("EmptyTrash: HEAD %q succeeded immediately after deleting %q", loc, loc)
373 if !os.IsNotExist(v.translateError(err)) {
374 v.logger.WithError(err).Warnf("EmptyTrash: HEAD %q failed", key)
377 err = v.bucket.Del("recent/" + key)
379 v.logger.WithError(err).Warnf("EmptyTrash: error deleting %q", "recent/"+key)
383 var wg sync.WaitGroup
384 todo := make(chan *s3.Object, v.cluster.Collections.BlobDeleteConcurrency)
385 for i := 0; i < v.cluster.Collections.BlobDeleteConcurrency; i++ {
389 for key := range todo {
395 trashL := s3awsLister{
399 PageSize: v.IndexPageSize,
400 Stats: &v.bucket.stats,
402 for trash := trashL.First(); trash != nil; trash = trashL.Next() {
408 if err := trashL.Error(); err != nil {
409 v.logger.WithError(err).Error("EmptyTrash: lister failed")
411 v.logger.Infof("EmptyTrash: stats for %v: Deleted %v bytes in %v blocks. Remaining in trash: %v bytes in %v blocks.", v.String(), bytesDeleted, blocksDeleted, bytesInTrash-bytesDeleted, blocksInTrash-blocksDeleted)
414 // fixRace(X) is called when "recent/X" exists but "X" doesn't
415 // exist. If the timestamps on "recent/X" and "trash/X" indicate there
416 // was a race between Put and Trash, fixRace recovers from the race by
417 // Untrashing the block.
418 func (v *S3AWSVolume) fixRace(key string) bool {
419 trash, err := v.head("trash/" + key)
421 if !os.IsNotExist(v.translateError(err)) {
422 v.logger.WithError(err).Errorf("fixRace: HEAD %q failed", "trash/"+key)
427 recent, err := v.head("recent/" + key)
429 v.logger.WithError(err).Errorf("fixRace: HEAD %q failed", "recent/"+key)
433 recentTime := *recent.LastModified
434 trashTime := *trash.LastModified
435 ageWhenTrashed := trashTime.Sub(recentTime)
436 if ageWhenTrashed >= v.cluster.Collections.BlobSigningTTL.Duration() {
437 // No evidence of a race: block hasn't been written
438 // since it became eligible for Trash. No fix needed.
442 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)
443 v.logger.Infof("fixRace: copying %q to %q to recover from race between Put/Touch and Trash", "recent/"+key, key)
444 err = v.safeCopy(key, "trash/"+key)
446 v.logger.WithError(err).Error("fixRace: copy failed")
452 func (v *S3AWSVolume) head(key string) (result *s3.HeadObjectOutput, err error) {
453 input := &s3.HeadObjectInput{
454 Bucket: aws.String(v.bucket.bucket),
455 Key: aws.String(key),
458 req := v.bucket.svc.HeadObjectRequest(input)
459 res, err := req.Send(context.TODO())
461 v.bucket.stats.TickOps("head")
462 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.HeadOps)
463 v.bucket.stats.TickErr(err)
466 return nil, v.translateError(err)
468 result = res.HeadObjectOutput
472 // Get a block: copy the block data into buf, and return the number of
474 func (v *S3AWSVolume) Get(ctx context.Context, loc string, buf []byte) (int, error) {
475 // Do not use getWithPipe here: the BlockReader interface does not pass
476 // through 'buf []byte', and we don't want to allocate two buffers for each
477 // read request. Instead, use a version of ReadBlock that accepts 'buf []byte'
480 count, err := v.readWorker(ctx, key, buf)
485 err = v.translateError(err)
486 if !os.IsNotExist(err) {
490 _, err = v.head("recent/" + key)
491 err = v.translateError(err)
493 // If we can't read recent/X, there's no point in
494 // trying fixRace. Give up.
502 count, err = v.readWorker(ctx, key, buf)
504 v.logger.Warnf("reading %s after successful fixRace: %s", loc, err)
505 err = v.translateError(err)
511 func (v *S3AWSVolume) readWorker(ctx context.Context, key string, buf []byte) (int, error) {
512 awsBuf := aws.NewWriteAtBuffer(buf)
513 downloader := s3manager.NewDownloaderWithClient(v.bucket.svc, func(u *s3manager.Downloader) {
514 u.PartSize = PartSize
515 u.Concurrency = ReadConcurrency
518 v.logger.Debugf("Partsize: %d; Concurrency: %d\n", downloader.PartSize, downloader.Concurrency)
520 count, err := downloader.DownloadWithContext(ctx, awsBuf, &s3.GetObjectInput{
521 Bucket: aws.String(v.bucket.bucket),
522 Key: aws.String(key),
524 v.bucket.stats.TickOps("get")
525 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.GetOps)
526 v.bucket.stats.TickErr(err)
527 v.bucket.stats.TickInBytes(uint64(count))
528 return int(count), v.translateError(err)
531 func (v *S3AWSVolume) writeObject(ctx context.Context, key string, r io.Reader) error {
533 // r == nil leads to a memory violation in func readFillBuf in
534 // aws-sdk-go-v2@v0.23.0/service/s3/s3manager/upload.go
535 r = bytes.NewReader(nil)
538 uploadInput := s3manager.UploadInput{
539 Bucket: aws.String(v.bucket.bucket),
540 Key: aws.String(key),
544 if loc, ok := v.isKeepBlock(key); ok {
545 var contentMD5 string
546 md5, err := hex.DecodeString(loc)
548 return v.translateError(err)
550 contentMD5 = base64.StdEncoding.EncodeToString(md5)
551 uploadInput.ContentMD5 = &contentMD5
554 // Experimentation indicated that using concurrency 5 yields the best
555 // throughput, better than higher concurrency (10 or 13) by ~5%.
556 // Defining u.BufferProvider = s3manager.NewBufferedReadSeekerWriteToPool(64 * 1024 * 1024)
557 // is detrimental to througput (minus ~15%).
558 uploader := s3manager.NewUploaderWithClient(v.bucket.svc, func(u *s3manager.Uploader) {
559 u.PartSize = PartSize
560 u.Concurrency = WriteConcurrency
563 // Unlike the goamz S3 driver, we don't need to precompute ContentSHA256:
564 // the aws-sdk-go v2 SDK uses a ReadSeeker to avoid having to copy the
565 // block, so there is no extra memory use to be concerned about. See
566 // makeSha256Reader in aws/signer/v4/v4.go. In fact, we explicitly disable
567 // calculating the Sha-256 because we don't need it; we already use md5sum
568 // hashes that match the name of the block.
569 _, err := uploader.UploadWithContext(ctx, &uploadInput, s3manager.WithUploaderRequestOptions(func(r *aws.Request) {
570 r.HTTPRequest.Header.Set("X-Amz-Content-Sha256", "UNSIGNED-PAYLOAD")
573 v.bucket.stats.TickOps("put")
574 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.PutOps)
575 v.bucket.stats.TickErr(err)
577 return v.translateError(err)
580 // Put writes a block.
581 func (v *S3AWSVolume) Put(ctx context.Context, loc string, block []byte) error {
582 // Do not use putWithPipe here; we want to pass an io.ReadSeeker to the S3
583 // sdk to avoid memory allocation there. See #17339 for more information.
584 if v.volume.ReadOnly {
585 return MethodDisabledError
588 rdr := bytes.NewReader(block)
589 r := NewCountingReaderAtSeeker(rdr, v.bucket.stats.TickOutBytes)
591 err := v.writeObject(ctx, key, r)
595 return v.writeObject(ctx, "recent/"+key, nil)
598 type s3awsLister struct {
599 Logger logrus.FieldLogger
603 Stats *s3awsbucketStats
604 ContinuationToken string
609 // First fetches the first page and returns the first item. It returns
610 // nil if the response is the empty set or an error occurs.
611 func (lister *s3awsLister) First() *s3.Object {
616 // Next returns the next item, fetching the next page if necessary. It
617 // returns nil if the last available item has already been fetched, or
619 func (lister *s3awsLister) Next() *s3.Object {
620 if len(lister.buf) == 0 && lister.ContinuationToken != "" {
626 // Return the most recent error encountered by First or Next.
627 func (lister *s3awsLister) Error() error {
631 func (lister *s3awsLister) getPage() {
632 lister.Stats.TickOps("list")
633 lister.Stats.Tick(&lister.Stats.Ops, &lister.Stats.ListOps)
635 var input *s3.ListObjectsV2Input
636 if lister.ContinuationToken == "" {
637 input = &s3.ListObjectsV2Input{
638 Bucket: aws.String(lister.Bucket.bucket),
639 MaxKeys: aws.Int64(int64(lister.PageSize)),
640 Prefix: aws.String(lister.Prefix),
643 input = &s3.ListObjectsV2Input{
644 Bucket: aws.String(lister.Bucket.bucket),
645 MaxKeys: aws.Int64(int64(lister.PageSize)),
646 Prefix: aws.String(lister.Prefix),
647 ContinuationToken: &lister.ContinuationToken,
651 req := lister.Bucket.svc.ListObjectsV2Request(input)
652 resp, err := req.Send(context.Background())
654 if aerr, ok := err.(awserr.Error); ok {
662 if *resp.IsTruncated {
663 lister.ContinuationToken = *resp.NextContinuationToken
665 lister.ContinuationToken = ""
667 lister.buf = make([]s3.Object, 0, len(resp.Contents))
668 for _, key := range resp.Contents {
669 if !strings.HasPrefix(*key.Key, lister.Prefix) {
670 lister.Logger.Warnf("s3awsLister: S3 Bucket.List(prefix=%q) returned key %q", lister.Prefix, *key.Key)
673 lister.buf = append(lister.buf, key)
677 func (lister *s3awsLister) pop() (k *s3.Object) {
678 if len(lister.buf) > 0 {
680 lister.buf = lister.buf[1:]
685 // IndexTo writes a complete list of locators with the given prefix
686 // for which Get() can retrieve data.
687 func (v *S3AWSVolume) IndexTo(prefix string, writer io.Writer) error {
688 prefix = v.key(prefix)
689 // Use a merge sort to find matching sets of X and recent/X.
690 dataL := s3awsLister{
694 PageSize: v.IndexPageSize,
695 Stats: &v.bucket.stats,
697 recentL := s3awsLister{
700 Prefix: "recent/" + prefix,
701 PageSize: v.IndexPageSize,
702 Stats: &v.bucket.stats,
704 for data, recent := dataL.First(), recentL.First(); data != nil && dataL.Error() == nil; data = dataL.Next() {
705 if *data.Key >= "g" {
706 // Conveniently, "recent/*" and "trash/*" are
707 // lexically greater than all hex-encoded data
708 // hashes, so stopping here avoids iterating
709 // over all of them needlessly with dataL.
712 loc, isblk := v.isKeepBlock(*data.Key)
717 // stamp is the list entry we should use to report the
718 // last-modified time for this data block: it will be
719 // the recent/X entry if one exists, otherwise the
720 // entry for the data block itself.
723 // Advance to the corresponding recent/X marker, if any
724 for recent != nil && recentL.Error() == nil {
725 if cmp := strings.Compare((*recent.Key)[7:], *data.Key); cmp < 0 {
726 recent = recentL.Next()
730 recent = recentL.Next()
733 // recent/X marker is missing: we'll
734 // use the timestamp on the data
739 if err := recentL.Error(); err != nil {
742 // We truncate sub-second precision here. Otherwise
743 // timestamps will never match the RFC1123-formatted
744 // Last-Modified values parsed by Mtime().
745 fmt.Fprintf(writer, "%s+%d %d\n", loc, *data.Size, stamp.LastModified.Unix()*1000000000)
750 // Mtime returns the stored timestamp for the given locator.
751 func (v *S3AWSVolume) Mtime(loc string) (time.Time, error) {
753 _, err := v.head(key)
755 return s3AWSZeroTime, v.translateError(err)
757 resp, err := v.head("recent/" + key)
758 err = v.translateError(err)
759 if os.IsNotExist(err) {
760 // The data object X exists, but recent/X is missing.
761 err = v.writeObject(context.Background(), "recent/"+key, nil)
763 v.logger.WithError(err).Errorf("error creating %q", "recent/"+key)
764 return s3AWSZeroTime, v.translateError(err)
766 v.logger.Infof("Mtime: created %q to migrate existing block to new storage scheme", "recent/"+key)
767 resp, err = v.head("recent/" + key)
769 v.logger.WithError(err).Errorf("HEAD failed after creating %q", "recent/"+key)
770 return s3AWSZeroTime, v.translateError(err)
772 } else if err != nil {
773 // HEAD recent/X failed for some other reason.
774 return s3AWSZeroTime, err
776 return *resp.LastModified, err
779 // Status returns a *VolumeStatus representing the current in-use
780 // storage capacity and a fake available capacity that doesn't make
781 // the volume seem full or nearly-full.
782 func (v *S3AWSVolume) Status() *VolumeStatus {
783 return &VolumeStatus{
785 BytesFree: BlockSize * 1000,
790 // InternalStats returns bucket I/O and API call counters.
791 func (v *S3AWSVolume) InternalStats() interface{} {
792 return &v.bucket.stats
795 // Touch sets the timestamp for the given locator to the current time.
796 func (v *S3AWSVolume) Touch(loc string) error {
797 if v.volume.ReadOnly {
798 return MethodDisabledError
801 _, err := v.head(key)
802 err = v.translateError(err)
803 if os.IsNotExist(err) && v.fixRace(key) {
804 // The data object got trashed in a race, but fixRace
806 } else if err != nil {
809 err = v.writeObject(context.Background(), "recent/"+key, nil)
810 return v.translateError(err)
813 // checkRaceWindow returns a non-nil error if trash/key is, or might
814 // be, in the race window (i.e., it's not safe to trash key).
815 func (v *S3AWSVolume) checkRaceWindow(key string) error {
816 resp, err := v.head("trash/" + key)
817 err = v.translateError(err)
818 if os.IsNotExist(err) {
819 // OK, trash/X doesn't exist so we're not in the race
822 } else if err != nil {
823 // Error looking up trash/X. We don't know whether
824 // we're in the race window
827 t := resp.LastModified
828 safeWindow := t.Add(v.cluster.Collections.BlobTrashLifetime.Duration()).Sub(time.Now().Add(time.Duration(v.RaceWindow)))
830 // We can't count on "touch trash/X" to prolong
831 // trash/X's lifetime. The new timestamp might not
832 // become visible until now+raceWindow, and EmptyTrash
833 // is allowed to delete trash/X before then.
834 return fmt.Errorf("%s: same block is already in trash, and safe window ended %s ago", key, -safeWindow)
836 // trash/X exists, but it won't be eligible for deletion until
837 // after now+raceWindow, so it's safe to overwrite it.
841 func (b *s3AWSbucket) Del(path string) error {
842 input := &s3.DeleteObjectInput{
843 Bucket: aws.String(b.bucket),
844 Key: aws.String(path),
846 req := b.svc.DeleteObjectRequest(input)
847 _, err := req.Send(context.Background())
848 b.stats.TickOps("delete")
849 b.stats.Tick(&b.stats.Ops, &b.stats.DelOps)
854 // Trash a Keep block.
855 func (v *S3AWSVolume) Trash(loc string) error {
856 if v.volume.ReadOnly {
857 return MethodDisabledError
859 if t, err := v.Mtime(loc); err != nil {
861 } else if time.Since(t) < v.cluster.Collections.BlobSigningTTL.Duration() {
865 if v.cluster.Collections.BlobTrashLifetime == 0 {
867 return ErrS3TrashDisabled
869 return v.translateError(v.bucket.Del(key))
871 err := v.checkRaceWindow(key)
875 err = v.safeCopy("trash/"+key, key)
879 return v.translateError(v.bucket.Del(key))
882 // Untrash moves block from trash back into store
883 func (v *S3AWSVolume) Untrash(loc string) error {
885 err := v.safeCopy(key, "trash/"+key)
889 err = v.writeObject(context.Background(), "recent/"+key, nil)
890 return v.translateError(err)
893 type s3awsbucketStats struct {
903 func (s *s3awsbucketStats) TickErr(err error) {
907 errType := fmt.Sprintf("%T", err)
908 if aerr, ok := err.(awserr.Error); ok {
909 if reqErr, ok := err.(awserr.RequestFailure); ok {
910 // A service error occurred
911 errType = errType + fmt.Sprintf(" %d %s", reqErr.StatusCode(), aerr.Code())
913 errType = errType + fmt.Sprintf(" 000 %s", aerr.Code())
916 s.statsTicker.TickErr(err, errType)