1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
23 "git.curoverse.com/arvados.git/sdk/go/arvados"
24 "github.com/AdRoll/goamz/aws"
25 "github.com/AdRoll/goamz/s3"
26 log "github.com/Sirupsen/logrus"
30 s3DefaultReadTimeout = arvados.Duration(10 * time.Minute)
31 s3DefaultConnectTimeout = arvados.Duration(time.Minute)
35 // ErrS3TrashDisabled is returned by Trash if that operation
36 // is impossible with the current config.
37 ErrS3TrashDisabled = fmt.Errorf("trash function is disabled because -trash-lifetime=0 and -s3-unsafe-delete=false")
39 s3AccessKeyFile string
40 s3SecretKeyFile string
45 s3RaceWindow time.Duration
51 maxClockSkew = 600 * time.Second
52 nearlyRFC1123 = "Mon, 2 Jan 2006 15:04:05 GMT"
55 type s3VolumeAdder struct {
59 // String implements flag.Value
60 func (s *s3VolumeAdder) String() string {
64 func (s *s3VolumeAdder) Set(bucketName string) error {
66 return fmt.Errorf("no container name given")
68 if s3AccessKeyFile == "" || s3SecretKeyFile == "" {
69 return fmt.Errorf("-s3-access-key-file and -s3-secret-key-file arguments must given before -s3-bucket-volume")
71 if deprecated.flagSerializeIO {
72 log.Print("Notice: -serialize is not supported by s3-bucket volumes.")
74 s.Config.Volumes = append(s.Config.Volumes, &S3Volume{
76 AccessKeyFile: s3AccessKeyFile,
77 SecretKeyFile: s3SecretKeyFile,
80 RaceWindow: arvados.Duration(s3RaceWindow),
81 S3Replication: s3Replication,
82 UnsafeDelete: s3UnsafeDelete,
83 ReadOnly: deprecated.flagReadonly,
89 func s3regions() (okList []string) {
90 for r := range aws.Regions {
91 okList = append(okList, r)
97 VolumeTypes = append(VolumeTypes, func() VolumeWithExamples { return &S3Volume{} })
99 flag.Var(&s3VolumeAdder{theConfig},
101 "Use the given bucket as a storage volume. Can be given multiple times.")
106 fmt.Sprintf("AWS region used for subsequent -s3-bucket-volume arguments. Allowed values are %+q.", s3regions()))
111 "Endpoint URL used for subsequent -s3-bucket-volume arguments. If blank, use the AWS endpoint corresponding to the -s3-region argument. For Google Storage, use \"https://storage.googleapis.com\".")
114 "s3-access-key-file",
116 "`File` containing the access key used for subsequent -s3-bucket-volume arguments.")
119 "s3-secret-key-file",
121 "`File` containing the secret key used for subsequent -s3-bucket-volume arguments.")
126 "Maximum eventual consistency latency for subsequent -s3-bucket-volume arguments.")
131 "Replication level reported to clients for subsequent -s3-bucket-volume arguments.")
136 "EXPERIMENTAL. Enable deletion (garbage collection) even when trash lifetime is zero, even though there are known race conditions that can cause data loss.")
139 // S3Volume implements Volume using an S3 bucket.
140 type S3Volume struct {
146 LocationConstraint bool
149 ConnectTimeout arvados.Duration
150 ReadTimeout arvados.Duration
151 RaceWindow arvados.Duration
160 // Examples implements VolumeWithExamples.
161 func (*S3Volume) Examples() []Volume {
164 AccessKeyFile: "/etc/aws_s3_access_key.txt",
165 SecretKeyFile: "/etc/aws_s3_secret_key.txt",
168 Bucket: "example-bucket-name",
171 RaceWindow: arvados.Duration(24 * time.Hour),
172 ConnectTimeout: arvados.Duration(time.Minute),
173 ReadTimeout: arvados.Duration(5 * time.Minute),
176 AccessKeyFile: "/etc/gce_s3_access_key.txt",
177 SecretKeyFile: "/etc/gce_s3_secret_key.txt",
178 Endpoint: "https://storage.googleapis.com",
180 Bucket: "example-bucket-name",
183 RaceWindow: arvados.Duration(24 * time.Hour),
184 ConnectTimeout: arvados.Duration(time.Minute),
185 ReadTimeout: arvados.Duration(5 * time.Minute),
190 // Type implements Volume.
191 func (*S3Volume) Type() string {
195 // Start populates private fields and verifies the configuration is
197 func (v *S3Volume) Start() error {
198 region, ok := aws.Regions[v.Region]
199 if v.Endpoint == "" {
201 return fmt.Errorf("unrecognized region %+q; try specifying -s3-endpoint instead", v.Region)
204 return fmt.Errorf("refusing to use AWS region name %+q with endpoint %+q; "+
205 "specify empty endpoint (\"-s3-endpoint=\") or use a different region name", v.Region, v.Endpoint)
209 S3Endpoint: v.Endpoint,
210 S3LocationConstraint: v.LocationConstraint,
216 auth.AccessKey, err = readKeyFromFile(v.AccessKeyFile)
220 auth.SecretKey, err = readKeyFromFile(v.SecretKeyFile)
225 // Zero timeouts mean "wait forever", which is a bad
226 // default. Default to long timeouts instead.
227 if v.ConnectTimeout == 0 {
228 v.ConnectTimeout = s3DefaultConnectTimeout
230 if v.ReadTimeout == 0 {
231 v.ReadTimeout = s3DefaultReadTimeout
234 client := s3.New(auth, region)
235 client.ConnectTimeout = time.Duration(v.ConnectTimeout)
236 client.ReadTimeout = time.Duration(v.ReadTimeout)
237 v.bucket = &s3bucket{
246 // DeviceID returns a globally unique ID for the storage bucket.
247 func (v *S3Volume) DeviceID() string {
248 return "s3://" + v.Endpoint + "/" + v.Bucket
251 func (v *S3Volume) getReaderWithContext(ctx context.Context, loc string) (rdr io.ReadCloser, err error) {
252 ready := make(chan bool)
254 rdr, err = v.getReader(loc)
261 theConfig.debugLogf("s3: abandoning getReader(): %s", ctx.Err())
268 return nil, ctx.Err()
272 // getReader wraps (Bucket)GetReader.
274 // In situations where (Bucket)GetReader would fail because the block
275 // disappeared in a Trash race, getReader calls fixRace to recover the
276 // data, and tries again.
277 func (v *S3Volume) getReader(loc string) (rdr io.ReadCloser, err error) {
278 rdr, err = v.bucket.GetReader(loc)
279 err = v.translateError(err)
280 if err == nil || !os.IsNotExist(err) {
284 _, err = v.bucket.Head("recent/"+loc, nil)
285 err = v.translateError(err)
287 // If we can't read recent/X, there's no point in
288 // trying fixRace. Give up.
296 rdr, err = v.bucket.GetReader(loc)
298 log.Printf("warning: reading %s after successful fixRace: %s", loc, err)
299 err = v.translateError(err)
304 // Get a block: copy the block data into buf, and return the number of
306 func (v *S3Volume) Get(ctx context.Context, loc string, buf []byte) (int, error) {
307 rdr, err := v.getReaderWithContext(ctx, loc)
313 ready := make(chan bool)
318 n, err = io.ReadFull(rdr, buf)
321 case nil, io.EOF, io.ErrUnexpectedEOF:
324 err = v.translateError(err)
329 theConfig.debugLogf("s3: interrupting ReadFull() with Close() because %s", ctx.Err())
331 // Must wait for ReadFull to return, to ensure it
332 // doesn't write to buf after we return.
333 theConfig.debugLogf("s3: waiting for ReadFull() to fail")
341 // Compare the given data with the stored data.
342 func (v *S3Volume) Compare(ctx context.Context, loc string, expect []byte) error {
343 errChan := make(chan error, 1)
345 _, err := v.bucket.Head("recent/"+loc, nil)
352 case err = <-errChan:
355 // Checking for "loc" itself here would interfere with
356 // future GET requests.
358 // On AWS, if X doesn't exist, a HEAD or GET request
359 // for X causes X's non-existence to be cached. Thus,
360 // if we test for X, then create X and return a
361 // signature to our client, the client might still get
362 // 404 from all keepstores when trying to read it.
364 // To avoid this, we avoid doing HEAD X or GET X until
365 // we know X has been written.
367 // Note that X might exist even though recent/X
368 // doesn't: for example, the response to HEAD recent/X
369 // might itself come from a stale cache. In such
370 // cases, we will return a false negative and
371 // PutHandler might needlessly create another replica
372 // on a different volume. That's not ideal, but it's
373 // better than passing the eventually-consistent
374 // problem on to our clients.
375 return v.translateError(err)
377 rdr, err := v.getReaderWithContext(ctx, loc)
382 return v.translateError(compareReaderWithBuf(ctx, rdr, expect, loc[:32]))
385 // Put writes a block.
386 func (v *S3Volume) Put(ctx context.Context, loc string, block []byte) error {
388 return MethodDisabledError
393 md5, err := hex.DecodeString(loc)
397 opts.ContentMD5 = base64.StdEncoding.EncodeToString(md5)
400 // Send the block data through a pipe, so that (if we need to)
401 // we can close the pipe early and abandon our PutReader()
402 // goroutine, without worrying about PutReader() accessing our
403 // block buffer after we release it.
404 bufr, bufw := io.Pipe()
406 io.Copy(bufw, bytes.NewReader(block))
411 ready := make(chan bool)
414 if ctx.Err() != nil {
415 theConfig.debugLogf("%s: abandoned PutReader goroutine finished with err: %s", v, err)
419 err = v.bucket.PutReader(loc, bufr, int64(size), "application/octet-stream", s3ACL, opts)
423 err = v.bucket.PutReader("recent/"+loc, nil, 0, "application/octet-stream", s3ACL, s3.Options{})
427 theConfig.debugLogf("%s: taking PutReader's input away: %s", v, ctx.Err())
428 // Our pipe might be stuck in Write(), waiting for
429 // io.Copy() to read. If so, un-stick it. This means
430 // PutReader will get corrupt data, but that's OK: the
431 // size and MD5 won't match, so the write will fail.
432 go io.Copy(ioutil.Discard, bufr)
433 // CloseWithError() will return once pending I/O is done.
434 bufw.CloseWithError(ctx.Err())
435 theConfig.debugLogf("%s: abandoning PutReader goroutine", v)
438 return v.translateError(err)
442 // Touch sets the timestamp for the given locator to the current time.
443 func (v *S3Volume) Touch(loc string) error {
445 return MethodDisabledError
447 _, err := v.bucket.Head(loc, nil)
448 err = v.translateError(err)
449 if os.IsNotExist(err) && v.fixRace(loc) {
450 // The data object got trashed in a race, but fixRace
452 } else if err != nil {
455 err = v.bucket.PutReader("recent/"+loc, nil, 0, "application/octet-stream", s3ACL, s3.Options{})
456 return v.translateError(err)
459 // Mtime returns the stored timestamp for the given locator.
460 func (v *S3Volume) Mtime(loc string) (time.Time, error) {
461 _, err := v.bucket.Head(loc, nil)
463 return zeroTime, v.translateError(err)
465 resp, err := v.bucket.Head("recent/"+loc, nil)
466 err = v.translateError(err)
467 if os.IsNotExist(err) {
468 // The data object X exists, but recent/X is missing.
469 err = v.bucket.PutReader("recent/"+loc, nil, 0, "application/octet-stream", s3ACL, s3.Options{})
471 log.Printf("error: creating %q: %s", "recent/"+loc, err)
472 return zeroTime, v.translateError(err)
474 log.Printf("info: created %q to migrate existing block to new storage scheme", "recent/"+loc)
475 resp, err = v.bucket.Head("recent/"+loc, nil)
477 log.Printf("error: created %q but HEAD failed: %s", "recent/"+loc, err)
478 return zeroTime, v.translateError(err)
480 } else if err != nil {
481 // HEAD recent/X failed for some other reason.
484 return v.lastModified(resp)
487 // IndexTo writes a complete list of locators with the given prefix
488 // for which Get() can retrieve data.
489 func (v *S3Volume) IndexTo(prefix string, writer io.Writer) error {
490 // Use a merge sort to find matching sets of X and recent/X.
492 Bucket: v.bucket.Bucket,
494 PageSize: v.IndexPageSize,
497 Bucket: v.bucket.Bucket,
498 Prefix: "recent/" + prefix,
499 PageSize: v.IndexPageSize,
501 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps)
502 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps)
503 for data, recent := dataL.First(), recentL.First(); data != nil; data = dataL.Next() {
504 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps)
506 // Conveniently, "recent/*" and "trash/*" are
507 // lexically greater than all hex-encoded data
508 // hashes, so stopping here avoids iterating
509 // over all of them needlessly with dataL.
512 if !v.isKeepBlock(data.Key) {
516 // stamp is the list entry we should use to report the
517 // last-modified time for this data block: it will be
518 // the recent/X entry if one exists, otherwise the
519 // entry for the data block itself.
522 // Advance to the corresponding recent/X marker, if any
524 if cmp := strings.Compare(recent.Key[7:], data.Key); cmp < 0 {
525 recent = recentL.Next()
526 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps)
530 recent = recentL.Next()
531 v.bucket.stats.Tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps)
534 // recent/X marker is missing: we'll
535 // use the timestamp on the data
540 t, err := time.Parse(time.RFC3339, stamp.LastModified)
544 fmt.Fprintf(writer, "%s+%d %d\n", data.Key, data.Size, t.UnixNano())
549 // Trash a Keep block.
550 func (v *S3Volume) Trash(loc string) error {
552 return MethodDisabledError
554 if t, err := v.Mtime(loc); err != nil {
556 } else if time.Since(t) < theConfig.BlobSignatureTTL.Duration() {
559 if theConfig.TrashLifetime == 0 {
561 return ErrS3TrashDisabled
563 return v.translateError(v.bucket.Del(loc))
565 err := v.checkRaceWindow(loc)
569 err = v.safeCopy("trash/"+loc, loc)
573 return v.translateError(v.bucket.Del(loc))
576 // checkRaceWindow returns a non-nil error if trash/loc is, or might
577 // be, in the race window (i.e., it's not safe to trash loc).
578 func (v *S3Volume) checkRaceWindow(loc string) error {
579 resp, err := v.bucket.Head("trash/"+loc, nil)
580 err = v.translateError(err)
581 if os.IsNotExist(err) {
582 // OK, trash/X doesn't exist so we're not in the race
585 } else if err != nil {
586 // Error looking up trash/X. We don't know whether
587 // we're in the race window
590 t, err := v.lastModified(resp)
592 // Can't parse timestamp
595 safeWindow := t.Add(theConfig.TrashLifetime.Duration()).Sub(time.Now().Add(time.Duration(v.RaceWindow)))
597 // We can't count on "touch trash/X" to prolong
598 // trash/X's lifetime. The new timestamp might not
599 // become visible until now+raceWindow, and EmptyTrash
600 // is allowed to delete trash/X before then.
601 return fmt.Errorf("same block is already in trash, and safe window ended %s ago", -safeWindow)
603 // trash/X exists, but it won't be eligible for deletion until
604 // after now+raceWindow, so it's safe to overwrite it.
608 // safeCopy calls PutCopy, and checks the response to make sure the
609 // copy succeeded and updated the timestamp on the destination object
610 // (PutCopy returns 200 OK if the request was received, even if the
612 func (v *S3Volume) safeCopy(dst, src string) error {
613 resp, err := v.bucket.PutCopy(dst, s3ACL, s3.CopyOptions{
614 ContentType: "application/octet-stream",
615 MetadataDirective: "REPLACE",
616 }, v.bucket.Name+"/"+src)
617 err = v.translateError(err)
621 if t, err := time.Parse(time.RFC3339Nano, resp.LastModified); err != nil {
622 return fmt.Errorf("PutCopy succeeded but did not return a timestamp: %q: %s", resp.LastModified, err)
623 } else if time.Now().Sub(t) > maxClockSkew {
624 return fmt.Errorf("PutCopy succeeded but returned an old timestamp: %q: %s", resp.LastModified, t)
629 // Get the LastModified header from resp, and parse it as RFC1123 or
630 // -- if it isn't valid RFC1123 -- as Amazon's variant of RFC1123.
631 func (v *S3Volume) lastModified(resp *http.Response) (t time.Time, err error) {
632 s := resp.Header.Get("Last-Modified")
633 t, err = time.Parse(time.RFC1123, s)
634 if err != nil && s != "" {
635 // AWS example is "Sun, 1 Jan 2006 12:00:00 GMT",
636 // which isn't quite "Sun, 01 Jan 2006 12:00:00 GMT"
637 // as required by HTTP spec. If it's not a valid HTTP
638 // header value, it's probably AWS (or s3test) giving
639 // us a nearly-RFC1123 timestamp.
640 t, err = time.Parse(nearlyRFC1123, s)
645 // Untrash moves block from trash back into store
646 func (v *S3Volume) Untrash(loc string) error {
647 err := v.safeCopy(loc, "trash/"+loc)
651 err = v.bucket.PutReader("recent/"+loc, nil, 0, "application/octet-stream", s3ACL, s3.Options{})
652 return v.translateError(err)
655 // Status returns a *VolumeStatus representing the current in-use
656 // storage capacity and a fake available capacity that doesn't make
657 // the volume seem full or nearly-full.
658 func (v *S3Volume) Status() *VolumeStatus {
659 return &VolumeStatus{
661 BytesFree: BlockSize * 1000,
666 // InternalStats returns bucket I/O and API call counters.
667 func (v *S3Volume) InternalStats() interface{} {
668 return &v.bucket.stats
671 // String implements fmt.Stringer.
672 func (v *S3Volume) String() string {
673 return fmt.Sprintf("s3-bucket:%+q", v.Bucket)
676 // Writable returns false if all future Put, Mtime, and Delete calls
677 // are expected to fail.
678 func (v *S3Volume) Writable() bool {
682 // Replication returns the storage redundancy of the underlying
683 // device. Configured via command line flag.
684 func (v *S3Volume) Replication() int {
685 return v.S3Replication
688 var s3KeepBlockRegexp = regexp.MustCompile(`^[0-9a-f]{32}$`)
690 func (v *S3Volume) isKeepBlock(s string) bool {
691 return s3KeepBlockRegexp.MatchString(s)
694 // fixRace(X) is called when "recent/X" exists but "X" doesn't
695 // exist. If the timestamps on "recent/"+loc and "trash/"+loc indicate
696 // there was a race between Put and Trash, fixRace recovers from the
697 // race by Untrashing the block.
698 func (v *S3Volume) fixRace(loc string) bool {
699 trash, err := v.bucket.Head("trash/"+loc, nil)
701 if !os.IsNotExist(v.translateError(err)) {
702 log.Printf("error: fixRace: HEAD %q: %s", "trash/"+loc, err)
706 trashTime, err := v.lastModified(trash)
708 log.Printf("error: fixRace: parse %q: %s", trash.Header.Get("Last-Modified"), err)
712 recent, err := v.bucket.Head("recent/"+loc, nil)
714 log.Printf("error: fixRace: HEAD %q: %s", "recent/"+loc, err)
717 recentTime, err := v.lastModified(recent)
719 log.Printf("error: fixRace: parse %q: %s", recent.Header.Get("Last-Modified"), err)
723 ageWhenTrashed := trashTime.Sub(recentTime)
724 if ageWhenTrashed >= theConfig.BlobSignatureTTL.Duration() {
725 // No evidence of a race: block hasn't been written
726 // since it became eligible for Trash. No fix needed.
730 log.Printf("notice: fixRace: %q: trashed at %s but touched at %s (age when trashed = %s < %s)", loc, trashTime, recentTime, ageWhenTrashed, theConfig.BlobSignatureTTL)
731 log.Printf("notice: fixRace: copying %q to %q to recover from race between Put/Touch and Trash", "recent/"+loc, loc)
732 err = v.safeCopy(loc, "trash/"+loc)
734 log.Printf("error: fixRace: %s", err)
740 func (v *S3Volume) translateError(err error) error {
741 switch err := err.(type) {
743 if (err.StatusCode == http.StatusNotFound && err.Code == "NoSuchKey") ||
744 strings.Contains(err.Error(), "Not Found") {
745 return os.ErrNotExist
747 // Other 404 errors like NoSuchVersion and
748 // NoSuchBucket are different problems which should
749 // get called out downstream, so we don't convert them
750 // to os.ErrNotExist.
755 // EmptyTrash looks for trashed blocks that exceeded TrashLifetime
756 // and deletes them from the volume.
757 func (v *S3Volume) EmptyTrash() {
758 var bytesInTrash, blocksInTrash, bytesDeleted, blocksDeleted int64
760 // Use a merge sort to find matching sets of trash/X and recent/X.
762 Bucket: v.bucket.Bucket,
764 PageSize: v.IndexPageSize,
766 // Define "ready to delete" as "...when EmptyTrash started".
768 for trash := trashL.First(); trash != nil; trash = trashL.Next() {
770 if !v.isKeepBlock(loc) {
773 bytesInTrash += trash.Size
776 trashT, err := time.Parse(time.RFC3339, trash.LastModified)
778 log.Printf("warning: %s: EmptyTrash: %q: parse %q: %s", v, trash.Key, trash.LastModified, err)
781 recent, err := v.bucket.Head("recent/"+loc, nil)
782 if err != nil && os.IsNotExist(v.translateError(err)) {
783 log.Printf("warning: %s: EmptyTrash: found trash marker %q but no %q (%s); calling Untrash", v, trash.Key, "recent/"+loc, err)
786 log.Printf("error: %s: EmptyTrash: Untrash(%q): %s", v, loc, err)
789 } else if err != nil {
790 log.Printf("warning: %s: EmptyTrash: HEAD %q: %s", v, "recent/"+loc, err)
793 recentT, err := v.lastModified(recent)
795 log.Printf("warning: %s: EmptyTrash: %q: parse %q: %s", v, "recent/"+loc, recent.Header.Get("Last-Modified"), err)
798 if trashT.Sub(recentT) < theConfig.BlobSignatureTTL.Duration() {
799 if age := startT.Sub(recentT); age >= theConfig.BlobSignatureTTL.Duration()-time.Duration(v.RaceWindow) {
800 // recent/loc is too old to protect
801 // loc from being Trashed again during
802 // the raceWindow that starts if we
803 // delete trash/X now.
805 // Note this means (TrashCheckInterval
806 // < BlobSignatureTTL - raceWindow) is
807 // necessary to avoid starvation.
808 log.Printf("notice: %s: EmptyTrash: detected old race for %q, calling fixRace + Touch", v, loc)
813 _, err := v.bucket.Head(loc, nil)
814 if os.IsNotExist(err) {
815 log.Printf("notice: %s: EmptyTrash: detected recent race for %q, calling fixRace", v, loc)
818 } else if err != nil {
819 log.Printf("warning: %s: EmptyTrash: HEAD %q: %s", v, loc, err)
823 if startT.Sub(trashT) < theConfig.TrashLifetime.Duration() {
826 err = v.bucket.Del(trash.Key)
828 log.Printf("warning: %s: EmptyTrash: deleting %q: %s", v, trash.Key, err)
831 bytesDeleted += trash.Size
834 _, err = v.bucket.Head(loc, nil)
835 if os.IsNotExist(err) {
836 err = v.bucket.Del("recent/" + loc)
838 log.Printf("warning: %s: EmptyTrash: deleting %q: %s", v, "recent/"+loc, err)
840 } else if err != nil {
841 log.Printf("warning: %s: EmptyTrash: HEAD %q: %s", v, "recent/"+loc, err)
844 if err := trashL.Error(); err != nil {
845 log.Printf("error: %s: EmptyTrash: lister: %s", v, err)
847 log.Printf("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)
850 type s3Lister struct {
859 // First fetches the first page and returns the first item. It returns
860 // nil if the response is the empty set or an error occurs.
861 func (lister *s3Lister) First() *s3.Key {
866 // Next returns the next item, fetching the next page if necessary. It
867 // returns nil if the last available item has already been fetched, or
869 func (lister *s3Lister) Next() *s3.Key {
870 if len(lister.buf) == 0 && lister.nextMarker != "" {
876 // Return the most recent error encountered by First or Next.
877 func (lister *s3Lister) Error() error {
881 func (lister *s3Lister) getPage() {
882 resp, err := lister.Bucket.List(lister.Prefix, "", lister.nextMarker, lister.PageSize)
883 lister.nextMarker = ""
888 if resp.IsTruncated {
889 lister.nextMarker = resp.NextMarker
891 lister.buf = make([]s3.Key, 0, len(resp.Contents))
892 for _, key := range resp.Contents {
893 if !strings.HasPrefix(key.Key, lister.Prefix) {
894 log.Printf("warning: s3Lister: S3 Bucket.List(prefix=%q) returned key %q", lister.Prefix, key.Key)
897 lister.buf = append(lister.buf, key)
901 func (lister *s3Lister) pop() (k *s3.Key) {
902 if len(lister.buf) > 0 {
904 lister.buf = lister.buf[1:]
909 // s3bucket wraps s3.bucket and counts I/O and API usage stats.
910 type s3bucket struct {
915 func (b *s3bucket) GetReader(path string) (io.ReadCloser, error) {
916 rdr, err := b.Bucket.GetReader(path)
917 b.stats.Tick(&b.stats.Ops, &b.stats.GetOps)
919 return NewCountingReader(rdr, b.stats.TickInBytes), err
922 func (b *s3bucket) Head(path string, headers map[string][]string) (*http.Response, error) {
923 resp, err := b.Bucket.Head(path, headers)
924 b.stats.Tick(&b.stats.Ops, &b.stats.HeadOps)
929 func (b *s3bucket) PutReader(path string, r io.Reader, length int64, contType string, perm s3.ACL, options s3.Options) error {
931 // goamz will only send Content-Length: 0 when reader
932 // is nil due to net.http.Request.ContentLength
933 // behavior. Otherwise, Content-Length header is
934 // omitted which will cause some S3 services
935 // (including AWS and Ceph RadosGW) to fail to create
939 r = NewCountingReader(r, b.stats.TickOutBytes)
941 err := b.Bucket.PutReader(path, r, length, contType, perm, options)
942 b.stats.Tick(&b.stats.Ops, &b.stats.PutOps)
947 func (b *s3bucket) Del(path string) error {
948 err := b.Bucket.Del(path)
949 b.stats.Tick(&b.stats.Ops, &b.stats.DelOps)
954 type s3bucketStats struct {
964 func (s *s3bucketStats) TickErr(err error) {
968 errType := fmt.Sprintf("%T", err)
969 if err, ok := err.(*s3.Error); ok {
970 errType = errType + fmt.Sprintf(" %d %s", err.StatusCode, err.Code)
972 s.statsTicker.TickErr(err, errType)