X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/f3b231c69407299133a6eb5ff6066ae6136608e9..feb463839301b0b596089e48a981660365d2c4a7:/services/keepstore/s3_volume.go diff --git a/services/keepstore/s3_volume.go b/services/keepstore/s3_volume.go index 4f5b3b165f..ca5b1a2eb9 100644 --- a/services/keepstore/s3_volume.go +++ b/services/keepstore/s3_volume.go @@ -9,17 +9,23 @@ import ( "fmt" "io" "io/ioutil" - "log" "net/http" "os" "regexp" "strings" "sync" + "sync/atomic" "time" "git.curoverse.com/arvados.git/sdk/go/arvados" "github.com/AdRoll/goamz/aws" "github.com/AdRoll/goamz/s3" + log "github.com/Sirupsen/logrus" +) + +const ( + s3DefaultReadTimeout = arvados.Duration(10 * time.Minute) + s3DefaultConnectTimeout = arvados.Duration(time.Minute) ) var ( @@ -143,7 +149,7 @@ type S3Volume struct { ReadOnly bool UnsafeDelete bool - bucket *s3.Bucket + bucket *s3bucket startOnce sync.Once } @@ -216,22 +222,45 @@ func (v *S3Volume) Start() error { // Zero timeouts mean "wait forever", which is a bad // default. Default to long timeouts instead. if v.ConnectTimeout == 0 { - v.ConnectTimeout = arvados.Duration(time.Minute) + v.ConnectTimeout = s3DefaultConnectTimeout } if v.ReadTimeout == 0 { - v.ReadTimeout = arvados.Duration(10 * time.Minute) + v.ReadTimeout = s3DefaultReadTimeout } client := s3.New(auth, region) client.ConnectTimeout = time.Duration(v.ConnectTimeout) client.ReadTimeout = time.Duration(v.ReadTimeout) - v.bucket = &s3.Bucket{ - S3: client, - Name: v.Bucket, + v.bucket = &s3bucket{ + Bucket: &s3.Bucket{ + S3: client, + Name: v.Bucket, + }, } return nil } +func (v *S3Volume) getReaderWithContext(ctx context.Context, loc string) (rdr io.ReadCloser, err error) { + ready := make(chan bool) + go func() { + rdr, err = v.getReader(loc) + close(ready) + }() + select { + case <-ready: + return + case <-ctx.Done(): + theConfig.debugLogf("s3: abandoning getReader(): %s", ctx.Err()) + go func() { + <-ready + if err == nil { + rdr.Close() + } + }() + return nil, ctx.Err() + } +} + // getReader wraps (Bucket)GetReader. // // In situations where (Bucket)GetReader would fail because the block @@ -243,6 +272,7 @@ func (v *S3Volume) getReader(loc string) (rdr io.ReadCloser, err error) { if err == nil || !os.IsNotExist(err) { return } + _, err = v.bucket.Head("recent/"+loc, nil) err = v.translateError(err) if err != nil { @@ -254,6 +284,7 @@ func (v *S3Volume) getReader(loc string) (rdr io.ReadCloser, err error) { err = os.ErrNotExist return } + rdr, err = v.bucket.GetReader(loc) if err != nil { log.Printf("warning: reading %s after successful fixRace: %s", loc, err) @@ -265,25 +296,13 @@ func (v *S3Volume) getReader(loc string) (rdr io.ReadCloser, err error) { // Get a block: copy the block data into buf, and return the number of // bytes copied. func (v *S3Volume) Get(ctx context.Context, loc string, buf []byte) (int, error) { - ready := make(chan bool) - var rdr io.ReadCloser - var err error - go func() { - rdr, err = v.getReader(loc) - close(ready) - }() - select { - case <-ctx.Done(): - theConfig.debugLogf("s3: abandoning getReader() because %s", ctx.Err()) - return 0, ctx.Err() - case <-ready: - if err != nil { - return 0, err - } + rdr, err := v.getReaderWithContext(ctx, loc) + if err != nil { + return 0, err } var n int - ready = make(chan bool) + ready := make(chan bool) go func() { defer close(ready) @@ -312,13 +331,13 @@ func (v *S3Volume) Get(ctx context.Context, loc string, buf []byte) (int, error) } // Compare the given data with the stored data. -func (v *S3Volume) Compare(loc string, expect []byte) error { - rdr, err := v.getReader(loc) +func (v *S3Volume) Compare(ctx context.Context, loc string, expect []byte) error { + rdr, err := v.getReaderWithContext(ctx, loc) if err != nil { return err } defer rdr.Close() - return v.translateError(compareReaderWithBuf(rdr, expect, loc[:32])) + return v.translateError(compareReaderWithBuf(ctx, rdr, expect, loc[:32])) } // Put writes a block. @@ -428,16 +447,19 @@ func (v *S3Volume) Mtime(loc string) (time.Time, error) { func (v *S3Volume) IndexTo(prefix string, writer io.Writer) error { // Use a merge sort to find matching sets of X and recent/X. dataL := s3Lister{ - Bucket: v.bucket, + Bucket: v.bucket.Bucket, Prefix: prefix, PageSize: v.IndexPageSize, } recentL := s3Lister{ - Bucket: v.bucket, + Bucket: v.bucket.Bucket, Prefix: "recent/" + prefix, PageSize: v.IndexPageSize, } + v.bucket.stats.tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps) + v.bucket.stats.tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps) for data, recent := dataL.First(), recentL.First(); data != nil; data = dataL.Next() { + v.bucket.stats.tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps) if data.Key >= "g" { // Conveniently, "recent/*" and "trash/*" are // lexically greater than all hex-encoded data @@ -459,10 +481,12 @@ func (v *S3Volume) IndexTo(prefix string, writer io.Writer) error { for recent != nil { if cmp := strings.Compare(recent.Key[7:], data.Key); cmp < 0 { recent = recentL.Next() + v.bucket.stats.tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps) continue } else if cmp == 0 { stamp = recent recent = recentL.Next() + v.bucket.stats.tick(&v.bucket.stats.Ops, &v.bucket.stats.ListOps) break } else { // recent/X marker is missing: we'll @@ -494,7 +518,7 @@ func (v *S3Volume) Trash(loc string) error { if !s3UnsafeDelete { return ErrS3TrashDisabled } - return v.bucket.Del(loc) + return v.translateError(v.bucket.Del(loc)) } err := v.checkRaceWindow(loc) if err != nil { @@ -597,9 +621,14 @@ func (v *S3Volume) Status() *VolumeStatus { } } +// InternalStats returns bucket I/O and API call counters. +func (v *S3Volume) InternalStats() interface{} { + return &v.bucket.stats +} + // String implements fmt.Stringer. func (v *S3Volume) String() string { - return fmt.Sprintf("s3-bucket:%+q", v.bucket.Name) + return fmt.Sprintf("s3-bucket:%+q", v.Bucket) } // Writable returns false if all future Put, Mtime, and Delete calls @@ -688,7 +717,7 @@ func (v *S3Volume) EmptyTrash() { // Use a merge sort to find matching sets of trash/X and recent/X. trashL := s3Lister{ - Bucket: v.bucket, + Bucket: v.bucket.Bucket, Prefix: "trash/", PageSize: v.IndexPageSize, } @@ -738,7 +767,9 @@ func (v *S3Volume) EmptyTrash() { v.fixRace(loc) v.Touch(loc) continue - } else if _, err := v.bucket.Head(loc, nil); os.IsNotExist(err) { + } + _, err := v.bucket.Head(loc, nil) + if os.IsNotExist(err) { log.Printf("notice: %s: EmptyTrash: detected recent race for %q, calling fixRace", v, loc) v.fixRace(loc) continue @@ -832,3 +863,91 @@ func (lister *s3Lister) pop() (k *s3.Key) { } return } + +// s3bucket wraps s3.bucket and counts I/O and API usage stats. +type s3bucket struct { + *s3.Bucket + stats s3bucketStats +} + +func (b *s3bucket) GetReader(path string) (io.ReadCloser, error) { + rdr, err := b.Bucket.GetReader(path) + b.stats.tick(&b.stats.Ops, &b.stats.GetOps) + b.stats.tickErr(err) + return NewCountingReader(rdr, b.stats.tickInBytes), err +} + +func (b *s3bucket) Head(path string, headers map[string][]string) (*http.Response, error) { + resp, err := b.Bucket.Head(path, headers) + b.stats.tick(&b.stats.Ops, &b.stats.HeadOps) + b.stats.tickErr(err) + return resp, err +} + +func (b *s3bucket) PutReader(path string, r io.Reader, length int64, contType string, perm s3.ACL, options s3.Options) error { + err := b.Bucket.PutReader(path, NewCountingReader(r, b.stats.tickOutBytes), length, contType, perm, options) + b.stats.tick(&b.stats.Ops, &b.stats.PutOps) + b.stats.tickErr(err) + return err +} + +func (b *s3bucket) Put(path string, data []byte, contType string, perm s3.ACL, options s3.Options) error { + err := b.Bucket.PutReader(path, NewCountingReader(bytes.NewBuffer(data), b.stats.tickOutBytes), int64(len(data)), contType, perm, options) + b.stats.tick(&b.stats.Ops, &b.stats.PutOps) + b.stats.tickErr(err) + return err +} + +func (b *s3bucket) Del(path string) error { + err := b.Bucket.Del(path) + b.stats.tick(&b.stats.Ops, &b.stats.DelOps) + b.stats.tickErr(err) + return err +} + +type s3bucketStats struct { + Errors uint64 + Ops uint64 + GetOps uint64 + PutOps uint64 + HeadOps uint64 + DelOps uint64 + ListOps uint64 + InBytes uint64 + OutBytes uint64 + + ErrorCodes map[string]uint64 `json:",omitempty"` + + lock sync.Mutex +} + +func (s *s3bucketStats) tickInBytes(n uint64) { + atomic.AddUint64(&s.InBytes, n) +} + +func (s *s3bucketStats) tickOutBytes(n uint64) { + atomic.AddUint64(&s.OutBytes, n) +} + +func (s *s3bucketStats) tick(counters ...*uint64) { + for _, counter := range counters { + atomic.AddUint64(counter, 1) + } +} + +func (s *s3bucketStats) tickErr(err error) { + if err == nil { + return + } + atomic.AddUint64(&s.Errors, 1) + errStr := fmt.Sprintf("%T", err) + if err, ok := err.(*s3.Error); ok { + errStr = errStr + fmt.Sprintf(" %d %s", err.StatusCode, err.Code) + } + s.lock.Lock() + if s.ErrorCodes == nil { + s.ErrorCodes = make(map[string]uint64) + } + s.ErrorCodes[errStr]++ + s.lock.Unlock() +}