6321: Add OSError to slurm shutdown actor retry conditions, to catch and retry

[arvados.git] / services / keepstore / azure_blob_volume.go

diff --git a/services/keepstore/azure_blob_volume.go b/services/keepstore/azure_blob_volume.go
index 836b8f18c80dd51314b1e387da6e7f61309029ef..657c3151caade8aeca6cbedcd96ffef70d3d7ac4 100644 (file)
--- a/services/keepstore/azure_blob_volume.go
+++ b/services/keepstore/azure_blob_volume.go
@@ -122,7 +122,7 @@ func (v *AzureBlobVolume) Get(loc string) ([]byte, error) {
        var deadline time.Time
        haveDeadline := false
        buf, err := v.get(loc)
-       for err == nil && len(buf) == 0 && loc[:32] != "d41d8cd98f00b204e9800998ecf8427e" {
+       for err == nil && len(buf) == 0 && loc != "d41d8cd98f00b204e9800998ecf8427e" {
                // Seeing a brand new empty block probably means we're
                // in a race with CreateBlob, which under the hood
                // (apparently) does "CreateEmpty" and "CommitData"
@@ -134,38 +134,34 @@ func (v *AzureBlobVolume) Get(loc string) ([]byte, error) {
                                break
                        }
                        deadline = t.Add(azureWriteRaceInterval)
+                       if time.Now().After(deadline) {
+                               break
+                       }
+                       log.Printf("Race? Block %s is 0 bytes, %s old. Polling until %s", loc, time.Since(t), deadline)
                        haveDeadline = true
-               }
-               if time.Now().After(deadline) {
+               } else if time.Now().After(deadline) {
                        break
                }
                bufs.Put(buf)
                time.Sleep(azureWriteRacePollTime)
                buf, err = v.get(loc)
        }
+       if haveDeadline {
+               log.Printf("Race ended with len(buf)==%d", len(buf))
+       }
        return buf, err
 }
 
 func (v *AzureBlobVolume) get(loc string) ([]byte, error) {
        rdr, err := v.bsClient.GetBlob(v.containerName, loc)
        if err != nil {
-               if strings.Contains(err.Error(), "404 Not Found") {
-                       // "storage: service returned without a response body (404 Not Found)"
-                       return nil, os.ErrNotExist
-               }
-               return nil, err
-       }
-       switch err := err.(type) {
-       case nil:
-       default:
-               log.Printf("ERROR IN Get(): %T %#v", err, err)
-               return nil, err
+               return nil, v.translateError(err)
        }
        defer rdr.Close()
        buf := bufs.Get(BlockSize)
        n, err := io.ReadFull(rdr, buf)
        switch err {
-       case io.EOF, io.ErrUnexpectedEOF:
+       case nil, io.EOF, io.ErrUnexpectedEOF:
                return buf[:n], nil
        default:
                bufs.Put(buf)
@@ -176,7 +172,7 @@ func (v *AzureBlobVolume) get(loc string) ([]byte, error) {
 func (v *AzureBlobVolume) Compare(loc string, expect []byte) error {
        rdr, err := v.bsClient.GetBlob(v.containerName, loc)
        if err != nil {
-               return err
+               return v.translateError(err)
        }
        defer rdr.Close()
        return compareReaderWithBuf(rdr, expect, loc[:32])
@@ -220,6 +216,14 @@ func (v *AzureBlobVolume) IndexTo(prefix string, writer io.Writer) error {
                        if err != nil {
                                return err
                        }
+                       if b.Properties.ContentLength == 0 && t.Add(azureWriteRaceInterval).After(time.Now()) {
+                               // A new zero-length blob is probably
+                               // just a new non-empty blob that
+                               // hasn't committed its data yet (see
+                               // Get()), and in any case has no
+                               // value.
+                               continue
+                       }
                        fmt.Fprintf(writer, "%s+%d %d\n", b.Name, b.Properties.ContentLength, t.Unix())
                }
                if resp.NextMarker == "" {
@@ -271,3 +275,17 @@ func (v *AzureBlobVolume) Writable() bool {
 func (v *AzureBlobVolume) Replication() int {
        return v.replication
 }
+
+// If possible, translate an Azure SDK error to a recognizable error
+// like os.ErrNotExist.
+func (v *AzureBlobVolume) translateError(err error) error {
+       switch {
+       case err == nil:
+               return err
+       case strings.Contains(err.Error(), "404 Not Found"):
+               // "storage: service returned without a response body (404 Not Found)"
+               return os.ErrNotExist
+       default:
+               return err
+       }
+}