10349: Add code snippet for combining collections.

[arvados.git] / services / keepstore / s3_volume.go

diff --git a/services/keepstore/s3_volume.go b/services/keepstore/s3_volume.go
index d34b8772c5eb90d8a5ac3ad89d3a2cee6a1f5d9c..3f08f6e1b8e362777286aad8581a1e7b5a9caa87 100644 (file)
--- a/services/keepstore/s3_volume.go
+++ b/services/keepstore/s3_volume.go
@@ -1,3 +1,7 @@
+// Copyright (C) The Arvados Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
 package main
 
 import (
@@ -129,7 +133,7 @@ func init() {
                &s3UnsafeDelete,
                "s3-unsafe-delete",
                false,
-               "EXPERIMENTAL. Enable deletion (garbage collection), even though there are known race conditions that can cause data loss.")
+               "EXPERIMENTAL. Enable deletion (garbage collection) even when trash lifetime is zero, even though there are known race conditions that can cause data loss.")
 }
 
 // S3Volume implements Volume using an S3 bucket.
@@ -239,6 +243,11 @@ func (v *S3Volume) Start() error {
        return nil
 }
 
+// DeviceID returns a globally unique ID for the storage bucket.
+func (v *S3Volume) DeviceID() string {
+       return "s3://" + v.Endpoint + "/" + v.Bucket
+}
+
 func (v *S3Volume) getReaderWithContext(ctx context.Context, loc string) (rdr io.ReadCloser, err error) {
        ready := make(chan bool)
        go func() {
@@ -331,6 +340,40 @@ 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(ctx context.Context, loc string, expect []byte) error {
+       errChan := make(chan error, 1)
+       go func() {
+               _, err := v.bucket.Head("recent/"+loc, nil)
+               errChan <- err
+       }()
+       var err error
+       select {
+       case <-ctx.Done():
+               return ctx.Err()
+       case err = <-errChan:
+       }
+       if err != nil {
+               // Checking for "loc" itself here would interfere with
+               // future GET requests.
+               //
+               // On AWS, if X doesn't exist, a HEAD or GET request
+               // for X causes X's non-existence to be cached. Thus,
+               // if we test for X, then create X and return a
+               // signature to our client, the client might still get
+               // 404 from all keepstores when trying to read it.
+               //
+               // To avoid this, we avoid doing HEAD X or GET X until
+               // we know X has been written.
+               //
+               // Note that X might exist even though recent/X
+               // doesn't: for example, the response to HEAD recent/X
+               // might itself come from a stale cache. In such
+               // cases, we will return a false negative and
+               // PutHandler might needlessly create another replica
+               // on a different volume. That's not ideal, but it's
+               // better than passing the eventually-consistent
+               // problem on to our clients.
+               return v.translateError(err)
+       }
        rdr, err := v.getReaderWithContext(ctx, loc)
        if err != nil {
                return err