Cancel S3 requests when dropped #794
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Today we don't cancel S3 requests when dropped. For our prefetcher that means we keep streaming (up to) 2GB of data that will never be used. This change cancels in-flight requests when dropped, so that the CRT will stop streaming them. Some bytes might still be in flight or delivered, which is fine. Canceling requests is a no-op if they've already completed. The tricky case for this change is PutObject. Our current implementation of `PutObjectRequest::write` blocks until the bytes it provides are consumed by the client. But sometimes the client might stop reading from the stream because the request has failed. That case happens to work today because we don't retain a reference to the meta request ourselves, and so the failed request's destructors run immediately after the failure, which unblocks the writer and returns it an error. But now we do hold onto a reference, and the destructors can't run until the last reference is released, so the writer is never unblocked. To fix this, we make the `write` and `complete` methods of the `PutObjectRequest` poll _both_ the write stream and the request itself in parallel. If the request completes, this gives us a chance to bail out of the write rather than blocking forever. Signed-off-by: James Bornholt <[email protected]>
Signed-off-by: James Bornholt <[email protected]>
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
had a problem deploying
to
PR integration tests
GitHub Actions
Failure
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
passaro
reviewed
Mar 4, 2024
| @@ -167,9 +169,51 @@ async fn test_put_object_abort() { | |||
|
|
|||
| drop(request); // Drop without calling complete(). | |||
|
|
|||
| // Allow for the AbortMultipartUpload to complete. | |||
| // Try to wait a while for the async abort to complete. For the larger object, this might be | |||
There was a problem hiding this comment.
This seems to still not be enough: https://github.com/awslabs/mountpoint-s3/actions/runs/8134885229/job/22228500040?pr=794#step:8:1228
There was a problem hiding this comment.
I'm still messing with it to confirm, but it looks like a CRT-side issue — the AbortMultipartUpload can race with in-flight UploadParts, which might succeed after the Abort succeeds and so re-create the upload. I'm not sure if that's worth fixing here (since we're not actually changing how PUT works), so I might just disable the test for large objects.
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
had a problem deploying
to
PR integration tests
GitHub Actions
Failure
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
The CRT abort is best-effort -- part uploads can succeed after the Abort succeeds, which effectively recreates the MPU. This is mentioned in the AbortMultipartUpload documentation. Signed-off-by: James Bornholt <[email protected]>
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
Signed-off-by: James Bornholt <[email protected]>
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
jamesbornholt
temporarily deployed
to
PR integration tests
GitHub Actions
Inactive
monthonk
approved these changes
Mar 6, 2024
passaro
added a commit
to passaro/mountpoint-s3
that referenced
this pull request
Nov 22, 2024
The issue was addressed in Cancel S3 requests when dropped [awslabs#794](awslabs#794). Signed-off-by: Alessandro Passaro <[email protected]>
passaro
added a commit
to passaro/mountpoint-s3
that referenced
this pull request
Nov 25, 2024
The issue was addressed in Cancel S3 requests when dropped [awslabs#794](awslabs#794). Signed-off-by: Alessandro Passaro <[email protected]>
Merged
github-merge-queue bot
pushed a commit
that referenced
this pull request
Nov 25, 2024
The issue has already been addressed in Cancel S3 requests when dropped [#794](#794). --- By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license and I agree to the terms of the [Developer Certificate of Origin (DCO)](https://developercertificate.org/). Signed-off-by: Alessandro Passaro <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description of change
Today we don't cancel S3 requests when dropped. For our prefetcher that
means we keep streaming (up to) 2GB of data that will never be used.
This change cancels in-flight requests when dropped, so that the CRT
will stop streaming them. Some bytes might still be in flight or
delivered, which is fine. Canceling requests is a no-op if they've
already completed.
The tricky case for this change is PutObject. Our current implementation
of
PutObjectRequest::writeblocks until the bytes it provides areconsumed by the client. But sometimes the client might stop reading from
the stream because the request has failed. That case happens to work
today because we don't retain a reference to the meta request ourselves,
and so the failed request's destructors run immediately after the
failure, which unblocks the writer and returns it an error. But now we do
hold onto a reference, and the destructors can't run until the last
reference is released, so the writer is never unblocked. To fix this, we
make the
writeandcompletemethods of thePutObjectRequestpollboth the write stream and the request itself in parallel. If the request
completes, this gives us a chance to bail out of the write rather than
blocking forever.
Relevant issues: fixes #510.
Does this change impact existing behavior?
No.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license and I agree to the terms of the Developer Certificate of Origin (DCO).