capistrant opened a new pull request, #20002:
URL: https://github.com/apache/druid/pull/20002

   ### Description
   
   Seems to trace back to https://github.com/apache/druid/pull/16481 where we 
made this multipart upload async. Before this change any failed part would 
indeed result in an exceptional event.
   
   Explicitly throw when RetryableS3OutputStream fails to complete a multipart 
upload and aborts the write to s3. Right now, if an error occurs during 
completion of the multipart upload, we log the failure for that part and abort 
the large upload (correct to avoid orphaning files) before quietly returning to 
caller implicitly indicating success. Problem is, downstream code will try and 
fail to consume the files that were supposed to be uploaded and fail due to the 
object not existing. This can result in a hard to troubleshoot error since the 
real problem was the failed upload with an error log buried somewhere else.
   
   A common presentation I have seen for this is MSQ compaction using durable 
storage. During the shuffle a multipart upload to durable storage has a part 
that fails and aborts the rest of the upload + logs an error. But it doesn't 
throw and something downstream tries to access what would have been written by 
this worker which results in a no such key error from s3 and a task failure. 
Once you dig in you see that the multipart failure happened and doomed the task 
to fail down the road. This proposal intends to explicitly throw and in this 
case cause the compaction worker to fail fast instead of leave a landmine for 
whoever comes to read the data that wasn't written later.
   
   #### Release note
   
   Fixes an issue when MSQ durable storage writes to s3 could fail but not 
immediately report their failure, causing downstream code to fail in a way that 
made troubleshooting root cause more difficult.
   
   <hr>
   
   ##### Key changed/added classes in this PR
    * `RetryableS3OutputStream`
   
   <hr>
   
   <!-- Check the items by putting "x" in the brackets for the done things. Not 
all of these items apply to every PR. Remove the items which are not done or 
not relevant to the PR. None of the items from the checklist below are strictly 
necessary, but it would be very helpful if you at least self-review the PR. -->
   
   This PR has:
   
   - [ ] been self-reviewed.
   - [ ] added documentation for new or modified features or behaviors.
   - [ ] a release note entry in the PR description.
   - [ ] added Javadocs for most classes and all non-trivial methods. Linked 
related entities via Javadoc links.
   - [ ] added or updated version, license, or notice information in 
[licenses.yaml](https://github.com/apache/druid/blob/master/dev/license.md)
   - [ ] added comments explaining the "why" and the intent of the code 
wherever would not be obvious for an unfamiliar reader.
   - [ ] added unit tests or modified existing tests to cover new code paths, 
ensuring the threshold for [code 
coverage](https://github.com/apache/druid/blob/master/dev/code-review/code-coverage.md)
 is met.
   - [ ] added integration tests.
   - [ ] been tested in a test Druid cluster.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to