github-actions[bot] commented on code in PR #64779:
URL: https://github.com/apache/doris/pull/64779#discussion_r3467215593
##########
be/src/io/fs/s3_file_writer.cpp:
##########
@@ -275,9 +311,9 @@ Status S3FileWriter::_close_impl() {
}
if (_pending_buf != nullptr) { // there is remaining data in buffer need
to be uploaded
- _countdown_event.add_count();
- RETURN_IF_ERROR(FileBuffer::submit(std::move(_pending_buf)));
+ auto st = _submit_upload_buffer(_pending_buf);
Review Comment:
Blocking: this error return lets async close complete before previously
submitted upload buffers have drained. A multipart writer can submit one or
more full buffers from `appendv()`, and those `UploadFileBuffer` callbacks
capture this writer. If the async close worker then fails to submit the final
`_pending_buf`, `_close_impl()` clears `_pending_buf` and returns here without
reaching `_complete()` and its `_wait_until_finish()` call. The close lambda
immediately fulfills `_async_close_pack->promise`, and the destructor path for
an outstanding async close waits only on that future. Destroying the writer
after this future is ready can therefore leave earlier upload tasks running
with callbacks that still dereference the freed `S3FileWriter`. Please make
this error path drain `_countdown_event` before completing the async close, or
otherwise preserve the writer until all earlier upload tasks have finished.
--
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]
