Skip to content

reproducer for delay in cancellation caused by CoalescePartitionsExec#1

Open
Samyak2 wants to merge 6 commits into
mainfrom
coalesce-delay-drop
Open

reproducer for delay in cancellation caused by CoalescePartitionsExec#1
Samyak2 wants to merge 6 commits into
mainfrom
coalesce-delay-drop

Conversation

@Samyak2

@Samyak2 Samyak2 commented May 4, 2026

Copy link
Copy Markdown
Owner

Run:

cargo test -p datafusion-physical-plan cancellation_delay_coalesce_repartition -- --ignored --nocapture

Output:

running 1 test
lower_final_partitioned_hash_repartition_drop_elapsed_ms=80
upper_final_partitioned_hash_repartition_drop_elapsed_ms=68
repartition_task_group=0 input_partition=0 kind=pull_from_input drop_elapsed_ms=68
repartition_task_group=0 input_partition=0 kind=wait_for_task drop_elapsed_ms=68
repartition_task_group=1 input_partition=0 kind=pull_from_input drop_elapsed_ms=80
repartition_task_group=1 input_partition=0 kind=wait_for_task drop_elapsed_ms=80
repartition_task_group=1 input_partition=1 kind=pull_from_input drop_elapsed_ms=85
repartition_task_group=1 input_partition=1 kind=wait_for_task drop_elapsed_ms=80
output_partitions=32 input_rows_per_partition=1024000 all_repartition_operator_drop_elapsed_ms=80
all_repartition_task_drop_elapsed_ms=85
all_observed_drop_elapsed_ms=85
input_plan_strong_count_after_cancel=0
test coalesce_partitions::tests::cancellation_delay_coalesce_repartition ... ok

zhuqi-lucas pushed a commit to zhuqi-lucas/arrow-datafusion that referenced this pull request May 7, 2026
## Which issue does this PR close?

<!--
We generally require a GitHub issue to be filed for all bug fixes and
enhancements and this helps us generate change logs for our releases.
You can link an issue to this PR using the GitHub syntax. For example
`Closes apache#123` indicates that this PR will close issue apache#123.
-->

- Closes apache#22016

## Rationale for this change

<!--
Why are you proposing this change? If this is already explained clearly
in the issue then this section is not needed.
Explaining clearly why changes are proposed helps reviewers understand
your changes and offer better suggestions for fixes.
-->

- `CoalescePartitionsExec` uses `RecordBatchReceiverStreamBuilder` which
was holding to an Arc ref of the input plan, until the stream was done.
- More details in the issue.

## What changes are included in this PR?

<!--
There is no need to duplicate the description in the issue here but it
is sometimes worth providing a summary of the individual changes in this
PR.
-->

- Now we drop the input plan early.
- The only other usage of `input` was in debug prints, to display the
input plan. We now take the display string early, but only when debug
logging is enabled.

## Are these changes tested?

<!--
We typically require tests for all PRs in order to:
1. Prevent the code from being accidentally broken by subsequent changes
2. Serve as another way to document the expected behavior of the code

If tests are not included in your PR, please explain why (for example,
are they covered by existing tests)?
-->

- Added a test that ensures this. Verified that the test fails without
the fix.

The reproducer (Samyak2#1) before the
fix:
```
repartition_task_group=0 input_partition=0 kind=pull_from_input drop_elapsed_ms=68
repartition_task_group=1 input_partition=0 kind=pull_from_input drop_elapsed_ms=80
repartition_task_group=1 input_partition=1 kind=pull_from_input drop_elapsed_ms=85
output_partitions=32 input_rows_per_partition=1024000 all_repartition_operator_drop_elapsed_ms=80
all_repartition_task_drop_elapsed_ms=85
all_observed_drop_elapsed_ms=85
```
~85ms to cancel repartition tasks.

After fix:
```
repartition_task_group=0 input_partition=0 kind=pull_from_input drop_elapsed_ms=16
repartition_task_group=1 input_partition=0 kind=pull_from_input drop_elapsed_ms=0
repartition_task_group=1 input_partition=1 kind=pull_from_input drop_elapsed_ms=0
output_partitions=32 input_rows_per_partition=1024000 all_repartition_operator_drop_elapsed_ms=0
all_repartition_task_drop_elapsed_ms=16
all_observed_drop_elapsed_ms=16
```
~16ms to cancel repartition tasks.

## Are there any user-facing changes?

<!--
If there are user-facing changes then we may require documentation to be
updated before approving the PR.
-->

<!--
If there are any breaking changes to public APIs, please add the `api
change` label.
-->

CPU/memory gets released earlier on cancellation

---------

Co-authored-by: Kumar Ujjawal <ujjawalpathak6@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant