[branch-52] Fix Arrow Spill Underrun (#20159)#20684
Merged
alamb merged 1 commit intoapache:branch-52from Mar 4, 2026
Merged
Conversation
## Which issue does this PR close? - Closes apache#19425 ## Rationale for this change This adjusts the way that the spill channel works. Currently we have a spill writer & reader pairing which uses a mutex to coordindate when a file is ready to be read. What happens is, that because we were using a `spawn_buffered` call, the read task would race ahead trying to read a file which is yet to be written out completely. Alongside this, we need to flush each write to the file, as there is a chance that another thread may see stale data. ## What changes are included in this PR? Adds a flush on write, and converts the read task to not buffer reads. ## Are these changes tested? I haven't written a test, but I have been running the example in the attached issue. While it now fails with allocation errors, the original error goes away. ## Are there any user-facing changes? Nope
12 tasks
Contributor
Author
hareshkh
changed the title
Contributor
|
Could you link in the comment description (so it ends up in the commit) to the original PR, and how you created this diff (e.g. "cherry picked abc123")? |
alamb
approved these changes
Mar 4, 2026
Contributor
alamb
left a comment
alamb
left a comment
There was a problem hiding this comment.
Looks good to me -- thank you @hareshkh and @adriangb
I linked the PR that is being backported and I verified the diff is the same, and removed the duplicated template in the PR description
I also ran the problem query and verified that that before this PR , it fails like this:
andrewlamb@Andrews-MacBook-Pro-3:~/Software/datafusion$ ~/Software/datafusion-cli/datafusion-cli-52.2.0 -m 1G -c "SELECT \"UserID\", extract(minute FROM to_timestamp_seconds(\"EventTime\")) AS m, \"SearchPhrase\", COUNT(*) FROM '/Users/andrewlamb/Software/datafusion/benchmarks/data/hits_partitioned' GROUP BY \"UserID\", m, \"SearchPhrase\" ORDER BY COUNT(*) DESC LIMIT 10;"
DataFusion CLI v52.2.0
Error: Arrow error: Io error: failed to fill whole bufferAnd after this PR it succeeds
andrewlamb@Andrews-MacBook-Pro-3:~/Software/datafusion$ ./target/profiling/datafusion-cli -m 1G -c "SELECT \"UserID\", extract(minute FROM to_timestamp_seconds(\"EventTime\")) AS m, \"SearchPhrase\", COUNT(*) FROM '/Users/andrewlamb/Software/datafusion/benchmarks/data/hits_partitioned' GROUP BY \"UserID\", m, \"SearchPhrase\" ORDER BY COUNT(*) DESC LIMIT 10;"
DataFusion CLI v52.2.0
+---------------------+----+--------------+----------+
| UserID | m | SearchPhrase | count(*) |
+---------------------+----+--------------+----------+
| 1313338681122956954 | 31 | | 589 |
| 1313338681122956954 | 28 | | 578 |
| 1313338681122956954 | 29 | | 572 |
| 1313338681122956954 | 33 | | 567 |
| 1313338681122956954 | 27 | | 557 |
| 1313338681122956954 | 32 | | 554 |
| 1313338681122956954 | 30 | | 552 |
| 1313338681122956954 | 34 | | 546 |
| 1313338681122956954 | 26 | | 540 |
| 1313338681122956954 | 10 | | 539 |
+---------------------+----+--------------+----------+
10 row(s) fetched.
Elapsed 184.374 seconds.
Contributor
|
Thanks again @hareshkh |
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Which issue does this PR close?
Rationale for this change
This adjusts the way that the spill channel works. Currently we have a spill writer & reader pairing which uses a mutex to coordindate when a file is ready to be read.
What happens is, that because we were using a
spawn_bufferedcall, the read task would race ahead trying to read a file which is yet to be written out completely.Alongside this, we need to flush each write to the file, as there is a chance that another thread may see stale data.
What changes are included in this PR?
Adds a flush on write, and converts the read task to not buffer reads.
Are these changes tested?
I haven't written a test, but I have been running the example in the attached issue. While it now fails with allocation errors, the original error goes away.
Are there any user-facing changes?
Nope