2010YOUY01 opened a new pull request, #26410:
URL: https://github.com/apache/datafusion/pull/26410
## 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 #123`
indicates that this PR will close issue #123.
-->
- Closes #.
## 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.
Please explain the problem you are trying to solve in terms of the
user-visible
behavior, rather than the implementation.
For example, "The code in `foo.rs` doesn't handle nulls" is a symptom of the
implementation. "COUNT(DISTINCT) returns wrong results when the column
contains
nulls" is the user-visible problem.
-->
See discussion for details:
[https://github.com/apache/datafusion/pull/25363](https://github.com/apache/datafusion/pull/25363)
TLDR: The optimization itself is great, but might not be a good perf v.s.
complexity tradeoff, so I overall think it's better to revert.
This PR introduces an integer-key prefilter for Hash Join, which overlaps
with the existing `ArrayMap` optimization.
For example, consider a Hash Join on a `UInt64` key, where the build-side
key range is `[1, 100]`:
- **ArrayMap:** Replaces the hash table with a flat array for direct lookup.
- **This PR:** Keeps the hash table but adds a bitmap prefilter to skip
unnecessary hash table lookups.
The tradeoff is:
- **ArrayMap:** Faster, but uses more memory.
- **Prefilter + HashMap:** Slower, but more memory-efficient, particularly
for sparse key ranges.
The concern is **whether the benefits justify the additional maintenance
overhead**:
1. It introduces 4 configuration options, while `ArrayMap` already has a
similar threshold configuration.
2. Having two closely related optimizations makes execution behavior harder
to predict and maintain.
This seem very hard to tune them right. I've considered simplifying the
thresholds or integrating the prefilter with existing mechanisms, but neither
seems straightforward.
In the [[original
discussion](https://github.com/apache/datafusion/pull/25363#issuecomment-6183169314)](https://github.com/apache/datafusion/pull/25363#issuecomment-6183169314),
@lyne7-sc provided benchmarks confirming the tradeoff, and also expressed
uncertainty about whether the extra maintenance is worthwhile.
**My overall recommendation is to revert this optimization for
simplification**, given our limited engineering capacity and current
maintenance burden. (I think the optimization is a good idea, we can still
think about introducing it in the future, but try to be more general-purpose
and easier to configure)
I'd love to hear others' thoughts.
## 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.
-->
commit 1 is pure change revert
commit 2 is some fixing due to main updates
## What is the testing strategy for this PR?
<!--
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
Briefly describe how this PR is tested, and point to the specific tests you
added. For example: 'This new feature is covered by the `sqllogictest` cases
added in `foo.slt`'.
If this PR does not add tests, explain why. For example, if the change is
already covered by existing tests, please mention it.
You should also check the `codecov` bot reply on this PR to confirm the
changed code is exercised.
-->
## 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.
-->
--
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]
