Uh oh!
There was an error while loading. Please reload this page.
feat: Create dynamic filters in SortMergeJoin - #21267
Closed
stuhood wants to merge 5 commits into
Closed
Conversation
stuhood
commented
Mar 31, 2026
ContributorAuthor
Supersedes #20455. |
stuhoodforce-pushed
the
stuhood.smj-dynamic-filter-creation
branch
2 times, most recently
from
March 31, 2026 05:29
ff61597 to
d7dcdd5Comparestuhood
marked this pull request as draft
March 31, 2026 05:49
stuhoodforce-pushed
the
stuhood.smj-dynamic-filter-creation
branch
from
March 31, 2026 15:06
d7dcdd5 to
155f789CompareDandandan
commented
Apr 1, 2026
Contributor
run benchmark tpch tpcds |
adriangbot
commented
Apr 1, 2026
🤖 Benchmark running (GKE) | trigger CPU Details (lscpu)Comparing stuhood.smj-dynamic-filter-creation (155f789) to ccaf802 (merge-base) diff using: tpch File an issue against this benchmark runner |
adriangbot
commented
Apr 1, 2026
🤖 Benchmark running (GKE) | trigger CPU Details (lscpu)Comparing stuhood.smj-dynamic-filter-creation (155f789) to ccaf802 (merge-base) diff using: tpcds File an issue against this benchmark runner |
adriangbot
commented
Apr 1, 2026
🤖 Benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)DetailsResource Usagetpch — base (merge-base)
tpch — branch
File an issue against this benchmark runner |
adriangbot
commented
Apr 1, 2026
🤖 Benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)DetailsResource Usagetpcds — base (merge-base)
tpcds — branch
File an issue against this benchmark runner |
Dandandan
commented
Apr 1, 2026
Contributor
🚀🚀🚀 |
adriangb
reviewed
Apr 2, 2026
adriangb
left a comment
Contributor
There was a problem hiding this comment.
Thank you for working on this. Can we split out the fixes for the other nodes into it's own PR that we can fast track (and maybe discuss any API improvements in)?
Then we can look at the SPM in more detail.
Comment on lines
+658
to
+665
| if !matches!(phase, FilterPushdownPhase::Pre) { | ||
| let mut result = FilterPushdownPropagation::if_all(child_pushdown_result); | ||
| if let Some(updated_child) = result.updated_node { | ||
| let mut new_self = self.clone(); | ||
| new_self.input = updated_child; | ||
| result.updated_node = Some(Arc::new(new_self) as _); | ||
| } | ||
| return Ok(result); |
Contributor
There was a problem hiding this comment.
Good catch! This is definitely a footgun. Any thoughts on how we can make the API less error prone? It would also be nice to put this into its own PR so we can fast track it.
| .iter() | ||
| .map(|f| PushedDownPredicate::unsupported(Arc::clone(f))) | ||
| .collect(), | ||
| parent_filters, |
| ) -> Result<FilterPushdownPropagation<Arc<dyn ExecutionPlan>>> { | ||
| let mut result = FilterPushdownPropagation::if_all(child_pushdown_result); | ||
| if let Some(updated_child) = result.updated_node { | ||
| result.updated_node = Some(Arc::new(GlobalLimitExec::new( |
Contributor
There was a problem hiding this comment.
Suggested change
| result.updated_node = Some(Arc::new(GlobalLimitExec::new( | |
| result.updated_node = Some(Arc::new(Self::new( |
?
| let mut result = FilterPushdownPropagation::if_all(child_pushdown_result); | ||
| if let Some(updated_child) = result.updated_node { | ||
| result.updated_node = | ||
| Some(Arc::new(LocalLimitExec::new(updated_child, self.fetch)) as _); |
Comment on lines
+436
to
+442
| let mut result = FilterPushdownPropagation::if_all(child_pushdown_result); | ||
| if let Some(updated_child) = result.updated_node { | ||
| let mut new_self = self.clone(); | ||
| new_self.input = updated_child; | ||
| result.updated_node = Some(Arc::new(new_self) as _); | ||
| } | ||
| Ok(result) |
Comment on lines
-1204
to
1211
| Ok(FilterPushdownPropagation::if_all(child_pushdown_result)) | ||
| let mut result = FilterPushdownPropagation::if_all(child_pushdown_result); | ||
| if let Some(updated_child) = result.updated_node { | ||
| let mut new_self = self.clone(); | ||
| new_self.input = updated_child; | ||
| result.updated_node = Some(Arc::new(new_self) as _); | ||
| } | ||
| Ok(result) | ||
| } |
Comment on lines
+1435
to
+1448
| fn handle_child_pushdown_result( | ||
| &self, | ||
| _phase: FilterPushdownPhase, | ||
| child_pushdown_result: ChildPushdownResult, | ||
| _config: &datafusion_common::config::ConfigOptions, | ||
| ) -> Result<FilterPushdownPropagation<Arc<dyn ExecutionPlan>>> { | ||
| let mut result = FilterPushdownPropagation::if_all(child_pushdown_result); | ||
| if let Some(updated_child) = result.updated_node { | ||
| let mut new_self = self.cloned(); | ||
| new_self.input = updated_child; | ||
| result.updated_node = Some(Arc::new(new_self) as _); | ||
| } | ||
| Ok(result) | ||
| } |
Contributor
There was a problem hiding this comment.
Really need to eliminate this foot gun 🦶🏻
Comment on lines
-280
to
+286
| Ok(FilterPushdownPropagation::if_all(child_pushdown_result)) | ||
| let mut result = FilterPushdownPropagation::if_all(child_pushdown_result); | ||
| if let Some(updated_child) = result.updated_node { | ||
| let mut new_self = self.clone(); | ||
| new_self.input = updated_child; | ||
| result.updated_node = Some(Arc::new(new_self) as _); | ||
| } | ||
| Ok(result) |
stuhood
commented
Apr 4, 2026
ContributorAuthor
I will do this! But I will be on vacation for the next two weeks: will leave it in draft until I've applied your feedback. |
| new_self.input = updated_child; | ||
| result.updated_node = Some(Arc::new(new_self) as _); | ||
| } | ||
| Ok(result) |
Contributor
There was a problem hiding this comment.
@stuhood looking at this again I'm not sure this is needed / there is a real bug. Do you have a regression test that fails?
FilterPushdownPropagation::if_all always sets updated_node: None:
datafusion/datafusion/physical-plan/src/filter_pushdown.rs
Lines 238 to 248 in ad6a507
And then in the optimizer rule:
datafusion/datafusion/physical-optimizer/src/filter_pushdown.rs
Lines 567 to 568 in ad6a507
datafusion/datafusion/physical-optimizer/src/filter_pushdown.rs
Lines 591 to 595 in ad6a507
Contributor
There was a problem hiding this comment.
So basically if any children have been updated we update the parent. Let me know if I'm missing something.
stuhood
commented
Jun 25, 2026
ContributorAuthor
We are going to be focused on partitioning (rather than sorting) for the near term, so I'm going to close this one for now. Sorry for the noise! |
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 freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
6 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?
SortMergeJoinExec#20443.Rationale for this change
This change fixes#20443, and adds support for generating dynamic filters for the left and right sides of a
SortMergeJoin, in order to allow for range-based pruning of both sides of the join. Some consumers of the dynamic filter may even be able to seek/skip ahead on their inputs.What changes are included in this PR?
To allow
SortMergeJointo actually propagate dynamic filters down to scans and get things working end to end, it was necessary to fix thehandle_child_pushdown_resultimplementations of a variety of nodes which were failing to clone and update themselves.Are these changes tested?
Yes.
Are there any user-facing changes?
Explain will now show dynamic filters for
SortMergeJoin, and for the scans that they consume.