[refactor](fe) Refactor SetPreAggStatus traversal to use per-scan context map instead of shared stack#65095
Draft
starocean999 wants to merge 1 commit into
Draft
[refactor](fe) Refactor SetPreAggStatus traversal to use per-scan context map instead of shared stack#65095starocean999 wants to merge 1 commit into
starocean999 wants to merge 1 commit into
Conversation
Contributor
|
Thank you for your contribution to Apache Doris. Please clearly describe your PR:
|
Contributor
Author
|
/review |
starocean999
changed the title
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Automated review completed. I did not find a substantiated blocking issue in this PR.
Critical checkpoint conclusions:
- Goal: the PR replaces the shared stack in
SetPreAggStatuswith per-RelationIdcontexts. The changed traversal matches that goal and avoids sharing scan context between sibling branches. - Scope: the change is focused to the pre-aggregation status rewriter plus two regression
explainexpectations. - Optimizer correctness: the snapshot/diff pattern was checked against
DefaultPlanRewriter.visitChildren,LogicalFilter,LogicalJoin,LogicalProject,LogicalAggregate, andLogicalRepeat. The handled nodes preserve context until the nearest aggregate consumes it; generic nodes remove only contexts introduced below that node, preserving pre-existing sibling/ancestor contexts. - Parallel paths: rule placement remains in the existing late rewrite stage, and the second registration path still uses the same custom rule. I did not find a missed sibling join/aggregate path in the changed code.
- Configuration, concurrency, lifecycle, persistence, compatibility, FE/BE protocol, and storage format: not applicable; this PR does not add or alter those surfaces.
- Tests: the regression suite updates the two right-join sibling aggregate cases from
preagg_t3OFF to ON, which targets the reported shared-context behavior. No.outartifact is expected for theseexplain { contains ... }assertions. - Validation: exact PR-range
git diff --checkpassed. I did not run FE build or regression tests because this runner lacks.worktree_initialized,thirdparty/installed, and executablethirdparty/installed/bin/protoc.
User focus: no additional user-provided review focus was present.
Subagent conclusions: optimizer-rewrite reported NO_NEW_VALUABLE_FINDINGS. tests-session-config proposed TSC-1 for trailing whitespace, but I dismissed it because it came from git show --check HEAD in a grafted checkout; the exact base-to-head git diff --check passes and the PR patch changes only two assertion strings. Final convergence round 1 ended with both subagents reporting NO_NEW_VALUABLE_FINDINGS against the same ledger and empty proposed inline comment set.
Contributor
Author
|
run buildall |
Contributor
FE UT Coverage ReportIncrement line coverage |
Contributor
TPC-H: Total hot run time: 30121 ms |
Contributor
TPC-DS: Total hot run time: 174162 ms |
Contributor
ClickBench: Total hot run time: 25.47 s |
Contributor
FE Regression Coverage ReportIncrement line coverage |
…text map instead of shared stack
Contributor
Author
|
/review |
Contributor
Author
|
run buildall |
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
2 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.
Related PR: #48502
Problem Summary:
The SetPreAggStatus rule uses a bottom-up traversal to determine whether
pre-aggregation can be enabled for OlapScan nodes. The original implementation
used a
Stack<PreAggInfoContext>as the traversal context, which had afundamental flaw: multiple scan nodes under sibling subtrees (e.g., both sides
of a Join) incorrectly shared the same stack-level context.
This caused two problems:
of the plan tree.
visitLogicalAggregateusedretainAll()to filter outscan nodes that were not actually children of the current aggregate — an
inelegant patch that partially masked the underlying issue.
Root cause: The stack-based context could not guarantee a one-to-one
correspondence between a scan node and its nearest aggregate ancestor.
Fix: Replace the
Stack<PreAggInfoContext>with aMap<RelationId, PreAggInfoContext>, where each scan node gets its own dedicated context entry.Each visitor method uses a snapshot-diff pattern:
At
LogicalAggregate, child scan contexts are consumed (removed from the map)and their results are stored — no more
retainAll()filtering needed.None
Check List (For Author)
Test
Behavior changed:
Does this need documentation?
Check List (For Reviewer who merge this PR)