[fix](be) Backport agg shuffle and runtime filter fixes to 4.1#64156
Closed
BiteTheDDDDt wants to merge 3 commits into
Closed
[fix](be) Backport agg shuffle and runtime filter fixes to 4.1#64156BiteTheDDDDt wants to merge 3 commits into
BiteTheDDDDt wants to merge 3 commits into
Conversation
Problem Summary: PR apache#62438 added `enable_local_exchange_before_agg` to allow skipping local exchange before non-finalizing aggregation. That optimization used `!_needs_finalize` as the condition, but non-finalizing aggregation includes both first-phase update/serialize aggregation and merge/serialize aggregation. When `experimental_use_serial_exchange` is enabled and `enable_local_exchange_before_agg` is disabled, a serial exchange source can be followed by the default `PASSTHROUGH` local exchange before a non-finalizing merge aggregation. For DISTINCT aggregation, the merge aggregation is the stage that deduplicates distinct keys after hash exchange. `PASSTHROUGH` only restores local parallelism and does not preserve key distribution, so duplicate distinct keys can be processed by different local tasks and later partial sums can produce incorrect results. This PR keeps the knob behavior for aggregation stages that can safely skip the local hash exchange, but excludes merge aggregation with a serial child from the skip path. That case falls through to the existing `HASH_SHUFFLE` / `BUCKET_HASH_SHUFFLE` distribution requirement. The PR also computes the merge flag during `AggSinkOperatorX::init()` because local exchange planning runs before `prepare()`. Fix occasional incorrect DISTINCT aggregate results when serial exchange is enabled. (cherry picked from commit b95038e)
…3766) Related PR: apache#63529, apache#62438 Problem Summary: `enable_local_exchange_before_agg=false` allows first-phase aggregation to skip the local hash exchange before agg for performance. This is only correct when the input still preserves local key distribution. After apache#62438, nested loop join and other operators may introduce non-hash local exchanges such as `ADAPTIVE_PASSTHROUGH`. Those exchanges can split rows with the same group/distinct key across local pipeline tasks. If agg still skips the hash local exchange, partial aggregation states for the same key are built in different tasks and later `COUNT(DISTINCT ...)` can over-count. The reproduced query in `output/ddl.txt` returned wrong counts such as `18/20` instead of `10`. This PR preserves correctness while keeping the knob usable: - Aggregation operators now skip local exchange with `enable_local_exchange_before_agg=false` only when the child preserves local key distribution. - The shared child-distribution check is reused by `AggSinkOperatorX`, `StreamingAggOperatorX`, and `DistinctStreamingAggOperatorX`. - `Pipeline::need_to_local_exchange()` also handles the case where the current pipeline source is a non-hash `LocalExchangeSource` but the downstream target requires hash distribution, so inherited hash-ish pipeline state cannot incorrectly suppress the required local exchange. - Regression coverage is added for the nested-loop-join + distinct aggregation wrong-result case, and unit tests cover the agg distribution decisions. (cherry picked from commit d232caa)
…pache#64102) related with: apache#57425 Runtime filters from a parent inner join could be pushed through an outer join into the null-generating child even when the probe expression was not null-propagating for that child. The problem can be reproduced with this SQL shape: ```sql create table rf_outer_join_nullable_a (pk int) duplicate key(pk) distributed by hash(pk) buckets 1 properties("replication_num" = "1"); create table rf_outer_join_nullable_b (pk int) duplicate key(pk) distributed by hash(pk) buckets 1 properties("replication_num" = "1"); create table rf_outer_join_nullable_c (pk int) duplicate key(pk) distributed by hash(pk) buckets 1 properties("replication_num" = "1"); insert into rf_outer_join_nullable_a values (1); insert into rf_outer_join_nullable_b values (1); insert into rf_outer_join_nullable_c values (0); set disable_join_reorder = true; select coalesce(b.pk, 0) as k, count(*) as cnt from rf_outer_join_nullable_a a left join rf_outer_join_nullable_b b on a.pk = b.pk inner join rf_outer_join_nullable_c c on coalesce(b.pk, 0) = c.pk group by 1 order by 1; ``` The correct result is empty. `a.pk = 1` matches `b.pk = 1` in the left outer join, then the parent inner join evaluates `coalesce(1, 0) = 0`, which is false. The wrong plan generated a runtime filter from the parent inner join, effectively `c.pk -> coalesce(b.pk, 0)`, and pushed it through the lower `LEFT OUTER JOIN` into the right side scan of `b`. If `b.pk = 1` is pre-filtered before the left outer join, the join emits a NULL-extended row for `b`; then `coalesce(NULL, 0) = 0` becomes true and incorrectly returns `(0, 1)`. Therefore the runtime filter `c.pk -> coalesce(b.pk, 0)` must not be planned on the null-generating side of the lower outer join. This PR blocks runtime filter pushdown through an outer join's null-generating child unless the probe expression preserves NULL semantics for slots from that child. Normal pushdown through preserved sides and null-propagating expressions is kept unchanged. The bug became observable after apache#57425 changed the target lookup for expression runtime filters from `ctx.probeExpr` to `ctx.probeSlot`. Before that change, an expression such as `coalesce(b.pk, 0)` could not resolve the target relation in this path and the unsafe pushdown was not generated. None - Test <!-- At least one of them must be included. --> - [x] Regression test - [x] Unit Test - [ ] Manual test (add detailed scripts or steps below) - [ ] No need to test or manual test. Explain why: - [ ] This is a refactor/code format and no logic has been changed. - [ ] Previous test can cover this change. - [ ] No code files have been changed. - [ ] Other reason <!-- Add your reason? --> Added regression case with `disable_join_reorder`, `qt_shape`, and empty result verification: `regression-test/suites/correctness_p0/test_runtime_filter_outer_join_nullable_side.groovy` Unit test: `./run-fe-ut.sh --run org.apache.doris.nereids.postprocess.RuntimeFilterTest#testDoNotPushDownNonNullPropagatingRuntimeFilterThroughOuterJoin,org.apache.doris.nereids.postprocess.RuntimeFilterTest#testPushDownNullPropagatingRuntimeFilterThroughOuterJoin` The SQL regression case was not run locally against the available 9333 cluster because that cluster was the unpatched repro cluster. - Behavior changed: - [ ] No. - [x] Yes. Runtime filters are no longer pushed through an outer join into its null-generating child when the probe expression can convert NULL to a non-NULL value. - Does this need documentation? - [x] No. - [ ] Yes. <!-- Add document PR link here. eg: apache/doris-website#1214 --> - [ ] Confirm the release note - [ ] Confirm test cases - [ ] Confirm document - [ ] Add branch pick label <!-- Add branch pick label that this PR should merge into --> (cherry picked from commit 691189a)
Contributor
|
Thank you for your contribution to Apache Doris. Please clearly describe your PR:
|
Contributor
Author
|
run buildall |
Contributor
There was a problem hiding this comment.
Pull request overview
Backports correctness fixes to the 4.1 branch for (1) aggregation/local-exchange distribution handling (to avoid wrong DISTINCT/group-by results when local key distribution is broken) and (2) runtime filter pushdown across outer joins (to avoid filtering the null-generating side when probe expressions are not NULL-propagating).
Changes:
- BE: Refine aggregation operators’ required distribution logic to preserve hash shuffle when upstream local exchanges/serial sources can break local key distribution, plus pipeline handling for non-hash local exchange sources.
- FE: Block runtime filter pushdown into the null-generating side of outer joins unless the probe expression is NULL-propagating for that side.
- Tests: Add/extend BE unit tests, FE unit tests, and regression suites + expected outputs to cover the above scenarios.
Reviewed changes
Copilot reviewed 21 out of 21 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| regression-test/suites/query_p0/join/test_agg_after_nested_loop_join_local_exchange.groovy | New regression to catch wrong DISTINCT agg results after NLJ-introduced non-hash local exchange. |
| regression-test/data/query_p0/join/test_agg_after_nested_loop_join_local_exchange.out | Expected output for the new NLJ + agg regression. |
| regression-test/suites/nereids_syntax_p0/agg_4_phase.groovy | Extends existing suite with serial-exchange distinct aggregation coverage and config toggles. |
| regression-test/data/nereids_syntax_p0/agg_4_phase.out | Expected output update for the added serial-exchange distinct agg query. |
| regression-test/suites/correctness_p0/test_runtime_filter_outer_join_nullable_side.groovy | New regression to ensure unsafe RF pushdown into outer-join nullable side is prevented. |
| regression-test/data/correctness_p0/test_runtime_filter_outer_join_nullable_side.out | Expected shape/result output for the new runtime filter regression. |
| fe/fe-core/src/test/java/org/apache/doris/nereids/postprocess/RuntimeFilterTest.java | Adds unit tests asserting RF behavior for NULL-propagating vs non-propagating probe expressions. |
| fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/RuntimeFilterPushDownVisitor.java | Implements the outer-join nullable-side RF pushdown guard based on NULL propagation. |
| be/src/runtime/runtime_state.h | Adds an accessor for enable_local_exchange_before_agg. |
| be/src/exec/pipeline/pipeline.cpp | Ensures downstream hash requirements aren’t suppressed after a non-hash LocalExchangeSource. |
| be/src/exec/operator/operator.h | Adds helper APIs to detect when a child breaks local key distribution. |
| be/src/exec/operator/operator.cpp | Implements the child local-key-distribution break detection logic. |
| be/src/exec/operator/streaming_aggregation_operator.h | Uses new child-distribution check to decide when skipping local exchange is safe. |
| be/src/exec/operator/distinct_streaming_aggregation_operator.h | Same distribution-safety logic for distinct streaming aggregation. |
| be/src/exec/operator/aggregation_sink_operator.h | Same distribution-safety logic for aggregation sink operator required distribution. |
| be/src/exec/operator/aggregation_sink_operator.cpp | Computes _is_merge earlier (during init) to support distribution planning before prepare. |
| be/src/exec/exchange/local_exchange_source_operator.h | Exposes local exchange type via required distribution + accessor to aid distribution checks. |
| be/test/exec/operator/streaming_agg_operator_test.cpp | Adds unit test covering hash-shuffle requirement after non-hash local exchange. |
| be/test/exec/operator/distinct_streaming_aggregation_operator_test.cpp | Adds unit test covering hash-shuffle requirement after non-hash local exchange. |
| be/test/exec/operator/agg_operator_test.cpp | Adds unit tests for agg sink required distribution + pipeline local exchange decision. |
| be/test/exec/operator/agg_operator_group_by_limit_opt_test.cpp | Updates mock agg sink evaluator init to compute _is_merge (test correctness). |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+589
to
+592
| bool enable_local_exchange_before_agg() const { | ||
| return _query_options.__isset.enable_local_exchange_before_agg && | ||
| _query_options.enable_local_exchange_before_agg; | ||
| } |
Comment on lines
752
to
756
| tnode.agg_node.__isset.agg_sort_infos ? tnode.agg_node.agg_sort_infos[i] : dummy, | ||
| tnode.agg_node.grouping_exprs.empty(), false, &evaluator)); | ||
| _aggregate_evaluators.push_back(evaluator); | ||
| _is_merge |= evaluator->is_merge(); | ||
| } |
Contributor
FE Regression Coverage ReportIncrement line coverage |
Contributor
Author
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
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.
What problem does this PR solve?
Issue Number: N/A
Related PR: #63529, #63766, #64102
Problem Summary:
Backport the following merged fixes to branch-4.1:
The BE aggregation backports are applicable to branch-4.1 because the branch still has the affected aggregation/local-exchange distribution paths and was missing these fixes. The runtime filter fix is also absent from branch-4.1 and was adapted to the branch-4.1 RuntimeFilterPushDownVisitor structure.
Release note
None
Check List (For Author)