[fix](cost model) scale group-by-less global agg cost by BE number when partitioned - #66354
[fix](cost model) scale group-by-less global agg cost by BE number when partitioned#66354foxtail463 wants to merge 3 commits into
Conversation
|
Thank you for your contribution to Apache Doris. Please clearly describe your PR:
|
|
run buildall |
dd90308 to
19f7f33
Compare
TPC-H: Total hot run time: 28674 ms |
TPC-DS: Total hot run time: 168752 ms |
ClickBench: Total hot run time: 23.9 s |
FE UT Coverage ReportIncrement line coverage |
FE Regression Coverage ReportIncrement line coverage |
|
run buildall |
FE Regression Coverage ReportIncrement line coverage |
3 similar comments
FE Regression Coverage ReportIncrement line coverage |
FE Regression Coverage ReportIncrement line coverage |
FE Regression Coverage ReportIncrement line coverage |
TPC-H: Total hot run time: 28709 ms |
TPC-DS: Total hot run time: 169538 ms |
ClickBench: Total hot run time: 23.83 s |
|
/review |
|
run buildall |
|
Codex automated review failed and did not complete. Error: Codex completed, but no new pull request review was submitted for the current head SHA. Please inspect the workflow logs and rerun the review after the underlying issue is resolved. |
FE UT Coverage ReportIncrement line coverage |
|
PR approved by anyone and no changes requested. |
TPC-H: Total hot run time: 28785 ms |
TPC-DS: Total hot run time: 168971 ms |
ClickBench: Total hot run time: 23.83 s |
FE Regression Coverage ReportIncrement line coverage |
|
run p0 |
|
run external |
FE Regression Coverage ReportIncrement line coverage |
|
run cloud_p0 |
FE Regression Coverage ReportIncrement line coverage |
|
PR approved by at least one committer and no changes requested. |
|
run buildall |
FE Regression Coverage ReportIncrement line coverage |
TPC-H: Total hot run time: 29369 ms |
TPC-DS: Total hot run time: 165580 ms |
ClickBench: Total hot run time: 24 s |
Problem Summary:
A scalar count(distinct) runs its global aggregate hash-shuffled by the distinct
key (non-empty partitionExpressions), i.e. in parallel across BEs. But the cost
model set the parallelism factor from group-by keys alone, so with no group by it
used factor = 1, costing the aggregate as single-point. This overestimated the
parallel plan and made CBO reject it.
Solution:
Treat a global aggregate as partitioned when it has group by keys or non-empty
partitionExpressions, and scale its cost by beNumber accordingly.