Open fabriziomello opened 4 days ago
Discovered this one recently as well. I had to move it before the chunkwise aggregation, or else it would lead to some very weird planning effects: chunkwise aggregation is used unconditionally and removes all the old paths, but after that we make a cost-based decision about these hash agg paths. This change is not merged yet.
Do you think we don't need it already? I can imagine it has value for vectorized aggregation, because currently Partial GroupAggregate
cannot be vectorized, only the Partial HashAggregate
. Would be interesting to look at the plan changes.
Discovered this one recently as well. I had to move it before the chunkwise aggregation, or else it would lead to some very weird planning effects: chunkwise aggregation is used unconditionally and removes all the old paths, but after that we make a cost-based decision about these hash agg paths. This change is not merged yet.
Do you think we don't need it already? I can imagine it has value for vectorized aggregation, because currently
Partial GroupAggregate
cannot be vectorized, only thePartial HashAggregate
. Would be interesting to look at the plan changes.
I'm not a planner expert but this is an attempt after a pvt conversation with @svenklemm ... the motivation of this test is to fix this SDC: https://github.com/timescale/Support-Dev-Collab/issues/1859
Codecov Report
All modified and coverable lines are covered by tests :white_check_mark:
Additional details and impacted files
```diff @@ Coverage Diff @@ ## main #7077 +/- ## ========================================== + Coverage 80.06% 81.78% +1.71% ========================================== Files 190 200 +10 Lines 37181 37350 +169 Branches 9450 9740 +290 ========================================== + Hits 29770 30547 +777 + Misses 2997 2887 -110 + Partials 4414 3916 -498 ```:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.