kamille

> ~Does this improve memory usage~ I forgot it is sketch > > What is the difference between blocked approach and Emit::First with block size? At the end, there are...

@2010YOUY01 make sense, it seems `emit_early_if_necessary` function is actually introduced in the spilling pr #7400. I am checking the related codes about memory control, too. But the `FirstBlocks` is just...

> THank you @Rachelint -- I took a look at this PR and here is some feedback: > > 1. I think it is important to spend time actually showing...

@2010YOUY01 After checking the codes about memory contorl, I think I got it. - `emit_early_if_necessary` is used in `Partial` - and `spill_previous_if_necessary` is used in the final phases They all...

> > @2010YOUY01 After checking the codes about memory contorl, I think I got it. > > > > * `emit_early_if_necessary` is used in `Partial` > > * and `spill_previous_if_necessary`...

> Hi @Rachelint -- please let me know if/when this PR is ready for another look. I think your plan as I understand it is to get this idea working...

@alamb @jayzhan211 @2010YOUY01 I have finish the blocked version common accumulators(avg, count, prim_op) yesterday. Following is the benchmark result. It can make some queries(especially in clickbench) faster 1.1x~1.2x. However, some...

Actually, I found `aggregate_grouping_sets_with_yielding_with_spill` and `aggregate_grouping_sets_source_not_yielding_with_spill` don't conver the spilling to disk logic in `main`... And this pr make the memory usage a bit bigger in the test, so that...

> Benchmark looks amazing 🚀 > > I have an idea to minimize this PR (and make progress easier): > #### Problem with existing code > > Now aggregation execution...

@alamb @jayzhan211 @2010YOUY01 I think this pr is ready now, I made benchmarks in my local, and seems 20% queries in clickbench obviously faster (especially in high cardinality cases). The...