[WIP] Manage group values by blocks in aggregation #11932

Rachelint · 2024-08-10T21:56:34Z

Which issue does this PR close?

Part of #11931

Define new group_idx format, but will just used in GroupValuesRows now, and not expose in interface functions.

block_id(16bit) | block offset(48bit)

Impl a conveter to convert the new format to old format for later compatibility (especially for GroupAccumulators).
Fix tests.
Benchmarks.

Rachelint force-pushed the page-based-group-manage branch from dd5b948 to 680bf3b Compare August 10, 2024 22:59

Rachelint changed the title ~~Manage group values by blocks in aggregation~~ [WIP] Manage group values by blocks in aggregation Aug 11, 2024

draft.

6269fb7

Rachelint force-pushed the page-based-group-manage branch from 680bf3b to 6269fb7 Compare August 11, 2024 05:07

Rachelint added 2 commits August 11, 2024 13:55

impl the new interface functions for all GroupValues impls.

7f5ccd2

just make GroupIdx an internal concept first.

c355368

Rachelint force-pushed the page-based-group-manage branch from 22dd236 to c355368 Compare August 11, 2024 07:47

Rachelint added 6 commits August 11, 2024 16:25

apply it to the main procedure.

f4a71bf

fix spilling(but maybe lead to performance problem currently).

8deecab

fix compile.

92be0ac

fix test.

73a7503

make emit first n exact.

06e7342

fix some cases in streaming_aggregate_test but still some faileds...

56399f7

Rachelint force-pushed the page-based-group-manage branch from ee0b369 to 56399f7 Compare August 11, 2024 11:12

Rachelint closed this Aug 19, 2024