Hi Tomas,
I rebased the code and resolved the comments you attached, some unresolved
comments are explained in 0002-fixes.patch, please take a look.
I also make the hash spill working for parallel grouping sets, the plan looks like:
gpadmin=# explain select g100, g10, sum(g::numeric), count(*), max(g::text)
from gstest_p group by cube (g100,g10);
QUERY PLAN
-------------------------------------------------------------------------------------------
Finalize MixedAggregate (cost=1000.00..7639.95 rows=1111 width=80)
Filtered by: (GROUPINGSETID())
Group Key: ()
Hash Key: g100, g10
Hash Key: g100
Hash Key: g10
Planned Partitions: 4
-> Gather (cost=1000.00..6554.34 rows=7777 width=84)
Workers Planned: 7
-> Partial MixedAggregate (cost=0.00..4776.64 rows=1111 width=84)
Group Key: ()
Hash Key: g100, g10
Hash Key: g100
Hash Key: g10
Planned Partitions: 4
-> Parallel Seq Scan on gstest_p (cost=0.00..1367.71 rows=28571 width=12)
(16 rows)
Thanks,
Pengzhou
Hi,
unfortunately this got a bit broken by the disk-based hash aggregation,
committed today, and so it needs a rebase. I've started looking at the
patch before that, and I have it rebased on e00912e11a9e (i.e. the
commit before the one that breaks it).
Attached is the rebased patch series (now broken), with a couple of
commits with some minor cosmetic changes I propose to make (easier than
explaining it on a list, it's mostly about whitespace, comments etc).
Feel free to reject the changes, it's up to you.
I'll continue doing the review, but it'd be good to have a fully rebased
version.
regards
--
Tomas Vondra https://urldefense.proofpoint.com/v2/url?u=http-3A__www.2ndQuadrant.com&d=DwIBAg&c=lnl9vOaLMzsy2niBC8-h_K-7QJuNJEsFrzdndhuJ3Sw&r=L968W84_Yb9HJKtAAZUSYw&m=hYswOh9Appfj1CipZAY8-RyPSLWnua0VLEaMDCJ2L3s&s=iYybgoMynB_mcwDfPDmJv3afu-Xdis45lMkS-_6LGnQ&e=
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
