On Wed, Jul 17, 2019 at 12:57 PM Thomas Munro <thomas.mu...@gmail.com> wrote:
> On Wed, Jul 17, 2019 at 12:44 PM Thomas Munro <thomas.mu...@gmail.com> wrote:
> > > #11 0x000055666e0359df in ExecShutdownNode
> > > (node=node@entry=0x55667033a6c8)
> > > at
> > > /build/postgresql-9.6-5O8OLM/postgresql-9.6-9.6.14/build/../src/backend/executor/execProcnode.c:830
> > > #12 0x000055666e04d0ff in ExecLimit (node=node@entry=0x55667033a428)
> > > at
> > > /build/postgresql-9.6-5O8OLM/postgresql-9.6-9.6.14/build/../src/backend/executor/nodeLimit.c:139
> >
> > https://github.com/postgres/postgres/blob/REL9_6_STABLE/src/backend/executor/nodeLimit.c#L139
> >
> > Limit thinks it's OK to "shut down" the subtree, but if you shut down a
> > Gather node you can't rescan it later because it destroys its shared
> > memory. Oops. Not sure what to do about that yet.
>
> CCing Amit and Robert, authors of commits 19df1702 and 69de1718.
Here's a repro (I'm sure you can find a shorter one, this one's hacked
up from join_hash.sql, basically just adding LIMIT):
create table join_foo as select generate_series(1, 3000) as id,
'xxxxx'::text as t;
alter table join_foo set (parallel_workers = 0);
create table join_bar as select generate_series(0, 10000) as id,
'xxxxx'::text as t;
alter table join_bar set (parallel_workers = 2);
set parallel_setup_cost = 0;
set parallel_tuple_cost = 0;
set max_parallel_workers_per_gather = 2;
set enable_material = off;
set enable_mergejoin = off;
set work_mem = '1GB';
select count(*) from join_foo
left join (select b1.id, b1.t from join_bar b1 join join_bar b2
using (id) limit 1000) ss
on join_foo.id < ss.id + 1 and join_foo.id > ss.id - 1;
--
Thomas Munro
https://enterprisedb.com
