race condition in CatchupInterruptHandler was:(Re: [HACKERS] Review: support for multiplexing SIGUSR1)

On Fri, Jul 17, 2009 at 3:30 PM, Jaime
Casanova<jcasanov(at)systemguards(dot)com(dot)ec> wrote:
>>
>> i wasn't able to repeat this on a new instalation and of
>> course i can't swear this is your patch fault...
>>
> this is not your patch fault but an existing bug, i repeat that
> behaviour in an unpatched source tree...
>

ok, i reproduced this again and again (i have tried only in 8.4.0)
with the following steps:

- open five terminals, and create empty databases pgbench1 and pgbench2
- on first execute: make installcheck
- on second: pgbench -i -s1000 pgbench1
- on third: pgbench -i -s1000 pgbench2
- on fourth: explain analyze
with q as (select * from generate_series(1, 1000000))
select * from q a, q b, q c, q d, q e, q f;

- on fifth:
execute two pg_cancel_backend for killing things happen in
terminals first, second or third (they have to actually cancel
something, eg: return true)
then pg_cancel_backend the explain analyze and repeat the
pg_cancel_backend, that interrupt will be pending and when you rerun
the explain analyze (or any other query in that same session) it will
be get cancelled immediately

ok, that's as far as i can go with this... now, i'm going to return to
my assignment as rrr

--
Atentamente,
Jaime Casanova
Soporte y capacitación de PostgreSQL
Asesoría y desarrollo de sistemas
Guayaquil - Ecuador
Cel. +59387171157