On Friday 09 May 2014 08:36:04 David G Johnston wrote:
> This seems to likely be the same, still open, bug reported previously:
>
> No Number Assigned:
> http://www.postgresql.org/message-id/CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpw
> si27vpm1z5sA@mail.gmail.com
>
> #8464
> http://www.postgresql.org/message-id/E1VN53g-0002Iy-Il@wrigleys.postgresql.o
> rg
>
> #8470 is referenced in the first thread as well...though that is
> specifically a performance issue and not a query bug.
>
> The recommended work-around is to move the sub-query using the "FOR UPDATE"
> into a CTE.
Thanks for those pointers, it certainly looks like the same issue (the only
difference being the size of the limit) and there has been a good amount of
discussion there.
I'll try the CTE workaround, although that leaves a bad taste in my mouth.
From the discussions in the 1st thread, I wonder wether raising the isolation
level to repeatable read would also fix the issue ?
In any case, testing will take time because the bug triggers less than once a
day and I haven't yet managed to reproduce it locally.
ps: sorry I'm only seeing your answer now, it helps if you cc me when
answering the list.
--
Vincent de Phily
