Re: Something is broken in logical decoding with CLOBBER_CACHE_ALWAYS

On 2015-01-04 21:54:40 +0100, Andres Freund wrote:
> On January 4, 2015 9:51:43 PM CET, Andrew Dunstan <andrew(at)dunslane(dot)net> wrote:
> >
> >On 12/15/2014 12:04 PM, Andres Freund wrote:
> >
> >>> I think the safest fix would be to defer catchup interrupt
> >processing
> >>> while you're in this mode. You don't really want to be processing
> >any
> >>> remote sinval messages at all, I'd think.
> >> Well, we need to do relmap, smgr and similar things. So I think
> >that'd
> >> be more complicated than we want.
> >>
> >
> >
> >
> >Where are we on this? Traffic seems to have gone quite but we still
> >have
> >a bunch of buildfarm animals red.
>
> I've a simple fix (similar too what I iriginally outkined) which I
> plan to post soonish. I've tried a bunch of things roughly in the vein
> of Tom's suggestions, but they all are more invasive and still
> incomplete.

Attached.

Note that part 1) referenced in the commit message is actually
problematic in all branches. I think we actually should backpatch that
part all the way, because if we ever hit that case the consequences of
the current coding will be rather hard to analyze. If others think so as
well, I'm going to split the commit into two parts, so commit messages
for < 9.4 won't reference logical decoding. Since there hasn't been a
report of that error for a long while (~8.1 era), I can also live with
not backpatching further than 9.4.

Greetings,

Andres Freund

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services