Re: all_visible replay aborting due to uninitialized pages

On Mon, Nov 11, 2013 at 01:42:07AM +0100, Andres Freund wrote:
> Hi,
>
> On 2013-11-10 17:40:31 -0700, Noah Yetter wrote:
> > Like your customer, this bug has blown up my standby servers, twice in the
> > last month: the first time all 4 replicas, the second time (mysteriously
> > but luckily) only 1 of them.
> >
> > At any rate, since the fix isn't available yet, is/are there any
> > configuration changes that can be made or maintenance procedures that can
> > be undertaken to prevent or reduce the probability of this bug popping up
> > again in the meantime? I really can't afford to be without my standby
> > servers during the holidays, even for the few hours it takes to build a new
> > one.
>
> The fix is included in 9.2.5, it's just not noted in the release notes.

Yes, I missed it because I didn't understand the importance of these
commit messages:

commit 17fa4c321ccf9693de406faffe6b235e949aa25f
Author: Robert Haas <rhaas(at)postgresql(dot)org>
Date: Thu Jun 6 10:15:45 2013 -0400

Ensure that XLOG_HEAP2_VISIBLE always targets an initialized page.

Andres Freund

commit 4c641d994e19676ef2fec574d52d2156ffc2b3ce
Author: Robert Haas <rhaas(at)postgresql(dot)org>
Date: Thu Jun 6 10:14:46 2013 -0400

Backport log_newpage_buffer.

Andres' fix for XLOG_HEAP2_VISIBLE on unitialized pages requires
this.

Sorry.

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ Everyone has their own god. +