Re: BUG #7648: Momentary index corruption while in hot standby

On Fri, Nov 9, 2012 at 3:35 PM, Jeff Davis <pgsql(at)j-davis(dot)com> wrote:
> On Fri, 2012-11-09 at 22:49 +0000, daniel(at)heroku(dot)com wrote:
>> PGError: ERROR: could not read block 556642 in file "base/16385/2904143.4":
>> read only 0 of 8192
>
> Does that seem wildly off to you, or a little off? Do you think that
> block may have existed in the past, but was truncated by a VACUUM or
> something?

A little off. I think that Tom Lane posted a pretty good mechanism in
-hackers: http://archives.postgresql.org/pgsql-hackers/2012-11/msg00409.php

> Just to clarify, that file belongs to the index in question, right? It's
> not the heap pointed to by the index?

Yes.

>> Immediately thereafter, no occurrences of the error resurfaced and any
>> attempts to reproduce the issue with an identical query were met with
>> failure.
>
> Or a lack of failure, I presume.

Yup.

>> This is reading from a 9.0.8 Postgres.
>
> Any indication whether it's present on other versions or does it appear
> to be isolated to 9.0.X?

This is the only known occurrence to me, ever, but given it's
incredibly ephemeral nature probably glossed over most of the time, I
can't say "it's the only time it's ever happened".

--
fdr