Re: TOAST table repeatedly corrupted

On Wed, May 9, 2018 at 11:54 AM, Niles Oien <noien(at)nso(dot)edu> wrote:
> I think this is probably a bug? Every time it happens
> it affects the same table, hmi.rdvtrack_fd05. The hardware
> we're running on is fairly new, disks and memory are
> not throwing errors, and we've never had problems with any
> other tables. We're running on CentOS 6.8 - again, not brand
> new but hardly decrepit, either. I've rebuilt the table
> several times. This time I'd like to give the PG community
> a chance to chase it down before I do that.

amcheck could potentially reveal more information about the problem.
The version on Github works with 9.5, and there are Redhat/CentOS pgdg
packages you can use. See:

https://github.com/petergeoghegan/amcheck

The "heapallindexed" option may be of particular interest to you. It
verifies that the table has matching rows for a target index, in
addition to testing the structure of the target B-Tree index itself.
It doesn't do anything special with TOAST tables just yet, but it's
still the best general corruption test we have.

--
Peter Geoghegan