On Sun, Sep 2, 2018 at 4:51 AM Stephen Frost <sfrost@xxxxxxxxxxx> wrote:
Greetings,
* Dave Peticolas (dave@xxxxxxxxxx) wrote:
> On Sat, Sep 1, 2018 at 5:09 PM Adrian Klaver <adrian.klaver@xxxxxxxxxxx>
> wrote:
>
> > On 09/01/2018 04:45 PM, Dave Peticolas wrote:
> >
> > > Well restoring from a backup of the primary does seem to have fixed the
> > > issue with the corrupt table.
> >
> > Pretty sure it was not that the table was corrupt but that transaction
> > information was missing from pg_clog.
> >
> > In a previous post you mentioned you ran tar to do the snapshot of
> > $PG_DATA.
> >
> > Was there any error when tar ran the backup that caused you problems?
>
> Well the interesting thing about that is that although the bad table was
> originally discovered in a DB restored from a snapshot, I subsequently
> discovered it in the real-time clone of the primary from which the backups
> are made. So somehow the clone's table became corrupted. The same table was
> not corrupt on the primary, but I have discovered an error on the primary
> -- it's in the thread I posted today. These events seem correlated in time,
> I'll have to mine the logs some more.
Has this primary been the primary since inception, or was it promoted to
be one at some point after first being built as a replica..?
It was the primary since inception. All the problems now appear to have stemmed from the primary due to a bug in 9.6.8 (see other thread). I've since upgraded to 9.6.10.
