On Mon, 2007-12-31 at 13:14 -0500, Tom Lane wrote:
> "Mason Hale" <masonhale@gmail.com> writes:
> > [postgres@dev-db-2 wal_archive]$ od -x 000000010000042200000058 | head -n15
> > 0000000 d05e 0002 0001 0000 0423 0000 0000 c100
> > 0000020 f7df 472e e701 4728 0000 0100 2000 0000
> > 0000040 a1db 81e6 0423 0000 0068 c000 0000 0000
> > 0000060 0048 0000 002c 0000 0000 0000 0423 0000
> > 0000100 0020 c100 0000 0000 0000 0000 0001 0000
> > 0000120 0000 0000 780b 2ede 9f68 00f5 7834 0000
> > 0000140 2f4d 0001 1f35 4774 0000 0000 0000 0000
> > 0000160 0000 0000 0000 0000 0000 0000 0000 0000
> > *
> > 0020000 d05e 0000 0001 0000 0422 0000 2000 5800
>
> Something else interesting about that: the apparent interloper page
> contains just a single WAL record, which appears to be a shutdown
> checkpoint bearing the timestamp "Thu Dec 27 2007, 16:55:01 EST".
> Not sure if Mason can correlate that with any recent activity...
OK, well thats clearly an archive overwrite then.
This is happening on the slave and it is a page-level overwrite, of WAL,
not an original page being propagated to WAL via full_page_writes. So it
is either unconnected to the index problems on the primary, or they are
linked in some way, which seems most likely. That tends to argue against
a pure hardware problem. So it could be device drivers, OS or Postgres
causing this. Don't see how it can be the latter...
--
Simon Riggs
2ndQuadrant http://www.2ndQuadrant.com