On Mon, 2007-12-31 at 12:33 -0500, Tom Lane wrote:
> Simon Riggs <simon@2ndquadrant.com> writes:
> > On Mon, 2007-12-31 at 11:53 -0500, Tom Lane wrote:
> >> The state of the ...0058 file might be explained by the theory that
> >> you'd archived it a bit too late (after the first page had been
> >> overwritten with newer WAL data),
>
> > The interlock with .ready and .done should prevent reuse of a file. So
> > the only way this could happen is if the archive_command queued a
> > request to copy, rather than performing the copy immediately.
> > So I was going to say "thats not possible", but perhaps rsync might
> > become confused by the file renaming mechanism we use?
>
> Actually, the other problem with that theory is that the slave swallowed
> the file without complaint.
No, it barfed. Mason showed us a recovery script, so it came from the
slave.
> Since the WAL reader code does check that
> the page header contains the expected address, this seems to imply that
> what the slave saw must have had 422/58 in it, not the 423/C1 we see
> now. So what needs to be explained is why what Mason is looking at now
> is different from what the slave saw ten days ago.
So the slave did see a problem ten days ago, though I take your point
that the problem we see now may not be the as it was back then.
--
Simon Riggs
2ndQuadrant http://www.2ndQuadrant.com