On Tue, Feb 15, 2011 at 7:59 AM, Christopher Browne
<cbbrowne at ca.afilias.info> wrote:
> Hah.  Almost certainly some file got unlinked, but some backend process
> remained holding onto the file descriptor, so the space did not get
> returned to the filesystem until that backend process terminated.
>
> It's pretty likely that you could have done something less dramatic than
> restarting PostgreSQL, but the restart was certainly an easy way to
> flush out any such hanging FDs.

Yup!  :)


> I'd anticipate that you'll find things work fine, now.

I expect so, I'll know for sure Saturday, when we're upgrading the master
(finally the database upgrade purchase request got approved so I have
a shiny new server) and then we'll turn replication back on.


> There's a mystery there, still, in terms of "what was that lingering
> file descriptor???", but it's entirely possible that nothing similar
> will ever recur.

I would be entirely satisfied if this problem never came back.

Thanks!!
Aleksey