On Tue, 2008-10-07 at 11:04 +0300, Heikki Linnakangas wrote:
> Jeff Davis wrote:
> > I see this problem on 8.3.3 now, too. Originally, I suppose my test was
> > not long enough, but now I see the problem after about 10 minutes of
> > running.
>
> I ran the script for about 30 minutes on CVS HEAD, and the index didn't
> grow at all after the first three iterations. Are you sure you didn't
> have a long-running transaction open that prevented vacuum from working?
> Did the heap bloat as well, or just the index?
In the VACUUM VERBOSE output I included, you can see that the heap is
only 785 pages (after 200M rows went through that table), and it
maintains that consistently. That means to me that the VACUUMs are
running and properly freeing the space in the heap. And the output seems
to indicate that it thinks it can re-use those index pages, it just
doesn't.
This is a bit non-deterministic: the index pages will be steady for a
while, and then jump all of a sudden. Maybe it only happens when
resources are strained? I will try on a few other machines today and see
if I can identify the conditions a little more clearly.
Regards,
Jeff Davis