On Fri, 3 May 2002, Holger Marzen wrote:
> On Thu, 2 May 2002, Scott Marlowe wrote:
>
> > And after that, the table started growing after all vacuum fulls. I can't
> > reproduce it now. I blew away the table, recreated it, and now it doesn't
> > do that. I blew away the database, and installed 7.2.1 and it won't do
> > it. I reinstalled 7.2 and blew away the database and it still won't do
> > it.
> >
> > So, I don't know why it WAS doing it, but it isn't now. And of course,
> > it's my personal workstation, so no log files. :-( I'll see if I can
> > reproduce this though, as I've seen similar problems pop up once in the
> > past with a seldom vacuumed database that reached a point where it stopped
> > freeing dead tuples. Very bothersome.
>
> Looks like something that happens only if counters or OIDs are big.
> Possible integer overflow in the vacuum code?
Prior to the test I had been running 100,000 transaction pgbench tests
repeatedly. I'll try running a few million transactions first and see if
the problem surfaces.
Thanks for the suggestions.