> > 1. When you have created temp table and just killed frontend. Backend
> > realize that connection is broken and somehow did not remove the table
> > from pg_class. Then it will exists on the disk and in pg_class.
>
> Hmm. I said to myself "no way", and did "create temp table foo ..."
> followed by killing psql from another window. By golly, the pg_temp
> file was still there, and it was still listed in pg_class, just as you
> said. But then I haven't been able to repeat it in quite a few tries.
> So there's a bug there, but it's not too easy to reproduce. Do you
> have any idea what contributing conditions might be involved?
I always get it when run vacuumlo from contribs on database with large (20000-30000)
number of LOs.
> > 2. When you do operations with large objects (they are treated as
> > relations also) and you do rollback or again connect broken. Or you do
> > lots of lo_unlink and rollback or just broke connection.
>
> lo_unlink is not safe to rollback, any more than a plain drop table is.
> This is a known deficiency and probably will be for a while. In the
> meantime the standard advice is "don't do that".
Actually this mean that postgres does not have safe support for BLOBs.
Hopefully tuple stuff will be in the next release. I've tested alpha patch and
it works quite fine except some queries was broken.
--
Sincerely Yours,
Denis Perchine
----------------------------------
E-Mail: dyp@perchine.com
HomePage: http://www.perchine.com/dyp/
FidoNet: 2:5000/120.5
----------------------------------