I'm probably wrong, but I thought vacuum may invalidate stuff which semi-required the cache to be flushed. :) I'll go take a look through as-well but it's hard to imagine this being overlooked for so long.
Sorry Alvaro, I haven't gone out to look at vacuum in awhile so I ain't much help.
On 9/16/05, Alvaro Herrera <alvherre@alvh.no-ip.org> wrote: On Fri, Sep 16, 2005 at 04:41:39PM -0400, Jonah H. Harris wrote:
> Was it relcache related?
I don't see how -- any user of a relcache entry needs to heap_open() or
relation_open() the table and acquire an appropiate lock. Any such call
would block because of the lock that VACUUM FULL acquires on the relation.
> On 9/16/05, Alvaro Herrera <alvherre@alvh.no-ip.org> wrote:
> >
> > Hackers,
> >
> > I'm reading the vacuum code and I just noticed that the routines
> > move_plain_tuple and move_chain_tuple expect the dest and source blocks
> > to be locked, and unlock them at exit. The only caller of both is
> > repair_frag, whose only caller in turn is full_vacuum_rel. Same thing
> > for update_hint_bits.
> >
> > So, the only callers of both has already acquired appropiate locks at
> > the relation level -- nobody is going to be modifying the blocks while
> > they proceed. So why bother locking the pages at all? Is there a
> > reason or is this an historical accident?
--
Alvaro Herrera -- Valdivia, Chile Architect, www.EnterpriseDB.com
"Now I have my system running, not a byte was off the shelf;
It rarely breaks and when it does I fix the code myself.
It's stable, clean and elegant, and lightning fast as well,
And it doesn't cost a nickel, so Bill Gates can go to hell."
--
Respectfully,
Jonah H. Harris, Database Internals Architect
EnterpriseDB Corporation
http://www.enterprisedb.com/