Re: Delete query takes exorbitant amount of time

On Tue, 29 Mar 2005, Simon Riggs wrote:

> On Tue, 2005-03-29 at 10:31 -0500, Tom Lane wrote:
> > Simon Riggs <simon(at)2ndquadrant(dot)com> writes:
> > > If action is NO ACTION or RESTRICT then
> > > we need to SELECT at most 1 row that matches the criteria
> > > which means we can use LIMIT 1
> >
> > > If action is CASCADE, SET NULL, SET DEFAULT then
> > > we need to UPDATE or DELETE all rows that match the criteria
> > > which means we musnt use LIMIT and need to use FOR UPDATE
> >
> > Huh? UPDATE/DELETE don't use FOR UPDATE. I think you have failed
> > to break down the cases sufficiently. In particular it matters which
> > side of the RI constraint you are working from ...
>
> OK... too quick, sorry. I'll hand over to Stephan for a better and more
> exhaustive explanation/analysis... but AFAICS we *can* always know the
> correct formulation of the query prepare time, whether or not we do
> currently.

We currently use FOR UPDATE on the NO ACTION check, because otherwise we
might get back a row that's already marked for deletion by a concurrent
transaction. I think that's intended to wait and succeed, not fail.