On 2014-11-20 11:22:37 -0500, Robert Haas wrote:
> On Thu, Nov 20, 2014 at 4:21 AM, Jeff Davis <pgsql@j-davis.com> wrote:
> >> The
> >> bad news, to borrow a phrase from Peter Geoghegan, is that it's an
> >> unprincipled deadlock; a user confronted with the news that her
> >> parallel scan has self-deadlocked will be justifiably dismayed.
> >
> > You seem to be raising this as a show-stopping problem, and I'm not
> > convinced that it is.
>
> Well, what I'm saying is that at very minimum we have to be able
> detect deadlocks, and we have two plausible designs for avoiding that:
>
> 1. Modify the deadlock detector to know about lock groups.
>
> 2. Propagate pre-existing locks from the user backend to all the workers.
>
> I initially proposed #1, but now I think #2 solves more of the
> problems for less code.
Except that it opens us up for all kinds of concurrency bugs. I'm pretty
strictly set against granting any self exclusive locks en-masse. If we
do this by default for all granted locks when starting a worker backend
it'll get *so* much harder to reason about correctness. Suddenly locks
don't guarantee what they used to anymore. We'll e.g. not be able to
rely that a CheckTableNotInUse() + AEL makes it safe to
drop/rewrite/whatever a relation - even if that happens in the main
backend.
Greetings,
Andres Freund
-- Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services
