Re: [GENERAL] Auto-timeout on all queries

Since were making a wishlist on this topic (Though it may be harder
than it at first seemed), I'd like a SET option, with a
transaction-based override. That way, a casual query can be
intercepted, at least warning me of the long query if I forgot a
constraint on a join for instance. THen, after I've looked at it, I
may want to force it through. Or I may rewrite it and try running it
again under the same TIME/TUPLE limits.

Kane Tao wrote:
>
> Been thinking about this time thing...wouldn't it be more flexible and more
> useful to do it by transaction rather than a global server setting? Plus
> time is a particularly bad measure as loaded servers take longer to execute
> queries etc...
>
> -----Original Message-----
> From: Peter Eisentraut <peter(at)pathwaynet(dot)com>
> To: Bruce Momjian <maillist(at)candle(dot)pha(dot)pa(dot)us>
> Cc: Bob Kruger <bkruger(at)mindspring(dot)com>; PostgreSQL-general
> <pgsql-general(at)postgreSQL(dot)org>
> Date: Tuesday, July 06, 1999 5:26 PM
> Subject: Re: [GENERAL] Auto-timeout on all queries
>
> >On Mon, 5 Jul 1999, Bruce Momjian wrote:
> >
> >> > I would like to make one small request for future releases. Is there a
> >> > setup parameter that could be enabled that would put a timeout for a
> query
> >> > that runs too long or endangers the integrity of the system? With the
> >> > systems that I use, there are times that queries are entered that
> contain
> >> > errors or run so long that they eventually cause a system crash. It
> would
> >> > be most helpful if a timeout limit on queries could be enabled that
> would
> >> > stop a query after so much time and release the resources back to the
> system.
> >>
> >> Does anyone thing this would be a nice feature? We can easily do it
> >> with alarm()/elog(), but I want to know if it would be valuable.
> >
> >Oh yeah!
> >
> >Time out in minutes, "time" out in tuples, all with an instructive error
> >message if exceeded. How about
> >
> >SET TIMEOUT '5 min';
> >SET TUPLE LIMIT 10000;
> >
> >If you can really "easily" do it, I say go for it.
> >
> >--
> >Peter Eisentraut
> >PathWay Computing, Inc.
> >
> >
> >