Thread: Auto-timeout on all queries
> 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. -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
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. Not so easy - we use SIGALRM in locking... But why not just use CANCEL to stop too long running queries? Vadim
> 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. > > Not so easy - we use SIGALRM in locking... Oh. :-) I forgot. > > But why not just use CANCEL to stop too long running queries? Good question. -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
> From: Bruce Momjian <maillist@candle.pha.pa.us> > Date: Mon, 5 Jul 1999 21:19:54 -0400 (EDT) > CC: PostgreSQL-general <pgsql-general@postgreSQL.org> > Content-Type: text/plain; charset=US-ASCII > Sender: owner-pgsql-general@postgreSQL.org > Precedence: bulk > > > 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. > > -- > Bruce Momjian | http://www.op.net/~candle > maillist@candle.pha.pa.us | (610) 853-3000 > + If your life is a hard drive, | 830 Blythe Avenue > + Christ can be your backup. | Drexel Hill, Pennsylvania > 19026 I think we'd use it. At least on our development machines, which have occasionally suffered from the ill effects of an excessively ambitious query. How would the interface to the user look? -- Karl DeBisschop <kdebisschop@spaceheater.infoplease.com> 617.832.0332 (Fax: 617.956.2696) Information Please - your source for FREE online reference http://www.infoplease.com - Your Ultimate Fact Finder http://kids.infoplease.com - The Great Homework Helper
> > > 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. > > > > -- > > Bruce Momjian | http://www.op.net/~candle > > maillist@candle.pha.pa.us | (610) 853-3000 > > + If your life is a hard drive, | 830 Blythe Avenue > > + Christ can be your backup. | Drexel Hill, Pennsylvania > > 19026 > > I think we'd use it. At least on our development machines, which have > occasionally suffered from the ill effects of an excessively ambitious > query. > > How would the interface to the user look? I assume it would be a new 'SET' option, like "SET TIMEOUT=60", but we already use alarm() for something else, so it may be more difficult than I thought. Not sure. -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
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.