Re: Connection pooling
От | Max Zorloff |
---|---|
Тема | Re: Connection pooling |
Дата | |
Msg-id | op.tyfdqndnll0p5y@1-rtt202dnf3uds обсуждение исходный текст |
Ответ на | Re: Connection pooling ("Scott Marlowe" <scott.marlowe@gmail.com>) |
Список | pgsql-general |
On Sat, 08 Sep 2007 19:28:52 +0400, Scott Marlowe <scott.marlowe@gmail.com> wrote: > On 9/7/07, Max Zorloff <zorloff@gmail.com> wrote: >> On Fri, 07 Sep 2007 10:58:36 +0400, Marko Kreen <markokr@gmail.com> >> wrote: >> >> >> The pgpool (I tried 3.1, 3.4 and pgpool-II 1.2) works fine but has >> the >> >> following problem - after some time it >> >> just "hangs", and if I try to connect to it with psql it just hangs >> >> indefinitely. After restart >> >> it works fine again. I turned off persistent connections in php so >> it's >> >> not that. Does anybody >> >> have the same problem? >> > >> > All symptoms point to the same problem - your app fails to >> > release server connections for reuse. >> > >> > If the problem is that PHP fails to disconnect connection, >> > although the transaction is finished, you could run pgbouncer >> > in more relaxed mode - pool_mode=transaction. Also setting >> > client_idle_timeout to something may help debugging. >> >> > If the problem is uncommitted transactions, you could set >> > query_timeout to some small number (1-3) to see where >> > errors appear. >> > >> > Both timeouts are not something I would put into productions >> > config, so the code should be fixed still... >> >> pgbouncer does not have this problem, only pgpool does. >> pgbouncer has the problem of being very slow. >> >> i thought php released connections at the end of script? >> and also if i had this problem pgpool would hang in a few seconds >> because the server has some load. > > It does if you're running without pg_pconnect (i.e. using regular > pg_connects) and if the script doesn't crash the apache/php backend > it's running in. > > Are you using pg_pconnect and / or having crashing apache backends? I specifically turn off pconnects and I don't think I crash backends, never saw mistakes of that type.
В списке pgsql-general по дате отправления: