Skip site navigation (1) Skip section navigation (2)

Re: Socket timeouts (probably)

From: Oliver Jowett <oliver(at)opencloud(dot)com>
To: Sergei Georgiev <sergei(dot)georgiev(at)gmail(dot)com>
Cc: pgsql-jdbc(at)postgresql(dot)org
Subject: Re: Socket timeouts (probably)
Date: 2005-02-24 05:12:02
Message-ID: 421D6222.500@opencloud.com (view raw or flat)
Thread:
Lists: pgsql-jdbc
Sergei Georgiev wrote:

> I have a really strange problem here. I've got two Fedora 2 machines -
> one for the database and one for a Java application to be run. When a
> transaction from the application is run against the database and the
> network cable is unplugged (or the machine is shut down) the database
> continues to hold the lock on the tables, which are used in the
> transaction. The lock is released in about 10-12 minutes, which in my
> case is not acceptable, because if another client starts a transaction
> against these locked tables his UI ties his hands until the lock on
> the tables is released and the transaction is complete.

I raised this on -hackers recently. Ideally I'd like to see a 
transaction timeout implemented on the server side for exactly this 
case. See 
http://archives.postgresql.org/pgsql-hackers/2005-02/msg00397.php (the 
archive indexes seem to be broken at the moment, but the direct link works).

You may want to try tweaking /proc/sys/net/ipv4/tcp_keepalive_* on the 
server to make TCP keepalives notice the dead connection faster (these 
are global settings, though). There are corresponding setsockopt() calls 
you could make on a per-socket basis if you are prepared to tweak the 
server code.

-O

In response to

Responses

pgsql-jdbc by date

Next:From: Markus SchaberDate: 2005-02-24 09:39:14
Subject: Re: [postgis-devel] JDBC & WKB - 400% overhead
Previous:From: Kris JurkaDate: 2005-02-24 05:00:15
Subject: Re: DatabaseMetaData.getIndexInfo and function-based indexes

Privacy Policy | About PostgreSQL
Copyright © 1996-2014 The PostgreSQL Global Development Group