Mailing-List: contact users-help@tomcat.apache.org; run by ezmlm
Precedence: bulk
Reply-To: "Tomcat Users List" <users@tomcat.apache.org>
Received-SPF: pass (athena.apache.org: domain of jak-tomcat-user@m.gmane.org
 designates 80.91.229.2 as permitted sender)
To: users@tomcat.apache.org
From: "Eric B." <ebenze@hotmail.com>
Subject: Re: dbcp pool freezing
Date: Wed, 4 Feb 2009 14:31:08 -0500
Lines: 53
Message-ID: <gmcqea$npp$1@ger.gmane.org>
References: <gm9u4r$rpu$1@ger.gmane.org>
 	<0AAE5AB84B013E45A7B61CB66943C17215B600C148@USEA-EXCH7.na.uis.unisys.com>
 <gma5q2$nlv$1@ger.gmane.org>
 <0AAE5AB84B013E45A7B61CB66943C17215B604A5F4@USEA-EXCH7.na.uis.unisys.com>
 <gma9am$4rr$1@ger.gmane.org> <4989DF6B.3030201@christopherschultz.net>
Sender: news <news@ger.gmane.org>

"Christopher Schultz" <chris@christopherschultz.net> wrote in message 
news:4989DF6B.3030201@christopherschultz.net...
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Eric,
>
> Eric B. wrote:
>> I'm happy to test out an validate that theory, but based on what I've 
>> seen
>> so far, this isn't the case at all.
>
> Try deploying nothing in your application but the following JSP. Browse
> to it, and intentionally leak a connection. Watch your catalina.out file
> to see if an error message occurs ~20 seconds later. If you see such
> messages, Chuck's suggestion that you have some runaway looping code are
> probably something to look into.


Thanks for the pointers.  After a bit more troubleshooting, I finally got 
the abandoned connection tracking to work, and indeed managed to find a 
connection leak in a very obscure place (not in a loop).  However, during 
this process I noticed the following:

The abandoned connection tracking only seems to fire when requesting a new 
connection.  It is not a seperate background thread that cleans up abandoned 
connections.  Consequently, if all connections are being legitimately used 
when a new one is requested, the new one will go into a wait state.  Once a 
legit connection is closed, the thread waiting for a connection will be able 
to continue processing.

However, if all the connection pool is maxed out with leak'ed connections 
that have not exceeded the abandonedTimeout, the thread requesting a new 
connection will go into a wait state and wait perpetually until a second new 
thread tries to access a connection and triggers the abandoned connection 
cleanup.

I realize that the abandoned connection tracking is not something that a 
good app should rely upon, however, it is definitely a useful feature to 
have to ensure that your app doesn't crap out.   However, it is a little 
misleading in the belief that abandoned connections will automatically be 
returned to the pool automatically rather than only be returned once a new 
connection is requested.

Has there ever been any thought to doing abandoned connection tracking as a 
background thread, or was the current implementation decided upon to save on 
resources?

Thanks!

Eric


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
For additional commands, e-mail: users-help@tomcat.apache.org