After a lot of pain, I found that the real root of the problem is that JK is having trouble communicating with Tomcat:
[Sun Aug 10 07:02:14 2003] [jk_ajp13_worker.c (635)]: Error connecting to the Tomcat process. [Sun Aug 10 07:02:14 2003] [jk_ajp13_worker.c (848)]: In jk_endpoint_t::service, send_request failed in send loop 0 [Sun Aug 10 07:02:14 2003] [jk_connect.c (143)]: jk_open_socket, connect() failed errno = 111 This is showing up every 15 requests or so for me. I have seen a ton of threads about this both here and elsewhere, but no clear cut way to resolve it. I don't think it is a timeout problem, and I disabled MaxClientsPerRequest in Apache to try and help. Nothing I've done so far has helped. Thanks, Cris -----Original Message----- From: Cristopher Daniluk Sent: Saturday, August 09, 2003 10:52 AM To: Tomcat Users List Subject: RE: Oracle connection pooling I agree. I think its something else, but the real problem is - why doesn't the Oracle connection pool at least check the TCP state of the conn before it works? Is it because it sucks and shouldn't be used? As best I can tell, we're following all the documentation for the oracle pool to the letter. -----Original Message----- From: Eric J. Pinnell [mailto:[EMAIL PROTECTED] Sent: Saturday, August 09, 2003 12:07 AM To: Tomcat Users List Subject: Re: Oracle connection pooling Hi, Firewalls, and I'm not speaking for all of them, as a rule of thumb close _idle_ connections after an hour. Your connections should be done well before that. -e On Fri, 8 Aug 2003, Cristopher Daniluk wrote: > We are running a fairly large ecommerce site consisting of 3 apache > 2.0.48 servers being balanced between using UltraMonkey/ldirector and > 2 tomcat servers with AJP1.3. Everything is Linux except the back end > which is a fairly hefty and firewalled Sun running Oracle8i. > > The webapps build a connection pool to oracle via the > oracle.jdbc.pool.* connection pool. The pool instantiates fine and > everything is great, but gradually the pool begins to break down. Idle > connections are closed by Oracle (not sure if its oracle itself or the > fw..), but the oracle pool doesn't figure it out. If someone happens > to get the dead connection, Tomcat completely hangs on all threads > until the session-timeout expires. Even the session replication code > stops and the apache servers and the other tomcat in the cluster mark > it offline. During the peak traffic it isn't so bad but at the end of > the day after load goes back down, it's a big problem. > > We initially used DBCP but it didn't work for crap with Oracle. Is > there a better pool to use with Oracle and Tomcat in a cluster > environment and if not, is there a way we can get the Oracle pool to > recycle some of these bad connections without blowing up the server? > > Thanks, > > Cris Daniluk > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]