On Jan 23, 2007, at 8:57 AM, Greg Cox wrote:
I've got a cluster of 3 machines running Apache 2.0.52 (hello, stuck-on-RHEL4!) as a front-end to 2 Weblogic apps. One app (A) runs fine all the time; one app (B) will hold open weblogic connections to an outside vendor when said vendor explodes. When B blocks, it starts sucking up spare worker threads on apache (since it's proxying to weblogic) until there aren't any left, and both A and B become unreachable from the web for even calls to server-status.
We can't really speak to the WebLogic plugin, but don't they have a timeout tunable that you could use to reap hung connections? Alternatively, what happens when you play with the timeout value on the front-end?
Are you using the WebLogic plugin? If you are using our own mod_proxy, it does have ways to time out the backend.
A and B have separate names and are IP-differentiated (not name- based) VirtualHosts. Is there a way to make one Apache daemon handle separate thread pools for the separate virtualhosts and their proxies? I'd rather not set up separate daemons, since the people who would administer it could easily get confused over the maintenance upgrades down the line, but I'm coming up blank.
You can easily set up two instances of httpd running from the same code. Just set up two server root directories and start both with the -d flag from your init.d scripts. As you are binding to separate IP addresses, you should have no problems there.
S. -- sctemme@xxxxxxxxxx http://www.temme.net/sander/ PGP FP: 51B4 8727 466A 0BC3 69F4 B7B8 B2BE BC40 1529 24AF
Attachment:
smime.p7s
Description: S/MIME cryptographic signature
|