We have a setup where nginx is used as a load balancer for a squid backend in accelerator mode, and nginx is getting timeout trying to connect to squid. Although at any time nginx only has a couple of hundred connections open to the squid server, the squid server has over 10k tcp connection to the nginx server in TIME_WAIT state. The squid error log also shows frequent "TCP connection failed" error for peers and parents, which makes me think it's running out of connections. The system message log doesn't show any error. Can anyone shed a light on where I should look at? software versions: CentOS 5.2, squid 2.6 stable 11, nginx 0.6.32 Thanks a bunch.
