Hi Jeff, > > Just iptables. The OUTPUT chain's policy on all tables is ACCEPT, and we > have no OUTPUT rules (only INPUT). OUTPUT is wide open. > > We did find this in /usr/share/pyshared/twisted/internet/process.py: > > 601 try: > 602 import resource > 603 maxfds = resource.getrlimit(resource.RLIMIT_NOFILE)[1] + 1 > 604 # OS-X reports 9223372036854775808. That's a lot of fds > to close > 605 if maxfds > 1024: > 606 maxfds = 1024 > 607 except: > 608 maxfds = 256 > > The 1024 was interesting because we seemed to be having the problem around > 500 calls, possibly 512? This is from python-twisted 8.1.0. We tried to > adjust line 606 to maxfds = 4096, and make what seemed like appropriate > updates in /etc/security/limits.conf: > > root soft nofile 1024 > root hard nofile 16384 # was 1024 > > This is all new to us so it's very possible we're on the wrong track. >
You shouldn't need to change anything there, because the relay already does it when it starts: http://projects.ag-projects.com/projects/mediaproxy/repository/entry/mediaproxy/relay.py#L93 > Today we split the traffic between two relays. Each handles around 425 calls > at peak usage. We made these twisted+limits updates on one relay only. The > quantity of the Traceback errors is lower at 22 so far today on one relay, > and 3 on the other. The first apparence was with less than 100 calls on the > relay. The two relay machines have the same configurations, except for IP > address. I don't have an explanation why they are higher on one relay than > the other. > Unfortunately I have no explanation for this right now. I guess that there are no network issues like IP addresses going up and down, are there? -- Saúl Ibarra Corretgé AG Projects _______________________________________________ Users mailing list [email protected] http://lists.opensips.org/cgi-bin/mailman/listinfo/users
