CSM Real server taking only a few sessions

chris.super163 · ‎07-19-2006

Running into a weird issue with a real server in one of our VIPs. This real starting yesterday will not take many sessions. Say normal traffic is about 800 sessions, yesterday out of the blue sessions went down to about 200. Predictor on the serverfarm was leastconns, changed to round robin and one of the reals started taking about 400 sessions while the other stayed about about 80 or so. Took bad server out of service and good server started taking the full 800 sessions. We are unable to get that bad server to take sessions corrently. We have rebuild the server, move the server to a different network connection and tried to server hardware. Sessions still appear down. Also the few sessions that the bad server takes seem to work but the total sessions goes way down as if connections to bad server are just dropping. No config changes to the CSM or Cisco 7609 it is plugged into.

Gilles Dufour · ‎07-19-2006

do you have some form of stickyness ?

Also, with leastconn, there is a slowstart algorithm so that if a server goes down, it does not take all the sessions when it comes back up. This slowstart alogorithm forces the server to receive low number of connections initially to slowly increase it until it reaches the same level as the others.

If you have sitcky configured, a server going down could result in sticky entries pointing only to the other servers.

Gilles.

chris.super163 · ‎07-21-2006

I switched this over to roundrobin because of the slowstart on leastconns. Also we let it sit for quite a long time and the server will max out a very few sessions. We do not have sticky sessions enabled. This server farm has not been changed in months and just started having this issue out of the blue.

Sidenote that if we put the bad server inservice with roundrobin enabled we loss basically half of the connections.

Gilles Dufour · ‎07-22-2006

strange one.

I don't know if you want to continue troubleshooting.

If you do, you should capture a sniffer trace of the csm portchannel and then add the bad server and see what's happening.

Capture a 'sho tech' just before and just after the test as well as a 'sho mod csm x vserv' and 'sho mod csm x real'

Gilles.