I have two azure VMs (IaaS) running Ubuntu 12.04, with a load-balanced web-service on ports 80 and 443. I followed the instructions from this document: http://www.windowsazure.com/en-us/manage/windows/common-tasks/how-to-load-balance-virtual-machines/#lbprobes
For the record, both load balancer probes are configured to hit a custom endpoint /status on port 80, which is configured to return 200 on success or 500/502 on error. (I have a python service running behind nginx as a proxy, so if I take down the python service, nginx will return a 502 response immediately in response to /status.)
The load balancer works fine when both servers are up, but things break down when I take down one server. I can see load balance probes come in every 15 seconds, and they all get a 502 response code from nginx, but Azure still directs traffic to the broken instance for about 5 minutes. That's much longer than the advertised time of 31 seconds, and totally unacceptable for a production environment.
Could anyone please tell me how to get the advertised behavior from the load balancer? I really don't want to have to ditch this platform-based load balancing and rolling my own with haproxy or the like..
P.S. There was a very similar question asked here, but without any response besides "open a support ticket": http://social.msdn.microsoft.com/Forums/windowsazure/en-US/afcdfb97-a752-4de4-8c06-99cc70155282/load-balancer-does-not-take-out-of-rotation-dead-server
I hope that I don't require paid support in order to utilize an Azure feature as advertised. Any help appreciated!