OpenStack Havana on CentOS, network bugs and workarounds

Standing up OpenStack Havana on CentOS with packstack gets most of the way there on its own. The remaining part was two networking issues that only showed up after the install, both tied to Open vSwitch and a stale default service.

My install and fix notes from that run, as a gist:

Two bugs in there worth calling out on their own.

packstack generates an ifcfg-br-ex that shouldn't exist according to the docs, but does, due to a packaging bug at the time. The fix was putting the old eth0 address on the new br-ex bridge rather than picking a fresh one, which is what keeps the network restart from cutting the SSH session it's running in.

The second is RabbitMQ failing to start with BOOT FAILED {could_not_start_tcp_listener,{"::",5672}}. CentOS 6 runs Matahari by default, and it also listens on port 5672, the same port RabbitMQ needs. Stopping qpidd clears it; if Matahari isn't needed for anything else, removing it and its related packages outright is better than just disabling the service, so the port conflict can't come back after an update re-enables it.

Both bugs are the same shape of problem: a default from the base OS quietly conflicting with something OpenStack expects to own. Worth checking for on any new platform before assuming the failure is in OpenStack itself.