Currently we have a single network node, which means that it's a SPOF and a potential bottleneck. We should make all compute nodes be their own network node.
To do that and keep having a fragmented IP space, we need some kind of integration with the networking gear. Ryan has made some patches to integrate with exaBGP an use BGP. We should also align our plans with what OpenStack Quantum can do for us.
Version: unspecified
Severity: enhancement
This is quite an old task. Quantum appears to be an old name for Neutron, but of course we're not using Neutron right now (I think there's some other relevant tasks lying around about this somewhere)
We did add a second network node to our single openstack deployment with a custom failover procedure, but more verbose versions of this will be tracked elsewhere. Failover is currently documented at https://wikitech.wikimedia.org/wiki/Portal:Wikimedia_VPS/Admin/Troubleshooting#Fail-over