RFE: Network Isolation Should Allow Non-Standard Networks on Nodes

[Dan Sneddon]

Currently TripleO network isolation assumes that certain networks will be on certain roles. These networks can be individually enabled/disabled, but it isn't possible to arbitrarily assign networks to roles. The Ceph OSD nodes, for instance, only have ports configured on the Control Plane, Storage, Storage Management, and Management networks. If one wanted to place a port from the Internal API network on the Ceph node, it would require customizing the Ceph Heat templates.

We should not be so prescriptive about which networks are on which nodes. If we just add all networks to all nodes, they will ordinarily be noop'd, and can be enabled for custom environments.

This work will mainly need to take place in the role templates for the overcloud nodes, so compute.yaml, ceph-storage.yaml, etc. and the Puppet equivalent role files. Since these networks would only be enabled in a custom environment, no changes to the existing NIC config templates are needed.

[Steven Hardy (shardy)]

This doesn't appear to have been started, and it has no assignee, so I'm bumping to n2

[Dan Sneddon (dsneddon)]

Note, this was 90% implemented in Mitaka. In fact, the work in Newton involves changes that could be considered bugfixes (improvements in routing, an additional VIP for the Management network).

[Dan Sneddon (dsneddon)]

Here is the main patch which implemented this change: https://review.openstack.org/#/c/249533/

Also, this was backported to stable/liberty: https://review.openstack.org/#/c/268833

[Steven Hardy (shardy)]

Dan, can you break down what remains before we can close this bug please? I'm going to defer to n-3 because atm it's not clear what work remains.

[Dan Sneddon (dsneddon)]

The only work remaining on this bug is the patch to make some corrections to IP routing:
https://review.openstack.org/#/c/318389/