Open Stack

So in a running OpenStack there are several categories of config.

 - per user - e.g. ssh key registration with nova: we repeat this sort
of config everytime we add a user.
 - per node - e.g. nova.conf or ovs-vsctl add-br br-ex : settings that
apply individually to machines
 - application state - e.g. 'quantum net-create ...' : settings that
apply to the whole cluster not on a per-user / per-tenant basis

We have four places we can do configuration in TripleO:
 - image build time
 - in-instance heat-driven (ORC scripts)
 - in-instance first-boot scripts
 - from outside via APIs

At the moment we have a mismash around our demo scripts with some bits
being done here and there with no real consistency or ability to
predict 'for this sort of config, it is done <...>'

So, I'd like to come up with a predictive description that covers most
(ideally all) cases, so we can make it easy for folk to decide where
and how to tackle things.

As a strawman:
 - per user config should be done from the outside via APIs, even for
users like 'admin' that we know we'll have.
 - application state changes should be done from outside via APIs
 - per-node config should be done via ORC
 - first-boot scripts should only be used for things required to make
boot work for which cloud-init is either not available [seed node] or
not suitable [???].

Thoughts?

-Rob

-- 
Robert Collins <rbtcollins at hp.com>
Distinguished Technologist
HP Cloud Services