Open Stack

Guys,

Currently, at the time of the first boot Mcollective on slaves is
configured by Cloud-init (host, port etc.) and by Nailgun agent when node
identity is written. This happens in any order and already caused several
problem which were fixed in bounds of
https://bugs.launchpad.net/fuel/+bug/1455489. The last fix relies on
hostname to figure that the provisioning is over and prevent Nailgun agent
from messing with Mcollective. This solution is quite hacky and, I think,
we need to come up with a better fix. So, the options are:

1. Do nothing
  + No effort, no regression
  - Existing solution is fragile and not simple enough to be sure that
there are obviously no more bugs.
  - Cloud-init does things on early boot stages which is a bit hard to
debug.

2. "Manual" Mcollective configuration, triggered by Nailgun agent.
  + Quite simple
  - Not "cross-distro"
  - Not consistent with the remaining configuration tasks (rewriting them
worsens the problem above)

3. Use Puppet to configure Mcollective. (copy necessary input data into the
chroot env and run puppet inside it)
  + Using puppet is consistent with the way Fuel configures things
  - It is not consistent with remaining configuration tasks, which is done
via Cloud-init and moving them to Puppet takes some time.

4. Some other tool?

New options, cons and pros are very welcome!

Best,
Georgy
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20160812/f0d44ec6/attachment.html>