Juju on MAAS agent tools upgrade mechanism

[Peter Grandi]

[ ... ]

> On your question about whether Juju on MAAS or private clouds
> is abandonware - definitely NOT. We have a whole lab set up
> which tests MAAS and Juju working together on a daily basis.

But consider my perspective: with my colleagues I got to
maintain a setup that uses "industry standards" and is tiny:

  * ULTS 14 with MAAS, Juju, OpenStack,
  * 12 identical nodes with around 70-80 LXCs and VMs on them
    (plus Ceph daemons), pretty much static, with perhaps 1-2
    VMs created/deleted every 2-3 days,
  * using LTS packages or "stable" versions (with the Juju
    1.23.3 "bugfix" version) for the same LTS release.

This has given quite a few problem: OpenStack huge internal
database inconsistencies, plus bizarre interactions with Juju
leading to changing address, Juju upgrade partially failed, and
difficulties getting new nodes acquired from MAAS, which itself
has had some problems acquiring new "metal", plus a failed
upgrade too.

Since there are probably MAAS/Juju/OpenStack deployments with
dozens of thousands of nodes working happily on AWS perhaps our
simple, static, case is not the current focus of maintenance and
development. The first time I looked at Juju I got indeed the
impression that AWS was the main focus:
  http://www.sabi.co.uk/blog/13-one.html#130319

Or perhaps distributed toolsets like that are just a bit too
fragile and try to do things that are a bit too hard or have too
many possible setups to lack occasional surprises... :-/