Open Stack

Hi everyone,

Last month I started a thread to discuss our rhythm and the possibility
of switching to one-year development cycles, starting with Rocky:

http://lists.openstack.org/pipermail/openstack-dev/2017-December/125473.html

The thread quickly exploded in various directions, objections and
tangents, which I tried to summarize here:

http://lists.openstack.org/pipermail/openstack-dev/2017-December/125688.html

At the core, I think the thread exposed a fundamental tension between
upstream and downstream OpenStack on that topic. On the upstream side,
we automated most of the process so the cost of releasing is limited.
The cycle boilerplate activities are a reasonable cost to pay to get
releases out often enough to get tight feedback loops and maintain pace
and quality. The downstream cost, however, is still significant:
marketing the new release, packaging it, upgrading to it or organizing
events around it is not getting simpler and in some cases resources are
getting more limited.

So while the concerns (especially downstream) are real, the proposed
solution is not a straight and consensual answer to them. We need to
look at alternative solutions to to reduce release downstream cost. We
need to have a wider look at how OpenStack is (or should be) consumed,
and discuss cycle length in relation to other efforts in this area
(fast-forward upgrades, past-EOL maintenance, support for OpenStack
deployments on mixed versions of components...). We can't really have
those complex discussions and the TC make any change in time for the
Rocky cycle, which will start in a couple of weeks.

Those discussions will happen in Dublin (PTG) and continue in Vancouver
(Forum at Summit). In the mean time, Rocky will be a 6-month development
cycle, as proposed in https://review.openstack.org/#/c/528772/

As a sidenote, Jay pointed us to a similar discussion[1] around
releasing less often, which is currently happening in the Kubernetes
community. While some of the context is different (especially their
releases "upstream" cost is still pretty high), a lot of their concerns
overlap with ours, so it makes an interesting complement read:

[1]
https://groups.google.com/forum/#!msg/kubernetes-dev/nvEMOYKF8Kk/n3Rjd2bMCAAJ

Looking forward to discussing this topic more with interested people in
Dublin.

-- 
Thierry Carrez (ttx)