5:54 a.m.
Hola rdoers,
The plan is to GA RDO Liberty today (woot!), so I wanted to send out a
status update for the RDO Manager installer. I would also like to gather
feedback on how other community participants feel about that status as
it relates to RDO Manager participating in the GA. That feedback can
come as replies to this thread, or even better there is a packaging
meeting on #rdo at 1500 UTC today and we can discuss it further then.
tldr;
RDO Manager installs with 3 controllers, 1 compute, and 1 ceph on
virtual hardware have been verified to work with GA bits, however bare
metal installs have not yet been verified.
I would like to start with some historical context here, as it seems we
have picked up quite a few new active community members recently (again
woot!). When RDO Kilo GA'd, RDO Manager was barely capable of a
successful end to end demo with a single controller and single compute
node, and only by using a special delorean server pulling bits from a
special github organization (rdo-management). We were able to get it
consistently deploying **virtual** HA w/ ceph in CI by the middle of the
Liberty upstream cycle. Then, due largely to the fact that there was
nobody being paid to work full time on RDO Manager, and the people who
were contributing in more or less "extra" time were getting swamped with
releasing RHEL OSP 7, CI on the Kilo bits became mostly red with brief
24 hour periods where someone would spend a weekend fixing things only
to have it break again early the following week.
There have been many improvements in the recent weeks to this sad state
of affairs. Firstly, we have upstreamed almost everything from the
rdo-management github org directly into openstack projects. Secondly,
there is a single source for delorean packages for both core openstack
packages and the tripleo and ironic packages that make up RDO Manager.
These two things may seem a bit trivial to a newcomer to the project,
but they are actually fixes for the biggest cause of the RDO Manager
Kilo CI breaking. I think with those two fixes (plus some work on
upstream tripleo CI) we have set ourselves up to make steady forward
progress rather than spending all our time troubleshooting complete
breakages. (Although this is still openstack so complete breakages will
still happen from time to time :p)
Another very easy to overlook improvement over where we were at Kilo GA,
is that we actually have all RDO Manager packages (minus a couple EPEL
dep stragglers[1]) in the official RDO GA repo. When RDO Kilo GA'd, we
did not even have everything officially packaged, rather only in our
special delorean instance.
All this leads to my opinion that RDO Manager should participate in the
RDO GA. I am unconvinced that bare metal installs can not be made to
work with some extra documentation or configuration changes. However,
even if that is not the case, we are in a drastically better place than
we were at the beginning of the Kilo cycle.
That said, this is a community, and I would like to hear how other
community participants both from RDO in general and RDO Manager
specifically feel about this. Ideally, if someone thinks the RDO Manager
release should be blocked, there should be a BZ with the blocker flag
proposed so that there is actionable criteria to unblock the release.
Thanks for all your hard work to get to this point, and lets keep it
rolling.
-trown
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1273541