On 03/07/2015 04:34 AM, Dan Prince wrote:
On Tue, 2015-03-03 at 17:30 -0500, James Slagle wrote:
Don't let the subject throw you off :)
I wasn't sure how to phrase what I wanted to capture in this mail, and
that seemed reasonable enough. I wanted to kick off a discussion about
what gaps people think are missing from TripleO before we can meet the
goal of realistically being able to use TripleO in production.
The things in my mind are:
Upgrades - I believe the community is trending away from the image
based upgrade rebuild process. The ongoing Puppet integration work is
integrated with Heat's SoftwareConfig/SoftwareDeployment features and
is package driven. There is still work to be done, especially around
supporting rollbacks, but I think this could be part of the answer to
how the upgrade problem gets solved.
+1 Using packages solves some problems very nicely. We haven't solved
all the CI related issues around using packages with upstream but it is
getting better. I mention this because it would be nice to have CI
testing on the upgrade process automated at some point...
HA - We have an implementation of HA in tripleo-image-elements today.
However, the Puppet codepath leaves that mostly unused. The Puppet
modules however do support HA. Is that the answer here as well?
In general most of the puppet modules support the required HA bits. We
are still working to integrate some of the final pieces here but in
general I expect this to proceed quickly.
going back to CI, I think this would benefit from an additional CI job
given we have a non-voting HA job running on precise/elements, I'd like
to add one running on fedora/puppet, maybe initially non-voting as well
this said, it would also be nice to have a job which deploys additional
block storage (cinder) and object storage (swift) nodes ...
... and to save some resources, maybe we can switch
'check-tripleo-ironic-overcloud-precise-nonha' to deploy a single
compute node instead of two
CLI - We have devtest. I'm not sure if anyone would argue that should
be used in production. It could be...but I don't think that was it's
original goal and it shows. The downstreams of TripleO that I'm aware
of each ended up more of less having their own CLI tooling. Obviously
I'm only very familiar with one of the downstreams, but in some
instances I believe parts of devtest were reused, and other times not.
That begs the question, do we need a well represented unified CLI in
TripleO? We have a pretty good story about using Nova/Ironic/Heat
to deploy OpenStack, and devtest is one such implementation of that
story. Perhaps we need something more production oriented.
I think this is an interesting idea and perhaps has some merit. I'd like
to see some specific examples showing how the unified CLI might make
things easier for end users...
I am of the same feeling; AFAIK devtest was meant to setup a development
environment, not a production environment, more on this later
Baremetal management - To what extent should TripleO venture into this
space? I'm thinking things like discovery/introspection, ready state,
and role assignment. Ironic is growing features to expose things like
RAID management via vendor passthrough API's. Should TripleO take a
role in exercising those API's? It's something that could be built
into the flow of the unified CLI if we were to end up going that
Bootstrapping - The undercloud needs to be
bootstrapped/deployed/installed itself. We have the seed vm to do
that. I've also worked on an implementation to install an undercloud
via an installation script assuming the base OS is already installed.
Are these the only 2 options we should consider, or are there other
ideas that will integrate better into existing infrastructure?
And also should we think about possibly renaming these? I find that many
times when talking about TripleO to someone new they find the whole
undercloud/overcloud thing confusing. Calling the undercloud the
baremetal cloud makes it click.
I don't think we need more; what I would like to have instead is a tool,
targeted at end users, capable of setting up an undercloud without going
through the seed
this said, I am really not sure if that should be a wrapper around
devtest --no-undercloud or a tool which turns the existing base os into
an undercloud; both seem to have pros and cons
Release Cadence with wider OpenStack - I'd love to be able to say on
the day that a new release of OpenStack goes live that you can use
TripleO to deploy that release in production...and here's how you'd do
personally, while I tried to join this conversation in the past, I am
still unsure whether for tripleo a stable/master approach would work
better or not than a synchronized release cadence
GPG KEY: 08D733BA
OpenStack Development Mailing List