[openstack-dev] [Cinder] PTL Candidacy

Mike Perez thingee at gmail.com
Thu Apr 2 18:46:10 UTC 2015


Hello all,

I'm announcing my candidacy for Cinder PTL for the Liberty release.

I have contributed to block storage in OpenStack since Bexar back when things
were within nova-volume, before Cinder, and the honor of serving as PTL for
Cinder in the Kilo cycle.

I've spoke in the past about focused participation, something I still feel is
needed in the projects that are the basic foundation of OpenStack. Compute,
storage and networking need to be solid. My work as core in Cinder and
continuing as PTL has involved a lot of evangelizing and making new
contributors feel comfortable with becoming part of the team. As a project
grows, communication needs to be excellent, coordination is key to making sure
reviews don't stick around too long for contributors to feel discouraged.
I think the Cinder team has done an excellent job in managing this as we grow,
based on the feedback received. I really do think participation in Cinder
is getting better, and it's wonderful to be part of that.

If we take the Kilo-3 milestone for example, we landed 44 blueprints in
a single milestone [1]. That's huge progress. I would like to believe this
happens because of focus, and that happens because of better tracking of what
is a priority and clear communication. Lastly participation, not just core
folks, but any contributor that feels welcomed by the team and not to be burnt
out on never ending patch revisions.

Most of 2014 in Cinder was a lot of discussions on third party CI's. Third
party CI's are a great way for vendors to verify if a proposed upstream patch
would break their integration. In addition, it identifies if a vendor really
does work with the current state of the OpenStack project. There have been
plenty of cases that vendors discovered that their integration in OpenStack
really didn't work until they ran these tests. Last year, there was a real lack
of coordination and communication with vendors on getting them on board with
reporting third party CI results. In 2015 I took on the responsibility of being
the point of contact for the 70+ drivers in Cinder, emailing the mailing list,
countless reminders on IRC, contacting maintainers directly, and actually
making phone calls to companies if maintainers were not responsive by email.

I'm happy to report that majority of vendors have responded back and are active
in the Cinder community to ensure their integration is solid. Compare that to
last year when we just had one or two vendors reporting and majority of vendors
not having a clue! It's very exciting to help build a better experience for
their users using OpenStack. The communities pouring support to me on this
issue was hugely appreciated, and is what keeps me coming back to help.

We added 14 new drivers to Cinder in the Kilo release. Coordination was
beautiful thanks to clear communication and coordination with the hard working
reviewers in the Cinder team.

My priorities for Cinder in the Kilo release was to make progress on rolling
upgrades. I have spent a greater deal of my time testing the work to allow
Cinder services to not be dependent on database schemas. This is a big change,
and doesn't completely solve rolling upgrades in Cinder, but is a building
block needed to begin solving the other rolling upgrade problems. I'm really
happy with the work done by the team in the Kilo release and excited with how
comfortable I feel in terms of stability of the work thanks to the amount of
testing we've done.

This work however not only benefits Cinder, but is a general solution into
Oslo, in attempt to help other OpenStack projects in upgrades. Upgrades are
a huge problem that needs to be solved across OpenStack, and I'm proud of the
Cinder team for helping do their part to help drive adoption. Long term I see
this work contributing to an ideal single upgrade solution, so that operators
aren't having to learn how to upgrade 12 different services they may deploy.

My plans for Liberty is to work with the team on creating a better use of
milestones for appropriate changes. While we started some initial requirements
like making new drivers focus on the first milestone only, I think stability
time needs to be stretched a bit longer, and I think others will agree Kilo
didn't have a lot of this as planned for Kilo-3.

Cinder  will continue on efforts for rolling upgrades by now focusing on
compatibility across Cinder services with RPC. This is a very important piece
for making rolling upgrades complete. We will continue to work through projects
like Oslo to make sure these solutions general enough to benefit other
OpenStack projects, so as a whole, we will improve together.

Cinder volumes that end up in a stuck state. This has been a problem for ages,
and I have heard from countless people at the Ops Midcycle Meetup that
I attended. I'm happy to say, as reported from my take on the Ops Midcycle
meetup [2], that this was something the Cinder team discussed at the Cinder
Midcycle Meetup this year and we will be working on a solution that resolves
these issues with your preferred storage backend. The current solution is silly
dangerous database update and requires the operator to verify the real state
a volume is in. This is not a great a solution and totally error prone.
Instead, we will be working on improving this greatly by having Cinder
communicate to the storage solution the desired state for the volume to be in.
The solution will then either resolve the volume state internally and allow
Cinder to update its state for that volume, or report that it's actually in
a stuck state that not even it can resolve.

Lastly storage policies. Having the ability to use your preferred vendor for
the awesome features it provides. A lot of things aren't being exposed through
Cinder today easily. From my talks with small and big OpenStack
deployers, they want
more flexibility with this directly from the Cinder interface. I have spoke at
the Cinder Midcycle meetup on this and got great feedback. This started a spec
[3] which will greatly improve the visibility of what policies are possible
with your storage solution your using from OpenStack via Cinder client and
Horizon. This will greatly improve Operators knowledge in what's possible with
their OpenStack deployment and storage solution, and eliminates a great deal of
error prone issues. This is very exciting work coming from the team and I will
be contributing on driving this forward after getting further feedback from
operators at the next summit.

It would be an honor if the community would let me serve as the Cinder PTL for
Liberty release to finish out the planned improvements.


[1] - https://launchpad.net/cinder/+milestone/kilo-3
[2] - http://superuser.openstack.org/articles/takeaways-from-openstack-s-mid-cycle-ops-meetup-a-little-more-conversation-a-little-more-action
[3] - https://review.openstack.org/150511

--
Mike Perez



More information about the OpenStack-dev mailing list