[OpenStack-Infra] Dublin PTG recap

Clark Boylan cboylan at sapwetik.org
Wed Mar 7 13:54:38 UTC 2018


I've made it home after the PTG and am jet lagged which means it is a great time to try and recap some of what happened at the PTG. I'm not going to go into a ton of detail for each topic as I think that would make this turn into a novel but if people are interested in specific items feel free to start new threads for them and we can dig in more there (this has already happened for at least one or two topics). Also you can refer back to the PTG etherpad, https://etherpad.openstack.org/p/infra-rocky-ptg, for more information and notes that were taken at the time.

First thing that everyone should know is that the weather did not cooperate with us and resulted in much travel uncertainty and disruption. Mention this so that we aren't surprised if some people are slow to respond or otherwise AFK while finding their way home. It also meant that by Thursday we had largely abandoned our PTG schedule. Apologies if this meant that a topic you were interested in was not covered.

For this PTG we ended up having three major themes. There was the cross project (helproom) time, zuul topics, and infra services conversations. I think we managed to do a reasonable job covering these themes despite the weather and illness.

During the cross project conversations we were able to spend time helping projects like neutron, ironic, swift, and glance better take advantage of Zuul v3. We covered how the gate works, how to convert jobs to native Zuul v3, cleaning up old unused jobs, and how to run multinode jobs. We also had a conversation with keystone about how they might run performance testing on infra. There was also quite a bit of time spent working with the QA team to work out how multinode devstack would work in Zuul v3, how to make irrelevant files lists more intuitive, and how to wind down grenade testing on our oldest supported branch.

Kashyapc brought up nested virt support again which is becoming more important with post meltdown/spectre performance slowdowns. This resulted in a few of us crashing the public cloud working group's room for a discussion on how to have better communication between devs and clouds. Rough plan there is to spin up a new neutral mailing list to spark conversations over tools like nested virt between all the involved parties.

We covered a number of Zuul topics ranging from scaling out the scheduler to dashboard improvements to tenant label restrictions and executor affinity. None of these topics seemed controversial and I think we captured good notes on the etherpad for these topics in particular. I will call out the tenant label restrictions and executor affinity features as items that seemed to come up quite a bit from various users for various use cases so I think the importance of these features may have gone up after the PTG. One Zuul topic that we couldn't quite get into due to illness and last minute travel changes was Zuul support for containers. As I understand it there was some pre PTG talk about this, but we should probably try to have a proper discussion on the mailing list once people are back to normal operating hours.

For infra services we talked about upgrading Gerrit, better multi arch support, and rolling out Bionic. The rough plan (details on etherpad) for the Gerrit upgrade is to update the operating system for Java 8 this cycle then early next cycle upgrade Gerrit to 2.14 or 2.15 depending on which process is simpler for us (we will have to test that between now and then). Multi arch support is now something we have to think about with the arm64 Linaro cloud roll out. We seem to have largely decided that things are moving and largely working and we'll tackle problems as they come up. We also have Bionic beta images available. The one restriction we'd like to see there is to not gate on them but projects can (and in some cases should) go ahead and start using them to determine future compatibility particularly with python3.6.

One last major item that came up was how the OpenStack Foundation's CI/CD focus area will affect the infra team. This was a topic at the board meeting in Dublin which I had to miss due to helping run the helproom, but there were discussions with members of the infra team later in the week. The most concrete result of that seemed to be a shared understanding of three facets to this focus area: 1) The how to and best practices of doing CI/CD properly and effectively 2) Zuul and related software as a set of tools to enable (1) and 3) the current set of services run by the infra team which may be useful to developers outside of OpenStack. We've been promised a proper thread of its own from the foundation to start this conversation more broadly with the infra team soon so keep an eye out for that.

I've probably forgotten/skipped/missed other topics that are important and worth calling out. It was a long week and I'm now jetlagged so you have my apologies. Feel free to respond to this thread or make a new one to call things out or fill in details. It was great seeing everyone, hope you all made it home ok, and I'm excited we now have this large list of TODOs to get done over the next cycle :)

Thanks,
Clark



More information about the OpenStack-Infra mailing list