I attended the Neutron meetings for the OpenInfra PTG in Shanghai last week. I was not in Shanghai, so I participated entirely remotely over BlueJeans.

Remote Participation

Typically I would work most of a day - 5-6 hours with a nap in the middle - and then be on the PTG from 3-5 hours in the evening. The timeshift was such that the scheduled block of meetings started at 8:00pm my time and ended at 3:30am. BJ worked really well with a couple of momentary hiccups that were quickly rectified. The audio quality was pretty good and I believe I spent part of the time on the screen next to the Etherpad, so I felt very included. I did not attend cross-project sessions as those tend to be very noisy, with a lot of crosstalk.

Deep and sincere thanks to Slawek for making sure that remote participation was possible. I was the only one who took advantage of it in the Neutron discussions, but it was available for anyone; it will be interesting to see if more people try it as time goes on.

Wednesday

The first session of the PTG was a retrospective on the Train cycle. I think we did a very good self-examination here, talking about the good points, the bad (CI issues), and checking that we have the right processes and meetings. I felt it was very productive.

The second session was the beginning of a presentation from Liu Yulong, core reviewer from China Telecom, on the enhancements to OVS that he has been working on. It sounds like they have been hacking on the OVS agent as they scale up to meet the issues they are facing, and they wanted to share those and get community approval. This discussion continued into Thursday and Friday; during this time on Wednesday we talked about the idea of killing the DHCP agent and using OVS flows like OVN does, which is problematic for baremetal use cases.

After this were cross-project meetings (Kuryr, Cyborg) so I dropped off.

Thursday

We started with a discussion that I had added to the agenda, which is to start the process of planning how we will deprecate in-tree ML2 plugins, specifically Linux Bridge, as we add OVN. We decided to reach out to the large installations still using Linux Bridge like GoDaddy, and unless that raises some big surprises we will mark it ‘deprecated’ so that it is clear new feature work will not need t be done for LB.

Then we talked about IPSec encryption for tenant network tunnelling, which sounds interesting but needs a spec.

The next topic was how to handle the metadata service in IPv6-only environments. The metadata service works by using an 169.254.x.x IP address. We decided to talk to the cloud-init people, and then to determine an IP address in the IPv6 space that could act as a comparable equivalent. I suggested we get it ratified by IANA as well.

After the team photo there was a long discussion about OVN. First the OVN squad did an overview of OVN, which was very well received. There were a number of discussion points from the community afterwords: - What about migration scripts for ML2/linuxbridge to OVN? - OVN code in-tree or out? - WHat about Smart NICs for OVS offload?

After this there was a short discussion about IPv6 and OVN devstack tempest test configuration that seemed pretty straightforward. This was followed by meetings with the Edge SIG and Nova, which I skipped.

I was really interested in the following topic, which was the proposal from the Keystone team to refresh how policies are handled into different scoped roles. There seemed to be support for this, and mlavalle will work on it. I am excited to have the “admin anywhere is admin everywhere” problem squarely in the rear-view mirror.

The really excellent profiling work mlavalle has done with Neutron was covered next. Long term we agreed this work [1] belongs in Oslo. And we finished the day by continuing Liu Yulong’s talk about OVS agent enhancements, this time talking about stopping traffic flooding on br-int.

Friday

We kicked off Friday with another topic I suggested, which is was reevaluating the Stadium projects. I had put together some quick metrics on stadium commits over the past 4 cycles [2]. Sadly that google spreadsheet was not available, so I put the numbers in an ethercalc [3] so it would be viewable behind the Great Firewall. We categorized projects thusly:

  • Healthy
    • networking-ovn
    • ovsdbapp
    • networking-odl
  • Dead: neutron-interconnection
  • “Puppy” (will be euthanized unless it is adopted and cared for):
    • neutron-bagpipe
    • neutron-bgpvpn
    • networking-sfc (pending check with tacker team)
    • neutron-dynamic-routing
    • neutron-fwaas and neutron-fwaas-dashboard
    • neutron-vpnaas and neutron-vpnaas-dashboard
  • On probation pending CI issue resolution: networking-midonet

Projects considered dead will be ejected from the stadium and moved to the x/ namespace in opendev.org.

I also checked on Tap-as-a-Service, which had applied for stadium membership. The general response was that interest in TaaS in the stadium had fallen off, and in the current environment it seems the stadium is closed for new applicants.

Next we discussed floating IPs on routed networks, which would be a nice completion for routed networks but which there is not an agreed-upon approach yet. After this followed a continuation of Liu Yulong’s topics. We started with having the OVS agent do port health checks, but this got pushback as being in the province of Monasca/Skydive/Vitrage. Then there were additional discussions using an easel that I could not easily see, so I listened in as much as I could to the discussion.

After lunch, we discussed our plethora of Zuul jobs. We decided to remove all our current experimental jobs, which are broken and legacy. We’ll make uwsgi the default, finally closing a gap. And there are a laundry list of other small job changes and adjustments; see the etherpad for mre details.

I had added a topic for stateless security groups, a feature that Nuage wanted to add but had not gotten traction upstream. It seemed to me that the reaction was completely positive for this, and Nuage is going to revive their upstream work. I look forward to seeing this brought to fruition.

To close out the PTG we reviewed 24 stagnant, old specs and closed almost all of them. We discussed the community goals for the U cycle, which we are already in good position for. And finally we talked about neutron-lib, which is a difficult topic and there was not a volunteer to fill Boden’s role as champion.

Summary

Congratulations to Slawek for putting together a really good agenda that spanned a lot of topics and interactions with other teams and guiding us through them on time. It’s a tough task to get this all organized, and Slawek you did great, especially since it’s your first time. Great job!

I am left feeling really good about the vitality of the Neutron community, and while I think we may shed some of the more hype-curve stadium projects the core is healthy and strong.