Here is my report on the 2018 OpenStack Summit in Berlin.
Day1 - Keynotes
Besides the many new developments in the OpenStack projects, my take away from the first keynotes is that the summit is being renamed Open Infrastructure to better represent the OpenStack Foundation's focus areas. Nick Barcet announced the next OSP version 14 as well as a new edge use case design named Virtual Central Office (VCO) which looks very exciting.
After the keynotes, I spent most of my time on the Forum floor. This is where projects' communities gathered to discuss various topics around an etherpad in the same fashion as the old design summit.
Cross-project Open API 3.0 support
At the API session we discussed improvements to the api-ref documentation schema to be directly used by clients and servers. The improvements are a challenge because the openstacksdk has complex logics baked in to support different micro versions. Moreover, we want to be able to test the new schema definitions and perhaps be able to generate clients' and servers' code from the definitions. I mentioned the GraphQL implementation we have been working on with Gilles Dubreuil since Vancouver, but this is a much bigger project that was out of scope for this discussion. I'm personally very interested in being able to use a more rigorous schema for the restfuzz project testing purpose. I'm looking forward new developments in this area.
Reusable Zuul Job Configurations
At the end of this first day I was too jetlagged to attend the Red Hat party and went straight to my hotel to get a good night sleep. :)
Day 2 - Keynotes
At the second day's keynotes, Monty Taylor presented Zuul's project update and I was pleasantly surprised that two of the four new features were based on Software Factory contributions: the React web interface and the Kubernetes driver. Then Tobias Henkel gave a great presentation on the CI needs of BMW automotive software development and how they leverage Zuul to accelerate and ensure a high quality standard. Besides the compute resource scaling, for Tobias, the key feature of Zuul is the project gating system which enables scaling development team.
After the keynote I went back to the Forum floor and had some interesting walk-alley discussions:
We found a solution to improve multi-tenant labels by designing an 'allowed-labels' option in the Zuul tenant configuration, and I went ahead and proposed an implementation. This enabled further discussion about another solution on the Nodepool side.
As a follow-up I asked to merge the Nodepool code in Zuul to simplify this kind of new feature and reduce code duplication. However, the Zuul core team had a compeling argument for CI setup and how both projects are effectively tested differently. In the end, we agreed that merging the Zookeeper modules into a new nodepoollib repository would be a good compromise.
We also discussed how it can be difficult to debug complex issues such as NODE_FAILURE errors which requires lots of hop. Tobias suggested that we could attach a transaction-id to log message for tracing all the events related to a single connection trigger event. Incidentally, Matthieu Huin has been working on a very similar feature in Cauth recently.
Finally, we discussed how to improve Zuul's metric interface to enable native Prometheus backend. Different implementations have been proposed and the next step is to collect all the existing metrics and find the best fit to refactor the code.
Then I gave a talk about log-classify and attended three more forum sessions:
Reduce your log noise using machine learning
The presentation went very well, more people than expected attended the talk. We got some great feedback and it seems like many operators do not have efficient log analysis in place. One of the key requests is to support streaming logs. We need a good internal interface to properly modelize such streams. Another interesting feedback was about GDPR compliance and that log-classify models is a good approach since the vectors are anonymised by design.
A marketplace for sharing Zuul jobs and roles
The BMW team presented a new service to provide a search interface and jobs and roles usage metrics. This lead to a discussion about further Zuul REST API improvements to better support such a use-case, and eventually integrate the service directly into Zuul.
OpenDev feedback and missing features
The former openstack-infra team discussed the OpenDev upcoming rename and how review.openstack.org namespaces are going to be moved to the new system. Source code replication to external systems is going to be improved so that it can be managed on a per-project basis.
Zuul - Project Onboarding
In the Zuul Project Onboarding, James Blair presented the new Zuul QuickStart and how it can be used to quickly setup a test environment using docker-compose.
RDO/Ceph community event
At the end of the second day, I went to the RDO/Ceph community event. It had a more relaxed atmosphere than the main venue and it was great to meet the community in this setting with good music and German beer. It was also a good opportunity to meet with the new community manager Rain Leander who took over Rich Bowen's role recently.
I didn't attend the third day of the summit because it was already time for me to go home. Once again, we have made great progress and I'm looking forward to further developments. Thanks you all for the great summit.