OpenStack Summit Boston 2017
A Week in Review
The OpenStack Summit. The largest gathering of OpenStack operators and developers in one city, under one roof. I won’t bore you with a retelling of my perspective of the keynotes. I encourage you to watch them, though, as they are very enlightening. This is my week in review. *Law & Order sound effect*
Monday. Kicking off the Summit. All the fresh faces with various logos from past Summits, other developer conferences. Some are reconnecting, others are just meeting for the first time. Some flew in the day before, others just that morning. The theme for the Boston Summit is definitely Kubernetes, with Edge Computing emerging as a concept. Though there was some emphasis on interoperability, the underlying tones were very Kubernetes-centric. In the afternoon, a lively discussion took place regarding the future of configuration management (ie. /etc/nova/nova.conf, not necessarily a certain framework), with etcd emerging as the dominant contender for tool to add to the stack.
Being ill prepared, the first day took a major toll on me, and I retired to my hotel after a team dinner, after learning that Kubernetes had some serious warts when using it atop OpenStack. Fortunately, a Walgreens was immediately in my path, so I ducked in for a quick pit stop before making it to bed.
Tuesday. Provisions were acquired to kickstart (heh) the day.
After streaming the keynotes and hydrating, I made my way to the convention center to listen to a talk about Tess Master1 and its place in an OpenStack-based cloud, which, again, had a very Kubernetes-heavy slant. It was clear the community has coalesced around K8s as the container-du-jour, and that is the future of container stories at many companies today. The faults seemed to be around the shortcomings of K8s, but the writing on the wall is that K8s is The Way. If your container story doesn’t include K8s, you’re doing something wrong. An enlightening talk was given over the performance of OpenContrail vRouter by offloading it to a programmable NIC, or SmartNIC.2
The evening was ended with StackCity making a return, this time at the
historic Fenway Park. They brought out the total Fenway experience, including
the metal detectors and long lines. The only thing missing was an actual game.
Many selfies and group photos were had. Glorious.
We’re past the halfway mark, and people are starting to look haggard. Some unlucky souls got roped into work outages, and have taken over every available power outlet. Paddy Power Betfair demonstrated how OpenStack can be immutable by containerizing the control plane. Next, the power of oslo.messaging was demonstrated so as to keep RabbitMQ as stable as possible, with 0MQ emerging as a broker backend.3 Immediately after, a talk on scaling a 22 node environment to 100 nodes and beyond was cancelled when the speakers did not show, after everyone waited 20 minutes. Sad times for all.
As unfortunate as that was, the next presentation I attended was about scaling
vanilla OpenStack to 5000 nodes and beyond.4 Georgy, from Mirantis, highlighted
on the smaller scales, but the focus was definitely on the upper end of the
spectrum. MySQL and RabbitMQ were again highlighted as the services to protect
the most, with varying opinions on which techniques to use and when not to use
them. I noticed a panel being put on about being a PTL5, so I sat in to see how
my own experiences as a PTL stacked up. Turns out, my experiences were pretty
typical, exacerbated by the decimation of the team a year ago. Lastly, I
attended a talk which was put on by an old friend, regarding orchestrating the
deployment of Hadoop on OpenStack using Heat templates.6
The evening ended again at Fenway, this time with the OpenContrail User Group, with a smaller dinner. Again, classic Fenway, baby. Beer, pizza, burgers, Fenway Franks, Cracker Jacks, you name it. A call to action took place to reboot the open source community, to make OpenContrail the best open source SDN available. Presentations were given by folks from several organizations, outlining their use cases of OpenContrail.
Last day! Holy shit, it’s already Thursday! It’s been one hell of a ride! After making it to the convention center, I attended a fishbowl on high availability of OpenStack where people with way more advanced clouds than I were leading the discussion, including the pain points, which was pretty tooling agnostic. This was followed up by another fishbowl about achieving resiliency at 1000+ nodes, which was mainly led by one person in the room who had done so. It shows the level of scale of the average operator.
After lunch, I attended a panel on growing the next generation of OpenStack contributors, where two interns highlighted their experiences of their internship and interaction with the greater community7. Following up was another panel, outlining how interoperability has been, and will be in the future8. Lastly, I attended one final panel about the COA (Certified OpenStack Administrator) exam and why I should take it (not that I wasn’t looking to take it anyway :) 9
In summary, these are some of the things I took away from the Summit:
- Kubernetes10 is11 The12 Path13 Forward.14
- If you’re not containerizing your control plane, consider it. Why aren’t you using Kubernetes?
- Chef OpenStack is still kicking, even if nobody is talking about it15 (check out page 43!) Holy shit, the #4 deployment choice! He’s not dead, Jim!
- The COA will test your skill, knowledge and ability to figure shit out on the fly. Your mettle will matter here more than rote memorization. It is not something you can memorize, and the older your prod clouds, the harder it will be. Beef up on DevStack master, or another framework (cough openstack-chef-repo16 cough).
- Look at Kubernetes some more. Expect it to be a major component in The Future.
And that’s a wrap. Good knowledge, good food and many steps were had with
friends old and new alike. People I have only known as bits on a screen were
revealed as being actual people, and it was glorious.
Bonus cat picture! Meet the new addition to the clowder, Odin: