It was great to be back in Amsterdam for DevOps Days. Having attended as a StackStorm founder and then a couple of times now as one of the founders of OpenEBS and CEO of MayaData, it feels a little bit like home (although there were many new faces).
In this blog I will talk a little bit about:
- Our first workshop! What did we learn from the packed house?
- Who attended DevOps Days? — some observations
- A few observations about the 12 months since the last DevOps Days Amsterdam
TL;DR -> production usage of stateful workloads on Kubernetes does seem to require or at least lead to new approaches for many users, as explained below.
When I walked into the workshop space 15 minutes before the start, I saw Jeffry and Murat, pictured above, alone in an empty room; I immediately wondered whether it was best to start at 9 am in Amsterdam. Within 20 minutes the room was full.
The workshop was designed by Jeffry and Murat to focus on a common need -> blue/green deployments on Kubernetes of stateful workloads. You can get all the materials from the workshop here:
I won’t go into depth about the workshop. I would just point out that unlike many other Kubernetes workshops I’ve attended, this one spent almost no time on basic installation and deployment of Kubernetes and whatever tools that were being discussed. Instead, users were expected to have minikube or access to another Kubernetes cluster; they then simply:
And they went in 2 minutes or less into more interesting subjects such as snapshots and clones and how to use them to adopt blue/green and canary patterns to stateful workloads.
A couple of data points regarding who attended:
- We had financials, governments, start-ups, DevOps consultants and born on the web type companies like Booking.com in the room. So the first point is that at least amongst attendees — admittedly a self-selected group — there was a wide variety in the types of companies for which they worked. This means that not only is Kubernetes widely used, which we already knew, but stateful workloads are increasingly moving with Kubernetes as well.
- There were not *any* storage administrators in the room (or at least no one admitted to being one :)). This is a crucial point that companies and projects that are trying to serve stateful workloads on Kubernetes too often disregard.
The last 12 months have passed quickly; it was 12 months ago that I started full time at MayaData and our CTO Jeffry Molanus started as well. I’d like to thank the OpenEBS community and members of the broader community such as Chris at the CNCF and Alexis at WeaveWorks for their coaching and feedback. A few things that have happened just since last year that I hope rewards the belief of the community and of our team in our approach of delivering storage services to Kubernetes by leveraging Kubernetes itself:
- A broad and growing community. If you are interested, you can track our statistics at devstats.openEBS.io — if you take a look quickly you’ll see that there are now many more contributions and contributors that some of the CNCF projects.
- Hackathons and Bangalore. Over the last 12 months, members of our team have launched the OpenSource Cafe — a floor of the MayaData building plus organizational support and lots of food and chai as needed — to support the broader OpenSource community. And the OpenSource Cafe has become one of the leading spots for tech and DevOps meetups in Bangalore. Also, our hackathons have reached across India — and into the broader open source community — and have helped to educate hundreds of engineers on how to contribute to open source projects. For example:
3. Contributing upstream to Kubernetes. I am maybe most proud of our efforts to contribute to projects that we are not leading. While OpenEBS and Litmus are each driven by our engineers, the storage components of Kubernetes are of course led by engineers from Google with help from others. So we view it as great validation that, for example, our design for improved management of local media for storage forms the basis of the Node Disk Manager subcomponent that will aid many use cases.
Hopefully the above is of use to others in the OpenEBS community and the broader Kubernetes community. There is a tremendous amount of momentum behind enabling Kubernetes to run stateful workloads in a cloud-native way — allowing DevOps engineers that understand Kubernetes to control their storage via an architecture that traditional centralized storage cannot match. We’re happy to both be contributing to this momentum and to be benefiting from all the work Google and others are doing in the area as well.
This article was first published on Jul 19, 2018 on MayaData's Medium Account