Leboncoin at KubeCon EU 18
Intro
If you are into the containers hype, then you definitely know that last week KubeCon Eu 18 took place from the 2nd to the 4th of May at the Bella Center in Copenhagen, Denmark.
We were not able to submit a talk this year since we are still in the process of deploying k8s in our infrastructure (keep following our blog and we’ll share with you what we did when the time has come ;). Nevertheless, three of our engineers attended this conference and here is their feedback.
First, a big thumbs up to the CNCF members and to the whole staff for the organization. Everything was smooth, most of the talks were excellent and you could feel that the expertise in this containers world has greatly evolved and reached high levels.
Keynotes summary
Day1
The first day of the conference started with an update about the project and the community. They were quite a few numbers about the growth between nowadays and last year’s edition of KubeCon. Reading these charts, we can state without any doubt that 2017/2018 was the year of the container. At the end of the session, the CNCF presented their new interactive portal of the CNCF landscape: https://l.cncf.io/
The second keynote focused on a few new projects that joined the CNCF (20 in total) or gained greater attention as they matured. There were 3 main categories of projects:
- Sandboxed
- Incubatored
- Graduated
New Sandboxed projects:
- Rook for storage
- OpenPolicyAgent, SPIFEE & SPIRE for security
Projects that reached Incubator:
- CoreDNS: Reached Beta since k8s 1.10
- LinkerD and Envoy gained substantial attention these last couple of months
- Prometheus, OpenTracing and Jaeger
- Nats : A new project in the queue ecosystem, between RabbitMQ and Kafka, even if this project is already more than 10 years old :)
- FluentD for log collection
- And for the others: CNI, GRPC, ContainerD, TUF, Notary, Vitess
Graduated Projects:
- k8s
Then, the first “real world” guest went on stage: CERN who presented their take on k8s, how they use it, what they did to make it useful in their context of heavy data science and processing a huge amount of data.
Day2
The starting keynote of day2 was the famous one from Google Cloud Platform.
Among the various announcements they made, we can list:
- gVisor: sandboxed containers
- Support for more and more different kinds of workloads (Batch, Cron, GPUs, StatefullSets, …)
- The rise of the Operator framework from CoreOS, or should we say RedHat now ?
- A demo of Skaffold to ease the development with k8s.
The third keynote was a journey feedback from TheFinancialTimes. It was really interesting, especially the feedback they made about the mentality and open-minded mindset needed for a successful transition.
One of the last keynotes for the day was by Spotify. They presented their feedback about OpenSource, community, involvement, organizations … (I have to admit that I have been confused by the presentation and did not manage to bind everything together as it should be on that one).
Day3
Day 3 was really focused on Machine Learning and we got a 1st introduction to Kubeflow, with a full demo.
Then a talk from Sahil Dua, Software Developer at Booking.com, who presented their workflow of deep learning for booking previsions and suggestions. He especially explained how the models are trained, recycled and deployed till production.
In the end, we had the chance to see an AMAZING talk by Simon Wardley about the evolution of IT and how to make choices that matter. This guy is really a rock star !
Sessions summary
SRE point of view
From an SRE point of view, they were a few very valuable talks on k8s usage “at scale” or feedbacks on “how you would transition to k8s”:
- Continuously deliver your k8s infrastructure
- The enterprise’s new Shoes
- Managing k8s at day 2
- Multi cluster ingress and K8s cluster registry
- Cluster as cattle
- 101 ways to break and recover your cluster
Among the other various topics around, we witnessed a real interest in security.
- One very good talk about it was from Matt Rickard and its presentation “Building Docker images without Docker”. This was one of our 1st introduction to gVisor and the topic to Rootless containers :)
- GVisor got a bigger attention during the talks “Best Practice for Container Security at Scale”.
The second hotter topic was mesh routing. They were a lot of talks speaking about Istio, Envoy, LinkerD etc;
If you look at the schedules:
- 15 refs to Istio
- 12 refs to mesh routing
- 9 refs to Envoy
- 5 refs to LinkerD
These projects look very interesting and shine with an impressive list of features.
But, the best part comes: In the context of 1 app, deployed on Kube, you have everything you care about, logically “tight” together. The integration level is the key. With the Operators, CustomResources, custom controllers and so on, you can really build these so called “cloud native” apps, and even better: in a composable way.
Monitoring and metrics were not at rest either. Prometheus and its integration with the rework of the AutoScaler manager was also very popular.
In the end, I have felt the “wow” effect on the presentation of Kris Nova about Kubeflow and all the tools that Heptio released this last year (Ksonnet, Ark, Sonoboy, Contour, …)
For people running in the cloud, we have got an update on AWS EKS but sadly still no GA date announcement… You will find more details in the notes links below and the slides.
Conclusion
KubeCon Eu 18 was a really good event that gathered more than 4000 peoples.
More over the keynotes and the talks, it gave us the opportunity to meet other IT professionals who try to solve the same kind of problems than us. We were also able to meet the core community thanks to the “Meeth the maintainer” sessions and other SIG’s users. They were “deep dive”, “hacking sessions” and workshops without forgetting the incredible Thursday night event at the Tivoli park.
Definitively a good conference :)
