Bdeo at KubeCon + CloudNativeCon Europe 2022

30 June 2022 · By Sergio Fernández

A month ago, Valencia hosted the event that all CloudNative and Kubernetes lovers in Europe have been waiting for over a year. If you just stumbled across this post, may I explain that Kubernetes is a de facto standard: the following link describes in detail its growth in recent years and how industry adoption has been exponential.

AI, Serverless and Nodeless

The second week of May saw several simultaneous events including IA Kubernetes Day and KnativeCon, along with others that I will discuss later. For me, the highlights of these events were:

🧠 A couple of presentations where Kubeflow was integrated with KServe for different use cases such as cancer detection. All in all, it seems that Kubeflow is the way to go for AI in K8s.

✈️ KNative has finally taken lambda scaling seriously at the function level. They presented a demo confirming that, for future releases, it will be integrated into the core. Nowadays many developers use KNative for Pod-level scaling, where it demonstrates its strength and stability, and for lambdas they use OpenFaas. This is an important step to address this kind of functionality.

☁️ The Nodeless theme was repeated throughout the different talks. In particular, we talked about how virtual-kubelet helps us to abstract from the cloud to use, for example, AWS Fargate without worrying about auto-scaling groups.

Another event that took place this day was the Kubernetes on EDGE Day. In this event we could see how Kubernetes fits very well for different use cases where having the cluster close to the end customer is crucial. An example of this use case was presented by the people from New Relic placing inference models on the Edge to comply with regional data protection policies and reduce access time to the models.

One technology that has been gaining momentum in recent years is e-bpf, which makes monitoring through sidecars a thing of the past. For those of you who are familiar with it, it allows any program launched by a user to be executed in isolation within the operating system kernel itself. To do this, this function creates a specific event, which has a system call, entry and exit functions and trace points.

There are many technologies that are making ebpf their workhorse in recent years. One of them is Cilium, with its mesh cluster that allows clusters to be interconnected as if they were one at a functional level, or Falco for security issues.

🍿 Other lectures of interest

Several lectures caught my attention the rest of the days. The most interesting were:

KEDA, an Event Driven autoscaler that is very well built and whose talk reached full capacity due to the speculation generated.
The state of the art of NATS, another one of those projects that has stood out in recent years, where new features such as key-value storage and Object storage were presented.
Clossplane as a tool for provisioning and managing infra resources.
There were also several on troubleshooting with Linkerd, or confidential containers and how they can be used to obfuscate sensitive information, this at a low level.

🚘 As use cases, it is worth highlighting the presentation by the people from Mercedes-Benz, where they presented their journey through the 7 years they have been using K8s, how they process orders in Wolf and use ML with Kubernetes, etc.

📺Below is a list containing the videos of the most relevant conferences that have been published in my opinion. Many of the others that I have mentioned above, as they are paid events, I doubt that they will be published (you will have to take my word for it ). The rest will be here.