Kubermatic Kubernetes Platform (KKP) has been pushed out in v2.16, seeing the open source project for managing Kubernetes cluster automation gain sought-after features such as Open Policy Agent (OPA) and ARM support.

OPA, an engine for enforcing policies across the cloud native stack, graduated the Cloud Native Computing Foundation’s maturing process just last week. While in-depth knowledge of the project isn’t exactly widely spread right now, giving it a closer look might become a necessity soon: after all, pod security policies will be deprecated in Kubernetes 1.21 and something will be needed to fill that gap. 

KKP users can ascertain whether OPA will be that something for them via the Kubermatic API for now, though it should become fully integrated into the UI with one of the upcoming patch releases. Another thing to try is the newly added functionality to deploy and manage ARM-based clusters, which is a necessity when looking to support edge scenarios, something that is recently seeing a lot of interest.

It surely won’t be the last thing the Kubermatic does with that aspect, given its past involvement in some collaborative reference projects as well as the recent hiring of Sascha Haase to fill the position of VP Edge. “General cloud scenarios are already covered rather well by the Kubermatic platform, now it’s just about expanding the scope to constrained devices” Haase said on a call with DevClass. First results of these efforts can be expected later this year.

With Kubermatic Kubernetes Platform 2.16, the Kubermatic team also took the first steps towards putting the platform on the map of organisations that need to set up and maintain machine learning infrastructure. According to Haase, Kubernetes lends itself to such scenarios, since it already is one of the most prominent tools to manage workloads. However “data scientists don’t really want to interact with the infrastructure, but use it for crunching numbers,” as Haase puts it.

For that reason, projects such as MLflow and Kubeflow have started to gain traction and do indeed a decent job – once they’re set up, that is. Rolling them out across clusters still needs a bit of fiddling to say the least, which is why the 2.16 release of KKP includes a tech preview for what Haase describes as a machine learning platform. 

The latter uses KubeCarrier along with Google’s Kubeflow and various other open source tools to provide a central management instance for spawning “number crunching clusters” as needed.

To keep things secure, KKP 2.16 contains new Preset Management functionalities admins can use to configure environments so that others have an easier time setting up dev or pre-production clusters. Additional control mechanisms include functionality to dynamically configure which data centers to use and keep a lid on infrastructure consumption.

Under the hood Kubermatic Kubernetes Platform has learned to work with Kubernetes 1.20, while support for v1.16 has been removed. Teams using it with CoreOS clusters might want to consider migrating those, since the project reached its end of life in May 2020 and isn’t supported in KKP anymore, starting with the current release.