However some don’t make it obvious at all! The easiest way I’ve found so far to determine if an image is multi-arch, is to use the experimental docker manifest command in the latest Docker versions. Some registries make this really easy, such as Docker Hub, where you’ll get a nice list of architectures on the Tags tab. If you’re using an existing public image, you will need to make sure that it is using a manifest list and supports both amd64 and arm64. You might think that this means we’ve reached multi-arch nirvana and everything will “just work” now, but unfortunately this is not the case. Previously you would have got the godforsaken “exec format error”. In newer container runtime versions (like Docker) this means if you do a docker pull nginx:alpine on your Raspberry Pi you’ll get an image for arm64 and on your Intel laptop you’ll get an amd64 image without any further effort. This allows you to specify a list of container images for a number of different architectures in a single “Manifest List”. This problem has gone away with Manifest Lists in the Image Manifest V2 spec.

This was both a pain for people building the images and for people trying to use them if you’re deploying a Helm Chart for example you would have to override the image tags if you wanted to run on ARM and there was no hope for you if you wanted to use mixture of arm64 and amd64 nodes in a cluster and have the same pods seamlessly schedule on either (without some extra tooling). Historically if you wanted to support multiple CPU architectures for your container image you would have to have two separate tags e.g. Let’s start with the container images, as that’s what you’ll need to tackle first if you want to start using arm64 in your environments. However the combination of all these events has caused a lot of the software to catch up and support for arm64 in Kubernetes distributions from Amazon EKS to k3s is now excellent! It is likely that a good proportion of the container images you’re using in production now support “multi-arch” and will run on both arm64 and amd64 machines. However now the Raspberry Pi 4 has a pretty decent quad-core CPU and up to 8GB RAM, Apple have blown away the industry with the M1 chips and AWS have launched Graviton2 instances which depending on who you ask have 20-40% better price/performance than the Intel equivalents.ĭespite all this, until recently I wasn’t convinced trying to use arm64 nodes in a production Kubernetes cluster was worth the effort. Until fairly recently, for most people, ARM CPUs were reserved for their phone or maybe a Raspberry Pi running their home DNS. The world of ARM processors has been getting very interesting over the last few years.

Kubernetes up and running