This article is more than one year old. Older articles may contain outdated content. Check that the information in the page has not become incorrect since its publication.

Dockershim: The Historical Context

Author: Kat Cosgrove

Dockershim has been removed as of Kubernetes v1.24, and this is a positive move for the project. However, context is important for fully understanding something, be it socially or in software development, and this deserves a more in-depth review. Alongside the dockershim removal in Kubernetes v1.24, we’ve seen some confusion (sometimes at a panic level) and dissatisfaction with this decision in the community, largely due to a lack of context around this removal. The decision to deprecate and eventually remove dockershim from Kubernetes was not made quickly or lightly. Still, it’s been in the works for so long that many of today’s users are newer than that decision, and certainly newer than the choices that led to the dockershim being necessary in the first place.

So what is the dockershim, and why is it going away?

In the early days of Kubernetes, we only supported one container runtime. That runtime was Docker Engine. Back then, there weren’t really a lot of other options out there and Docker was the dominant tool for working with containers, so this was not a controversial choice. Eventually, we started adding more container runtimes, like rkt and hypernetes, and it became clear that Kubernetes users want a choice of runtimes working best for them. So Kubernetes needed a way to allow cluster operators the flexibility to use whatever runtime they choose.

The Container Runtime Interface (CRI) was released to allow that flexibility. The introduction of CRI was great for the project and users alike, but it did introduce a problem: Docker Engine’s use as a container runtime predates CRI, and Docker Engine is not CRI-compatible. To solve this issue, a small software shim (dockershim) was introduced as part of the kubelet component specifically to fill in the gaps between Docker Engine and CRI, allowing cluster operators to continue using Docker Engine as their container runtime largely uninterrupted.

However, this little software shim was never intended to be a permanent solution. Over the course of years, its existence has introduced a lot of unnecessary complexity to the kubelet itself. Some integrations are inconsistently implemented for Docker because of this shim, resulting in an increased burden on maintainers, and maintaining vendor-specific code is not in line with our open source philosophy. To reduce this maintenance burden and move towards a more collaborative community in support of open standards, KEP-2221 was introduced, proposing the removal of the dockershim. With the release of Kubernetes v1.20, the deprecation was official.

We didn’t do a great job communicating this, and unfortunately, the deprecation announcement led to some panic within the community. Confusion around what this meant for Docker as a company, if container images built by Docker would still run, and what Docker Engine actually is led to a conflagration on social media. This was our fault; we should have more clearly communicated what was happening and why at the time. To combat this, we released a blog and accompanying FAQ to allay the community’s fears and correct some misconceptions about what Docker is and how containers work within Kubernetes. As a result of the community’s concerns, Docker and Mirantis jointly agreed to continue supporting the dockershim code in the form of cri-dockerd, allowing you to continue using Docker Engine as your container runtime if need be. For the interest of users who want to try other runtimes, like containerd or cri-o, migration documentation was written.

We later surveyed the community and discovered that there are still many users with questions and concerns. In response, Kubernetes maintainers and the CNCF committed to addressing these concerns by extending documentation and other programs. In fact, this blog post is a part of this program. With so many end users successfully migrated to other runtimes, and improved documentation, we believe that everyone has a paved way to migration now.

Docker is not going away, either as a tool or as a company. It’s an important part of the cloud native community and the history of the Kubernetes project. We wouldn’t be where we are without them. That said, removing dockershim from kubelet is ultimately good for the community, the ecosystem, the project, and open source at large. This is an opportunity for all of us to come together to support open standards, and we’re glad to be doing so with the help of Docker and the community.