DEV Community

Cover image for What if your Kubernetes applications didn’t ship their own runtimes?
Bruno Borges
Bruno Borges

Posted on

What if your Kubernetes applications didn’t ship their own runtimes?

When a Python, Node.js, JVM, or .NET runtime needs a security update, teams often rebuild application images to pick up the new runtime, even when the application itself hasn’t changed a bit. What if the runtime were managed as infrastructure instead?

That’s the idea behind Brewlet, an open-source project a few folks are building at Microsoft. It lets developers publish application code as OCI artifacts, while platform teams approve and manage language runtimes and dependencies on Kubernetes nodes.

A typical flow looks like this:

  1. A developer publishes an application artifact with explicit launch metadata.
  2. The platform team approves a digest-pinned runtime for a node pool.
  3. A regular Kubernetes Deployment selects that runtime.
  4. A containerd shim starts the application using the approved runtime.

When the runtime needs a security patch update, the platform team can rotate it without rebuilding the application artifact. Running processes aren’t hot-patched, however: they need new task instances to use the updated runtime. Application dependencies still need their own updates, too.

Why not Kubernetes image volumes? They’re a good native alternative and can also separate runtime and application images. Brewlet adds integrated runtime inventory, node-pool provisioning, and authorization checks when a container starts, including automatic restarts. The trade-off is a privileged node integration and a custom containerd shim. If you are interested in the comparison, here’s a summary.

Brewlet is pre-GA; please evaluate it in a disposable environment, not a production cluster.

I’d love feedback from platform engineers: would centrally governed runtimes be valuable enough to justify a custom node integration in your environment?

Top comments (0)