Posts

A Compose file with a models: section runs an AI application out of the box on a laptop. Shipping the same stack to Kubernetes used to mean writing the model server’s Deployment, Service, ConfigMap, and PVC by hand, and remembering to point your application’s environment variables at the right place. Compose Bridge now does that for you, in two distinct topologies. This is the final post of the Compose Bridge Deep Dive series. Parts 1 and 2 covered the fundamentals and customization. This one focuses on a concrete, end-to-end scenario built on top of the new model-runner support in the default transformers. ...

The defaults shipped with Compose Bridge produce sensible Kubernetes manifests, but every organisation has its own rules: which labels are required, which ingress class to use, which securityContext is mandatory, which Pod Security Standard applies. Hard-coding all of that in every Compose file gets old fast. The clean way to bake those rules in is a custom transformer. This is the second post of the Compose Bridge Deep Dive series. Part 1 explained the transformer concept. This post shows how to fork the default templates, plug your rules in, and use the x-* extension fields to keep your Compose files clean. ...

Compose excels at local development. Production usually runs on Kubernetes. Maintaining two parallel descriptions of the same application is where drift, bugs, and “works on my laptop” stories come from. Compose Bridge bridges that gap by turning a Compose file into a deployable Kubernetes artifact, without asking you to maintain a second source of truth. This is the first part of a three-post deep dive on Compose Bridge. Part 1 covers what it is and how it works. Part 2 goes into custom transformers and enterprise customization. Part 3 shows how to deploy an AI application with Docker Model Runner to Kubernetes. ...

Linux isolates containers using kernel namespaces. Sometimes you need the opposite: two containers that can see each other’s processes or share memory. The pid and ipc directives give you that escape hatch. Sharing a PID namespace pid: service:<name> lets a container see and act on processes inside another service: services: app: image: myapp debugger: image: alpine pid: service:app cap_add: - SYS_PTRACE command: sleep infinity The debugger container’s ps, strace, and /proc all reflect app’s processes. Combined with cap_add: SYS_PTRACE, you can attach strace or gdb to a running production-style container without baking debug tools into its image. ...

docker compose up --wait (Tip #51) waits for services to become healthy. docker compose wait does something different: it waits for services to exit, and returns their exit code. Basic usage docker compose wait <service> The command blocks until the specified service stops, then prints the exit code. If you echo $? after, it’s the same value. This is perfect for one-shot services: migrations, batch jobs, test runners, anything that runs and exits. ...