# Adapt A Workload Image Use this guide when turning a Den starter container into a workload image for a specific service, lab job, or internal tool. ## Choose A Support Level Start by classifying the workload: - `runnable today`: application container built from `den:dev`. - `prototype-ready`: VM appliance or lab gateway that integrates workload packages into the Den system generation. - `blueprint`: kiosk, POS, GPU, HPC scheduler, or provider cloud image work that needs platform support Den does not have yet. Do not present a blueprint workload as supported until the missing platform pieces are implemented and smoke-tested. ## Start From The Den Base Image Build the local Den OCI base image: ```bash ct run oci-image ``` Use it as the first line in the application Dockerfile: ```Dockerfile FROM den:dev ``` Create application environments under `/opt/app` so the Den runtime under `/opt/den` stays separate from the workload. ## Install Application Dependencies Use conda inside the image build: ```Dockerfile RUN conda create \ --yes \ --prefix /opt/app \ --override-channels \ --channel conda-forge \ python=3.12 \ && \ conda clean --all --force-pkgs-dirs --yes ``` Put `/opt/app/bin` first: ```Dockerfile ENV PATH=/opt/app/bin:/opt/den/bin:/run/current-system/sw/bin:/bin:/usr/bin:/sbin:/usr/sbin ``` ## Build And Run ```bash docker build --platform linux/arm64 -t :local docker run --rm :local ``` Use `linux/arm64` for a `linux-aarch64` Den base image and `linux/amd64` for a `linux-64` Den base image. For a network service: ```bash docker run --rm -p 8080:8080 :local ``` For a batch job that writes output: ```bash mkdir -p build/example-output docker run --rm \ -v "$PWD/build/example-output:/output" \ :local ``` ## When To Use A Den Package Instead Use a Den package in `den-forge/recipes/` when the workload must be part of the booted system itself: - it needs a systemd unit in the Den VM, - it owns rootfs files or symlinks, - it must participate in Den generations and rollback, - it should be published as part of a Den release train. That path is an OS integration task, not the default application-container starter path. ## Verify The Image Use the fastest check that covers the changed surface: | Change | Smoke | | --- | --- | | Den base image changed | `ct run oci-smoke` | | application Dockerfile changed | `docker build --platform ...` | | HTTP service changed | `curl http://127.0.0.1:/healthz` | | batch job changed | `docker run --rm :local` | | booted VM integration changed | `ct run boot-systemd-smoke` | If the workload must run inside a booted Den VM through Docker or Podman, treat that as a future container-host feature. It needs a runtime package, kernel and cgroup validation, and a dedicated smoke test before the docs should present it as supported.