The CMD / Entry Point tab on a custom container’s configuration screen controls what process is launched when the container starts. The two fields exposed by the platform map one-to-one onto the two Docker primitives — ENTRYPOINT and CMD — and let you override what was baked into the image without rebuilding it.
The two fields
- Entry Point — the executable that runs when the container starts. Overrides the image’s
ENTRYPOINT. If the image declares an entry point and you leave this field blank, the image’s value is used. - Run Command — context-dependent:
- When no entry point is configured (in this field or in the image), the run command is treated as the command to execute on container start.
- When an entry point is configured, the run command is passed as the argument list to the entry point. This matches Docker’s behaviour of appending
CMDtoENTRYPOINT.
Each field shows a Reset to default arrow icon once you have entered a custom value. Clicking it reverts the field to whatever the image’s manifest declares.
When you need to set these fields
The platform does not introspect images to detect a missing run command. If neither the image nor these fields supply a command, the container starts and exits immediately with nothing running — the layer ends up in a restart loop. Two situations where you need to set the fields explicitly:
- Distroless or minimal base images that ship without a default
CMD. Set the executable path in Entry Point. - Multi-purpose images that you want to start with a non-default command — for example, a single image that contains both an API server and a worker, started by different arguments.
For images that already declare both ENTRYPOINT and CMD in their Dockerfile (the typical pattern for Docker Hub stack images), you can leave both fields empty.
Applying the change
A change to entry point or run command requires a container restart. After saving, click Apply on the layer or the topology wizard. The platform restarts the affected containers in a coordinated rollout — for a single-instance layer this is a short downtime; for a multi-instance layer the rollout is sequential.