The jnlp container has currently a special status: it is automatically injected in the pod template definition if not manually declared.

For new users, it is not obvious that it should be declared in a pod template in order to customize the agent configuration.

Also, once you start defining the jnlp container, you have to provide the full configuration for it, including the entrypoint and command to run.

There has been some system properties to customize the agent image, the default resources, etc. but these are ad-hoc fixes and do not cover the agent configuration to its full extent.

The goal of this issue is to capture what would make sense to improve usability of this configuration.

• Would it make sense to define the agent configuration at the cloud level for all pod templates?
• Provide agent configuration per pod template?
• Define fields to override per pod template?