By default, we decided that 2-4 hours was a decent amount of time to wait before shutting down a user's services. Perhaps this limit could be higher if we find that there is less resource contention than we expect to see (8-12 hours might support a full work day without allowing users to eat up resources indefinitely). If we have time before the beta release, we have decided to enforce this as an explicit system-wide timeout.
In any case, we will likely need the ability to specify (likely as an environment variable) the timeout for user services. Similar to the "Token Timeout" parameter that the API server currently accepts, this would simply need to be an integer number of hours to keep services alive.
The API server would then need to implement logic to periodically check the runtime for each controller in each user's namespace. Any controllers whose Uptime exceeds the integer specified in the YAML will be shut down.
This ticket is complete when the API server accepts and respects a system-wide "timeout" for user services.
- Apply per account