What would you like to be added: The Flow Control layer should dynamically configure its recognized priority bands based on the InferenceObjective resources registered to the pool. This will remove the current need for a static, at-startup configuration of the priority domain.
Why is this needed: Currently, only priority 0 is configured by default, and any other priority (including negative values for load shedding) must be hardcoded. This change will allow users to define and use any priority value in their InferenceObjective resources without needing to fork, modify, and redeploy the EPP. This is a significant usability improvement and makes the priority system far more flexible.
What would you like to be added: The Flow Control layer should dynamically configure its recognized priority bands based on the
InferenceObjectiveresources registered to the pool. This will remove the current need for a static, at-startup configuration of the priority domain.Why is this needed: Currently, only priority 0 is configured by default, and any other priority (including negative values for load shedding) must be hardcoded. This change will allow users to define and use any priority value in their
InferenceObjectiveresources without needing to fork, modify, and redeploy the EPP. This is a significant usability improvement and makes the priority system far more flexible.