Hey team, I had a question about executing the pipelines using clearml agent setup in K8s. When we define a pipeline script to be executing remotely and submit it to the queue for execution we see following things happening

• User runs the pipeline script with stepA and stepB locally where it creates the pipeline and the controller task with task uuid say uuidA and submits it to the queue.
• ClearML also captures the pipeline arguments through some inspection (we use click mostly) and submits it along with the pipeline.
• The clearml agent running in k8s pod picks up the submitted pipeline script from the services queue and starts a worker pod which runs this same controller task uuid uuidA and executes the pipeline script remotely.
• It starts the worker pods that executes the individual two steps in the pipeline and these have their own task and task uuid i.e uuidStepA and uuidStepBSo I guess the question is, why does ClearML need to run the pipeline script twice i.e once when we submit it locally from our machines and once remotely in K8s? is there any specific reason for this design?