Actually, my current approach looks like this:

carla-server-task : Launch carla server instance on a random port, set the port as param and then block the task/process, so I can kill carla when this task is aborted. This task keeps running the whole time.
start-carla-task : Launch a carla-server-task and wait for the port parameter to be set. Set the launched carla-server-task task-id and the port as param. Set task completed.
main-task : Run experiment when all start-carla-task are finished with carla-servers running on the ports specified in the start-carla-task params.
stop-carla-task : Get the task-id from the start-carla-task and programmtically abort the carla-server-task

Since i want to be able to start multiple carla-servers my pipeline looks like this:

                    pipeline
        /                              \

start-carla-task -1 start-carla-task -2 -----> These manually start carla-server-task-1 and carla-server-task-2 and block until they got a port back.
|\ /|
| main-task |
|/ |
stop-carla-task -1 stop-carla-task -2
\ /
finished

(can you see my diagram?)