When the Elastic Agent container is enrolling (which happens before run), then it cannot check the /liveness endpoint. This causes the container to not be marked healthy and be killed by kubernetes in the case that enrollment doesn't succeed in the time window set by the checks.