We had a situation where spawning 3000+ child jobs with perform_async inside a batch.jobs block caused all child jobs to “bunch up” and prevented other jobs from executing. The pseudo code looks like: