When calling compute_parallel the full script is relaunched as a subprocess for each worker, which results in recursive processes that eventually die because they're being spawned too quickly.