We are able to paralelize up to 60 jobs, but we are taking too long to identify when they finish, preventing new jobs start.
We could parallelize the verification of finished jobs using threads.

The change can increase the memory and cpu consume in the worker-sql-jobs. Considering it’s already above the limit (in production and sandbox) we might need to increase the requirements in Infra. It’s important to keeping only one worker for this end. Celery documentation does not advice to have replicas for processes that run scheduled tasks.