Halting a workflow should purge its messages from RabbitMQ so that existing and new Sisyphus workers won't retry any of its tasks until the user resumes the workflow.
If a worker acknowledges a purged message, that needs to react gracefully.
This fix might have to wait until each workflow has its own queue.
RabbitMQ is not part of the user model so users should not need to know how to administer it to recover from problems.
Users should be able to halt a workflow, delete any stuck workers (as in #24), then fix a bug and start new workers without them also getting stuck.
RabbitMQ is not part of the user model so users should not need to know how to administer it to recover from problems.
Users should be able to halt a workflow, delete any stuck workers (as in #24), then fix a bug and start new workers without them also getting stuck.