Happy to discuss possible solutions. The main issue is that there are a lot of cases where "something" could happen that interrupts a playbook run and keeps a system unfinished. IMO Ansible is designed to just execute the playbook again to bring the system into the desired state, especially in the described case this would only be a temp issue, e.g. network/target unreachable and a re-run will successfully download the artifact and bring the system back into a desired state.

I agree that there are lot of cases for failures and for sure not all of them can be covered in Ansible due to the way Ansible works. So there is AFAIK no "rollback" which would be needed to cover all of them. Nevertheless I would like to at least cover the low hanging fruits, which means that the system should not start if the "gathering" phase already contains errors. So everything which is part of this phase should happen before the scripts actually make a change.

For the current issue I could offer to send a PR with the solution I had in mind.

Sure, happy to review.

See #68