If you use --start-at-task with ansible-playbook and config-download, the task (or a subsequent one) usually fails pretty quickly because gather_facts has been skipped.
We gather_facts once for all hosts as one of the first tasks, and subsequent tasks are heavily dependent on those facts having been gathered. If you then use --start-at-task, there's a good chance you skip gather_facts and your tasks all fail.
Whether or not we gather_facts on each task ought to be configurable with a variable that can be overridden from the cli. It should default to False to preserve the current behavior.
We should also use a fact cache, so that if we do skip gather_facts, things at least have a chance of working. The fact cache should live in the working dir of config-download, which is where the playbooks/tasks have been downloaded (under /var/lib/mistral/... when using the workflow).
Related fix proposed to branch: master /review. openstack. org/548651
Review: https:/