RFE: rule for injected ansible var usage (e.g. not using ansible_facts)

[mwhahaha]

Summary

In the investigation of an ansible performance problem related to facts (ansible/ansible#73654) , it was found that having INJECT_FACTS_AS_VARS set to true (which is the default) can lead to a performance impact when processing result. The larger the amount of host facts, the longer task result processing takes. It would be beneficial to warn if someone is relying on these facts. Since it's dynamic, it might be difficult to determine which vars, but there are common ones like ansible_distribution* or ansible_{hostname,fqdn,nodename,os}

Additionally it might be beneficial to warn about an unfiltered setup module call as pulling in all the facts can have unintended performance issues.

Related changes: https://review.opendev.org/q/owner:aschultz%2540redhat.com+topic:ansible-facts

Issue Type

• Feature Request

STEPS TO FIND SIMILAR VARS

I found our usage of these things by looking in our yaml files with something like:

grep ansible_ * -r | egrep -v "_(facts|become|user|host|limit|check_mode|ssh|connection|managed|verbose|job|home|log|loop|async|limit|filter|version|_inventory|\.py|group_vars)" | grep ansible_

Desired Behaviour

Warn when things like ansible_hostname, ansible_fqdn, ansible_distribution, etc are used.

[myllynen]

ansible-core upstream is changing the default so that facts will not be "injected" as vars in the future, see:

ansible/ansible@931c923

It would really helpful if ansible-lint count help detecting these cases. While one could alternatively set "inject_facts_as_vars = False" in ansible.cfg that requires running the code and then the code wouldn't work past the first instance so having ansible-lint reporting all occasions alongside with any other possible findings would make things a lot easier. Thanks.