RFC: Ansible >= 12 integration following deprecation of strategy plugins

[moreati]

Ansible 12 (ansible-core 2.19) will deprecate third-party strategy plugins. Strategy plugins are currently the hook that Mitogen hangs all its Ansible integration off. Mitogen will need to adapt/redesign its approach.

This issue is venue for discussion, proposals, etc.

refs #1258, #1272

[moreati]

As current caretaker of Mitogen I agree with the deprecation and accept it. I won't be advocating that the Ansible project alter this decision, and I encourage others to do the same. Strategy plugins are intimately linked to the internals of Ansible, there's not a clean interface.

It 'is Mitogen's responsibility to find a better way to integrate, ideally with much looser coupling and without reaching so deeply into Ansible's private implementation details.

[elek]

For the record, the motivation of Ansible project can be read here:

The current strategy plugin interface is deeply tied to the internals of ansible-core. This makes it very difficult to implement a properly functioning plugin, and even more difficult to avoid breaking plugins when changes are made in ansible-core.

The ability to provide custom strategy plugins should be deprecated, and later removed. The existing strategy plugins in ansible-core will remain. In the future a new mechanism may be provided for customizing playbook execution in a manner similar to strategy plugins, but there are no immediate plans for doing so.

ansible/ansible#84725

[dnmvisser]

Do I understand correctly that this could be the end of the ansible_mitogen strategy plugin?

[moreati]

Do I understand correctly that this could be the end of the ansible_mitogen strategy plugin?

It is the end for this mechanism. Mitogen will need to find other ways to integrate with Ansible.

[pioio]

Mitogen provides obvious speedups. I'd be sad to see it go.
I can imagine that incorporating entire mitogen to ansible core would be challenging.
At the same time, my understanding is that mitogen uses variety of techniques to achieve.
I suspect that 20% of those techniques provide 80% of speedups seen by most people. E.g. executing tasks on a local host to configure a laptop -> 3x speedup with mitogen!

would it be at least reasonable to make those 20% part of a new (or existing) ansible core strategy?

[gilesw]

The speedups are incredible over 50% in my test. I wouldn't want to run ansible without it.

[Nihlus]

How much of what Mitogen does today could we reasonably port directly to Ansible? That might be the best bang for our buck, so to speak.