Proposal: Molecule to require minimal py36

[ssbarnea]

As py27 is effectively dead and I can no longer keep up the py27 jobs green (two different outages in last 24h), I propose we drop support for py27 in molecule.

The next major version v3.1 will no longer be tested or work with python older than 3.6.

Please be aware that this requirement is about molecule itself and that you can still test hosts using any version of python supported by ansible. Still, if you happen to test a py37 based host using delegated driver, you may be in trouble. You would have to spawn two hosts, one that is used by molecule and the tested one.

Please use thumbs up and down to vote on this change. Once I receive enough feedback, I will implement the change and unblock current PRs.

[geerlingguy]

I mostly run Molecule from local workstation (where I have Python 3.7), in Travis CI (which has Python 2 and Python 3 available... depends on the environment), and GitHub Actions (which defaults to Python 3).

There may be some speed bumps once this lands, but I approve of the general idea, since Python 2.7 is getting harder and harder to support unless you live in RHEL 6 or 7-land.

[nmusatti]

Wouldn't this make it unusable on RHEL/CentOS 7? The problem I see is that at least for the time being there is no Python 3 binding for libselinux on EL7, so you have to run Ansible on Python 2.7

[tadeboro]

@nmusatti This change will make it almost impossible to run Molecule on RHEL/CentOS 7. But, as @ssbarnea already said, you can still use Molecule to test your stuff on those systems.

[ssbarnea]

@nmusatti Yes and no at the same time. AFAIK, Centos 7.8 will include libselinux for python3. Also, this requirement is for the controller host running molecule itself. You can still test managed hosts that have any version of python, or even not having python at all.

Obviously that if your use case was "running with delegated driver on centos-7", you will pretty much have to pin down molecule or reconfigure the scenarios to avoid this.

We already have few jobs on OpenStack TripleO that are going to be impacted, but they will be gone soon, replaced by either centos-8 or using workarounds.

[sshnaidm]

While I'm for gradual removing py2 everywhere, I think it should be done within a reasonable time frame with a deprecation notice. At least people who install in CI latest version of molecule on CentOS-7 will have time to pin it to last 2.7 supporting version.

[ssbarnea]

@sshnaidm Users using semantic versioning ranges will not be affected. In fact we already do this in several places where we have something like molecule>=3.0.2,<3.1.

We are forced to do the drop ASAP because CI is currently broken and nothing else can merge. I tried fixing and few good hours later I realised that the battle to keep py27 alive is... futile.

[sshnaidm]

@ssbarnea for CI you need to use upper-constraints. Total removing of py2 support is big overhead for solving this problem.

Users using semantic versioning ranges will not be affected

Yes, and there are very few of such. Need to focus on users that WILL be affected, not to search which one wouldn't be.

[gundalow]

I just wanted to say I really appreciate that everybody took the time to give feedback on this.

THANK YOU ALL.