Fix CI on CentOS

[smortex]

Pull Request (PR) description

We are experiencing CI failures on CentOS because the service cannot start because it cannot create it's PID file.

Running the very same test locally works as expected, so the root cause of the different behavior on the CI is unfortunately unknown. Yet, ensuring the directory for the PID file exist seems to fix the CI issue.

This Pull Request (PR) fixes the following issues

n/a

[smortex]

Cannot reproduce locally with:

BEAKER_PUPPET_COLLECTION=puppet6 BEAKER_setfile='centos7-64{hostname=centos7-64.example.com,image=centos:7.6.1810}' bundle exec rake beaker

[ekohl]

An alternative is to only do this in CI. https://github.com/voxpupuli/voxpupuli-acceptance/#per-host-configuration describes that you can create spec/setup_acceptance_node.pp. Perhaps that's a better place for a workaround?

Also, it only appears to show up on CentOS 7 (at least now). Perhaps scope it to that while you're at it?

[oranenj]

I don't think this is at all correct; I just looked at a CentOS 7 system using chrony, and /var/run is a symlink to /run and its contents get lost on boot, so managing files in there is not correct. Something else seems to be the problem.

[oranenj]

chrony seems to create the directory by itself when it's started; it starts as root, so unless the test container is restricting writes to /run somehow I don't see why it would fail.

[alexjfisher]

I don't think this is at all correct; I just looked at a CentOS 7 system using chrony, and /var/run is a symlink to /run and its contents get lost on boot, so managing files in there is not correct. Something else seems to be the problem.

indeed, it's tmpfs. I think @ekohl 's suggestion is the best to only apply the workaround in CI.

[smortex]

💯

I added commits to test this and stumbled on voxpupuli/voxpupuli-acceptance#27. When fixed, I can remove the last commit and merge all other commits in a single one.

[alexjfisher]

I've just released https://rubygems.org/gems/voxpupuli-acceptance/versions/1.0.1