Make systemd service state less slow

[k00mi]

As has been noted in #10710, execution of a systemd service state is quite slow. The suggested solution of using systemctl is-enabled only helps for the enable operation. Instead, these commits address the expensive function _untracked_custom_unit_found(), which is used in several other places.

Previously our highstate runs took about 45s to several minutes due to a bug in systemd/logind which leaks unit files. With these patches we are down to about 36s in all cases.

[cachedout]

@k00mi Thanks for this. I am concerned that this may regress the fix that was made here: #11921

Thoughts?

[cachedout]

@k00mi Additionally, it looks like this is causing a test to fail so if we were to merge this, we would need to have that fixed. Could you please take a look? Thanks.

[k00mi]

@cachedout Indeed parametrized units were not handled correctly. I think I fixed it now by asking systemctl list-unit-files if the unit was not found in list-units and added a comment explaining why both are necessary. systemd-sysv-generator takes care of legacy units in /etc/init.d, so they appear as regular units in list-units.

The CentOS VMs don't even bootstrap properly and those tests that actually reach the systemd module fail with KeyError: 'cmd.run_stdout'. I have zero clue why this happens. Other functions in this module use that function, as did the previous version and I can't reproduce said error on any of our systems.

Looking at the definition of test_available(), I think the issue is with the test itself. It mocks get_all() (which would later call cmd.run_stdout), but instead of calling that I use cmd.run_stdout directly. I don't know enough about your test system to fix this, can someone more familiar take a look?

[cachedout]

@k00mi That all sounds great. Thank you!

@jfindlay Any chance you could jump in here and help with the test errors?

[jfindlay]

@k00mi, I have updated the test on your branch, k00mi#1