Improve detection of containerized environments - fix issue #1732

[flavio]

The code determining if Portus is running containerized is broken on Kubernetes because in this case PID 1 doesn't belong to the docker cgroup slice.

That causes the code to invoke hostnamectl, that tries to access kdbus from within a container. That causes Rails to fail immediately.

fixes #1732

[flavio]

@mssola the code is gone from master. However I think we should fix it inside of the v2.3 branch.

[jordimassaguerpla]

I can't test it right now but given the detailed bug report, this makes a lot of sense , hence I am approving this. Thanks for this fix.

[mssola]

Thanks a lot 👏

[Monnoroch]

@flavio also, docker is not the only container runtime. There's also containerd, rkt and probably more. Checking for kubepods is great for kubernetes, but this check has to be redesigned to handle more runtimes.

[flavio]

@Monnoroch by default kubernetes places all the user PODs under this cgroup namespace, despite of the CRI in use.

[Monnoroch]

There is life outside kubernetes too. Besides, what if they change the cgroup name? Current implementation looks like a hack, really. I think there should be a way for a user to force assuming a containerized environment. Or, better yet, making the software environment-agnostic.

[flavio]

As far as I've seen this code is gone from the master branch of Portus. I see this just as a temporary fix.

@mssola can you confirm what I've said, we are not going to have to deal with this code anymore. Am I right?

[mssola]

As far as I've seen this code is gone from the master branch of Portus. I see this just as a temporary fix.

That's correct. This fix only applied to the v2.3 branch. For the master branch we are using this portusctl, which does not need this kind of check.