Crickets? @jfrazelle @crosbymichael @tianon

@rhatdan What kind of interactions will this enable? I'm assuming list is one thing, but what about the rest? Is the goal to expand support for more machinectl commands afterwards?

Yes I would like to see as much machinectl support, as possible. We would probably need to work with systemd guys on clarifying which commands will not work with docker containers. machinectl status, show, list, terminate, kill

Might be interesting getting copy-to and copy-from to work.

machinectl login, poweroff, reboot might work if the --init=systemd patch was accepted.

My biggest goal with this is just to get

journalctl -M CONTAINERID

To work in the situation where journald is running within the container.

Just making sure here: the rest of machinectl/Docker integration will be done in the machinectl codebase? It just needs the initial information that the container exists, right?

Right now we have a race condition with this. If a container exits before RegisterMachine gets called, it will log a message. I can add a patch to search for the Container PID on error and just ignore the error. In a perfect world we could prevent the container PID from disappearing until after the RegisterMachine completes.

Also should we move this code to runc rather then having it in the docker daemon at all.

I'd like to see registerMachine and terminateMachine behind an interface.
I could see wanting to be able to register containers with more than just systemd.

@cpuguy83 Added a machine interface, with linux and non linux. If someone adds another machine/container registration, then we could look at adding a buildtag for selinux

@rhatdan I was pretty upset when machinectl terminate NAME killed my X server. So, I dunno what to say about design :)

@LK4D4 It is a feature. You needed to get a way from the keyboard for a few minutes.

hahahahahaha

On Thu, Jul 23, 2015 at 1:55 PM, Daniel J Walsh notifications@github.com
wrote:

@LK4D4 https://github.com/LK4D4 It is a feature. You needed to get a
way from the keyboard for a few minutes.

—
Reply to this email directly or view it on GitHub
#13526 (comment).

So, this stuff is working somehow. But I think this is malicious code because of systemd bugs(killing not what it supposing to kill). So if include it at all in current design, then only in experimental branch.
ping @docker/core-maintainers

We are looking at moving this code into runc pre/post scripts rather then docker.
That way we could eliminate the race condition, where the container exits before it is registered.

@rhatdan As I see current PR makes no sense, because it only extends go systemd API. Maybe you should propose this changes to coreos/go-systemd first and then when we'll have pre/post scripts you will open PR with real implementation. Maybe systemd will be more generous to my X servers that time.

@LK4D4 Actually the calls have been lost from the patch. I sent patches for RegisterMachine/TerminateMachine to coreos a while ago, The current patch does not need some of the hire level stuff coreos version of systemd did.

I am going to close this and rework it for runc hooks, if and when this gets merged.