Quoting Tianon Gravi ([email protected]): > I absolutely agree that standardizing all the different methods of mounting > the cgroups would be a great idea. Ubuntu currently does this with their > "cgroup-lite" package (http://packages.ubuntu.com/precise/cgroup-lite), > which does pretty much exactly what we need. The only problems with it I > can see are that it has some weird love/hate relationship with "cgroup-bin" > that I can't quite figure out.
Hi, until he unfortunately had to bow out as maintainer, Jon Bernard and I had been talking about merging cgroup-lite (and sysvinit based equivalent) back into cgroup-bin. The idea was that the inherent racy parts of libcgroup (in particular the reclassification of programs after they start) would be yanked, leaving a simple startup and the useful binaries. > I would be very interested in and willing to maintain such a "cgroup > mounting" package in Debian if there's interest in it (especially from the > maintainers of the other packages needing cgroups mounted such as libvirt > and LXC). I think the approach taken by "cgroup-lite" is basically exactly > what we need, and I wonder if Serge Hallyn (who appears to be the most > active maintainer of the Ubuntu package from > http://changelogs.ubuntu.com/changelogs/pool/main/c/cgroup-lite/cgroup-lite_1.8/changelog) > would be amenable to my basing a new Debian package off the work there. Of course, whatever you need. For cgroup-lite I really wanted something that would just start as early and fast as possible, and be stupid enough to not risk breaking any libvirt or lxc installations. I just wanted cgroups mounted in time. Whether you would keep the new package as simple, just adding sysvinit scripts; or add some smarts so users can set initial cgroups; either way would be useful. > CC'ing Serge so we can hopefully get some added knowledge. :) > > ♥, > - Tianon Note though that my current focus is on cgmanager, and we hope to have lxc start to use it whenever possible instead of cgroupfs. Cgmanager offers a dbus interface for managing cgroups, and its goal is to allow even unprivileged and nested containers to manage their cgroups, subject to hierarchical constraints set up by logind and enforced by the kernel. thanks, -serge -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
