Updating to openrc-0.11.8, the cgroupfs is mounted at /sys/fs/cgroup/openrc/ despite of the setting rc_controller_cgroups="NO" in /etc/rc.conf I think a line has bee lost in /etc/init.d/sysfs at line 77ff: mount_cgroups() { * HERE A LINE MAY BE GONE, the next line is shifted in! * mountinfo -q /sys/fs/cgroup || return 0 local agent="@LIBEXECDIR@/sh/cgroup-release-agent.sh" mkdir /sys/fs/cgroup/openrc Compared to openrc-0.9.9.3, the missing line might be yesno ${rc_controller_cgroups:-YES} && [ -e /proc/cgroups ] && \ From a quick discussion on #openrc it turns out, that there is already a fix upstream (qnikst: 3df0bd64cb8d520cb4e65b1ac307d0b9fa4e83c3 <- that patch)
This will be in version 0.12.
Dear William, while performing the yearly update i have to deal with this for sys-apps/openrc-0.12.4, again. IMHO the mentioned patch restores the missing statements at another line. But this is done after the first mount and therefore it don't work as expected to me - the switch now will just disable the mounting of cgroup controllers but not mounting an instance of the cgroup fs itself. Beside, the line with the first mountinfo is still shifted in. mount_cgroups() { - yesno ${rc_cgroups:-YES} && [ -e /proc/cgroups ] && \ mountinfo -q /sys/fs/cgroup || return 0 local agent="@LIBEXECDIR@/sh/cgroup-release-agent.sh" @@ -83,6 +82,8 @@ mount_cgroups() -o none,nodev,noexec,nosuid,name=openrc,release_agent="$agent" \ openrc /sys/fs/cgroup/openrc echo 1 > /sys/fs/cgroup/openrc/notify_on_release + + yesno ${rc_controller_cgroups:-YES} && [ -e /proc/cgroups ] || return 0 while read name hier groups enabled rest; do case "${enabled}" in 1) mkdir /sys/fs/cgroup/${name}
Hi Guido, (In reply to Guido Jäkel from comment #2) > IMHO the mentioned patch restores the missing statements at another line. > But this is done after the first mount and therefore it don't work as > expected to me - the switch now will just disable the mounting of cgroup > controllers but not mounting an instance of the cgroup fs itself. Beside, > the line with the first mountinfo is still shifted in. The file system that is mounted on /sys/fs/cgroup is only a tmpfs; it isn't a cgroup fs. The /sys/fs/cgroup/openrc cgroupfs does not haveany controllers attached to it, so it shouldn't interfere with anything. The rc_controller_cgroups switch is now working as designed. Its purpose is to control the mounting of the controller cgroups, but that's it. The indentation has been fixed in git; that line should not be shifted in.
Hi William, > The file system that is mounted on /sys/fs/cgroup is only a tmpfs; it isn't > a cgroup fs. Right (, this is done elsewhere), but the "openrc" is one and this was disabled by the (renamed) switch before. > The /sys/fs/cgroup/openrc cgroupfs does not have any controllers attached to > it, so it shouldn't interfere with anything. > > The rc_controller_cgroups switch is now working as designed. Its purpose is > to control the mounting of the controller cgroups, but that's it. I found the thread #423317 - the point there is very similar. I read from it that "you" need this for openrc to (optional) kill tasks of a service. From the comments in /etc/rc.conf i see that this will become a successor of pid file based task termination. Therefore i have to "give in" and follow the standards, i.e. to work on the stuff that's not aware to arbitrary locations and multiple instances of cgroup file systems and it's controllers and not to "fix" something here which is broken elsewhere. Thank you for quick reply and the discussion. Guido
