On Sun, Oct 12, 2014 at 06:09:46PM +0200, Marcin Szewczyk wrote: > On Sun, Oct 12, 2014 at 05:25:43PM +0200, Marcin Szewczyk wrote: > > cgmanager: DEBUG per_ctrl_move_pid_main: 31385, memory - > > user.slice/user-1000.slice/session-c13.scope > > > > That means that when systemd-shim calls cgmanager to MovePid/MovePidAbs > > it uses a special controller name "all". But it seems "all" doesn't > > include some controllers -- for example "name=systemd". > > > > Questions are: > > 1) Should cgmanager move pid to "name=systemd" when called with "all" > > controller? > > So, I think it should and it would. I omitted very important line from > my previous log: > > cgmanager: DEBUG per_ctrl_move_pid_main: 31385, memory - > user.slice/user-1000.slice/session-c13.scope > cgmanager: Found no cgroup entry for pid 31388 controller memory > > The thing is -- cgmanager jumps out from the "all" loop after some > problem with the "memory" controller. I do not know what to do with it > yet.
I have applied a patch (attached) to cgmanager and everything seems to work now. I will send it to cgmanager maintainer and upstream to ask if it makes sense. -- Marcin Szewczyk http://wodny.org mailto:[email protected] <- remove b / usuĊ b xmpp:[email protected] xmpp:[email protected]
>From f9e01540decd7f55cbb8a2057249b93c5384801c Mon Sep 17 00:00:00 2001 From: Marcin Szewczyk <[email protected]> Date: Sun, 12 Oct 2014 18:45:58 +0200 Subject: [PATCH] check "enabled" column when parsing /proc/cgroups --- fs.c | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/fs.c b/fs.c index 827e95f..cdfcc92 100644 --- a/fs.c +++ b/fs.c @@ -552,6 +552,15 @@ static bool collect_kernel_subsystems(void) continue; *p = '\0'; + // TODO: How stable is /proc/cgroups interface? + // Check the 'enabled' column + p = strrchr(p+1, '\t'); + if (!p) + continue; + + if (*(p+1) != '1') + continue; + if (!save_mount_subsys(line)) { nih_fatal("Error storing subsystem %s", line); goto out; -- 2.1.1
