Curtis: can you paste a log to confirm it's the same issue as Colin?
It looks like some of the cgroup:mixed logic recently introduced in cgfsng (to match that of the old cgfs backend) is a bit wrong here and attempts to create a cpu/cpuset,cpuacct symlink instead of symlinking cpu to cpuset,cpuacct.
Curtis: can you paste a log to confirm it's the same issue as Colin?
It looks like some of the cgroup:mixed logic recently introduced in cgfsng (to match that of the old cgfs backend) is a bit wrong here and attempts to create a cpu/cpuset,cpuacct symlink instead of symlinking cpu to cpuset,cpuacct.
Serge: can you take a look?