The kernel’s control-group mechanism exists
to partition processes and to
provide resource guarantees (and limits) for each. Processes running
within a properly configured control group are unable to deprive
those running in a different group of their allocated resources (CPU time,
memory, I/O bandwidth, etc.), and are equally protected from interference
by others. With few exceptions, control groups are not used to take
direct actions on processes; Christian Brauner’s cgroup.kill
patch set is meant to be one of those exceptions.

Source: LWN.net – [$] A “kill” button for control groups