Erlang cgroups interface

An interface for cgroups manipulation that handles cgroup version details (i.e., differences between v1 and v2) and provides safe usage of the cgroups filesystem mount.

Build

rebar compile

Example

Update or create the cgroup "group1/nested1" with the OS pid 19368, then delete the cgroup path after moving the OS pid back to the root cgroup (executed as root).

# erl -pz ebin
1> application:start(cgroups).
2> OSPid0 = 19368.
3> CGroupPath = "group1/nested1".
4> {ok, CGroups} = cgroups:new().
5> MemoryLimit = case cgroups:version(CGroups) of 1 -> "memory.limit_in_bytes"; 2 -> "memory.high" end.
6> cgroups:update_or_create(CGroupPath,
                            [OSPid0],
                            [{MemoryLimit, "10000000"}],
                            CGroups).
7> cgroups:update("", [OSPid0], [], CGroups).
8> cgroups:delete_recursive(CGroupPath, CGroups).
9> cgroups:destroy(CGroups).

Troubleshooting

cgroups:update/4 and cgroups:update_or_create/4 Errors:

Either root execution of beam.smp or
setcap cap_sys_admin=+ep /path/to/beam.smp before execution is required
for most usage of cgroups
A Linux/systemd setup may have the control group setup mode set to hybrid
due to it being the recommended default
for distributions. However, that mode blocks the use of cgroup controllers
(the file /sys/fs/cgroup/unified/cgroup.controllers is empty)
because the cgroup controllers can only be mounted in one hierarchy
(v1 or v2). If cgroup v1 should be used, the Linux kernel argument
systemd.legacy_systemd_cgroup_controller=1 can be used.
If cgroup v2 should be used, the Linux kernel argument
systemd.unified_cgroup_hierarchy=1 can be used
(with systemd >= v226 and kernel >= 4.2) or
cgroup_no_v1=all can be used (with kernel >= 4.6).
Any non-root use should only use cgroups:update/4 with more limited
controllers (memory, pids, possibly cpu and/or io) and all OS pids added
elsewhere as root