Post #1546493

@regehr definitely doable with some cgroup management; setting `memory.oom.group` to 1 means the OOM granularity would be one entire command execution and any children it might've forked. or even by polling the memory stats and taking action faster than the kernel's oom-killer, like oomd does.