Ironically enough, LP actually makes a great case for what an inconsistent and semantically disparate interface cgroups really are. This gives quite a bit of va

Ironically enough, LP actually makes a great case for what an inconsistent and semantically disparate interface cgroups really are. This gives quite a bit of validation to Al Viro's acerbic rants against cgroups (among other questionably designed kernel features). As such, for a Linux-based service manager, one would be wise to skip cgroups entirely in favor of subscribing to a Netlink channel to get cn_proc events to get the "reliable process tracking" aspect, and optionally communicate with a well-defined external cgroup writer for delegating on cgroupfs operations (if at all desirable to support them beyond conforming to standards).