Shuttleworth: Losing graciously

Posted Feb 15, 2014 16:49 UTC (Sat) by mezcalero (subscriber, #45103)
In reply to: Shuttleworth: Losing graciously by stgraber
Parent article: Shuttleworth: Losing graciously

Here's one thing you keep repeating over and over again, which is however a bit misleading. The "single-writer" request by Tejun does not necessarily imply that all cgroup handling is implemented by the same process. It can happen with closely cooperating processes, and this is in fact what we do with systemd, where different systemd instances are responsible for different parts of the tree. So it's only systemd that manages the tree but from separate processes.

Lennart

Shuttleworth: Losing graciously

Posted Feb 15, 2014 17:33 UTC (Sat) by mbunkus (subscriber, #87248) [Link] (1 responses)

A question, if I may: when you say »from separate processes« do you mean multiple instances of /sbin/systemd running in various virtual machines in which each instance is managing the tree for its own virtual machine? Or is this still only one single host, no virtual machines involved, and you mean several processes of the systemd package that cooperate this way?

Shuttleworth: Losing graciously

Posted Feb 15, 2014 20:01 UTC (Sat) by mezcalero (subscriber, #45103) [Link]

The term "virtual machine" is usually reserved for kvm and suchlike, i.e. virtualization that emulates hardware. In such vm setups systemd on vms knows nothing of systemd on the host.

If your are refering to container virtualization (i.e. multiple userspaces on the same kernel) then yes, the idea is that the systemd in the container cooperates friendly with the host's systemd when it is interested in cgroups.

Lennart