Systemd 253 released

Posted Feb 18, 2023 18:03 UTC (Sat) by dankamongmen (subscriber, #35141)
In reply to: Systemd 253 released by intelfx
Parent article: Systemd 253 released

if there's a means to create a toplevel slice within the framework of systemd (and thus not violate the single-writer principle), nope, this works (so long as one can go freely editing the systemd configs, which i'm willing to concede as being roughly equivalent to the access necessary to configure cgroups in the first place).

Systemd 253 released

Posted Feb 18, 2023 18:55 UTC (Sat) by dankamongmen (subscriber, #35141) [Link] (1 responses)

well, not quite, though. for instance, my application is provided a configuration. from this it derives a set of network interfaces it'll be working with. from these it derives a set of processor packages it'll be running on (only those packages which are local to the relevant PCIe root complexes). it wants exclusive ownership of these cores, though if there are no unused processors, it yields by necessity one physical core of each package back to the system.

i'm aware of no way to expose this in systemd.

Systemd 253 released

Posted Feb 18, 2023 19:01 UTC (Sat) by intelfx (subscriber, #130118) [Link]

> i'm aware of no way to expose this in systemd.

I'm not sure what exactly do you wish to expose. You can ask systemd for any desired cpuset for any cgroup in the system (also, of course, you can do it programmatically).

Systemd 253 released

Posted Feb 18, 2023 18:59 UTC (Sat) by intelfx (subscriber, #130118) [Link]

> if there's a means to create a toplevel slice within the framework of systemd

Well, yes, you can create a slice that is sibling to system.slice/user.slice/init.scope (and configure the three default ones above to get out of your way). I don't see what could even suggest otherwise.

This should give you everything you had with cgroups v1.