Michal Koutný

FWIW, the proposed changes are effectively redundant to [the service setting](https://github.com/kdave/btrfsmaintenance/blob/master/btrfs-balance.service#L9).

Beware that the service changes scheduling policy (cf. priority), i.e. `ps -o pri,cls` will show that. Also, from the reports it seems that lower IO scheduling has little effect (for...

OK, I will do it, however, it may take some time till I get down to it.

I'm afraid I won't help here. If the regression is observed on systems with comparable number of unit files, then it needs profiling PID 1 and systemctl to find the...

Cross-linking systemd/systemd#20328 or systemd/systemd#14730 (IOW this issue may be another reason to optimize it in systemd).

Thanks for the descriptive report, @MoritzS. The non-determinism is interesting, let me ask more then. When you reach the state when the `cpuset` controller is missing in `cgroup.subtree_control`, could you...

Nothing wrong here (`DelegeteControllers=` is set as expected). Do you also have the `systemd-analyze dump` data (there are some more details about `[email protected]`)? EDIT: corrected `systemctl dump` to `systemd-analyze dump`,...

I can't see where the loophole could be so I was trying to reproduce your problem but to no avail. The `subtree_control ` always lists `cpuset` controller after I start...

> systemd[921]: Failed to enable controller cpuset for /user.slice/user-1000.slice/[email protected] (/sys/fs/cgroup/user.slice/user-1000.slice/[email protected]/cgroup.subtree_control): Operation not permitted This is very interesting. It implies that `open(2)(cgroup.subtree_control)` succeeded (otherwise there'd be an earlier debug message) but...

Thanks for checking. So what about this -- are there any tasks in the `[email protected]` cgroup subtree that don't run as uid 1000? (That could fail the cpuset enabling attempt.)