CPUQuota does not imply CPUAccounting in systemd-run #9647
Description
systemd version the issue has been seen with
239-5 in Debian testing
Used distribution
Debian testing
Expected behaviour you didn't see
Assume the unified cgroup hierarchy.
systemd-run --user --scope --unit=foo -p "CPUQuota=50%" /bin/bash implies CPUAccounting as written in the manpage of systemd.resource-control, like
$ systemctl --user status -l foo.scope
● foo.scope - /bin/bash
Loaded: loaded (/run/user/1000/systemd/transient/foo.scope; transient)
Transient: yes
Active: active (running) since Wed 2018-07-18 19:35:32 UTC; 17s ago
Tasks: 2 (limit: 2365)
Memory: 2.3M
CPU: 24ms
CGroup: /user.slice/user-1000.slice/user@1000.service/foo.scope
├─401 /bin/bash
└─405 systemctl --user status -l foo.scope
Unexpected behaviour you saw
$ systemd-run --user --scope --unit=foo -p "CPUQuota=50%" /bin/bash
Running scope as unit: foo.scope
$ cat /sys/fs/cgroup/user.slice/user-1000.slice/user@1000.service/foo.scope/cpu.max
50000 100000
CPUQuota is functioning as requested.
$ systemctl --user show foo.scope | grep CPU
CPUUsageNSec=[not set]
CPUAccounting=no
CPUWeight=[not set]
StartupCPUWeight=[not set]
CPUShares=[not set]
StartupCPUShares=[not set]
CPUQuotaPerSecUSec=500ms
CPUAccounting is not implied?! Actually,
$ systemctl --user status -l foo.scope
● foo.scope - /bin/bash
Loaded: loaded (/run/user/1000/systemd/transient/foo.scope; transient)
Transient: yes
Active: active (running) since Wed 2018-07-18 19:36:27 UTC; 1min 11s ago
Tasks: 2 (limit: 2365)
Memory: 2.5M
CGroup: /user.slice/user-1000.slice/user@1000.service/foo.scope
├─406 /bin/bash
└─413 systemctl --user status -l foo.scope
CPU accounting information is not shown above. I am happy with the current behavior and suggest updating the documentation.
Steps to reproduce the problem
See above. It is observed on a Debian testing virtual machine with the workaround to systemd-run resource control issue. Systemd-run resource control does not work on a plain Debian testing installation as far as I see.