centos9+systemd: setting cpu-period resets cpu.idle

[kolyshkin]

Description

One of our integration tests is failing on CentOS Stream 9 when using systemd cgroup driver.

What happens is cpu.idle setting is getting reset from 1 to 0 when setting cpu-period.

Steps to reproduce the issue

Minimal repro in bats:

#!/usr/bin/env bats

load helpers

function teardown() {
        teardown_bundle
}

function setup() {
        setup_busybox
        set_cgroups_path
}

@test "update cgroup cpu.idle" {
        requires cgroups_cpu_idle
        [ $EUID -ne 0 ] && requires rootless_cgroup

        runc run -d --console-socket "$CONSOLE_SOCKET" test_update
        [ "$status" -eq 0 ]

        check_cgroup_value "cpu.idle" "0"
        runc update --cpu-idle 1 test_update
        check_cgroup_value "cpu.idle" "1"

        # test update other option won't impact on cpu.idle
        runc update --cpu-period 10000 test_update
        [ "$status" -eq 0 ]
        check_cgroup_value "cpu.idle" "1"
}

Describe the results you received and expected

[root@localhost runc]# export RUNC_USE_SYSTEMD=yes
[root@localhost runc]# bats  tests/integration/u2.bats 
1..1
not ok 1 update cgroup cpu.idle
# (from function `check_cgroup_value' in file tests/integration/helpers.bash, line 265,
#  in test file tests/integration/u2.bats, line 28)
#   `check_cgroup_value "cpu.idle" "1"' failed
# runc spec (status=0):
# 
# runc run -d --console-socket /tmp/bats-run-102847/runc.ooH08F/tty/sock test_update (status=0):
# 
# current 0 !? 0
# runc update --cpu-idle 1 test_update (status=0):
# 
# current 1 !? 1
# runc update --cpu-period 10000 test_update (status=0):
# 
# current 0 !? 1

What version of runc are you using?

[root@localhost runc]# ./runc --version
runc version 1.1.0+dev
commit: e6cc9a67
spec: 1.1.0-rc.1
go: go1.19
libseccomp: 2.5.2

Host OS information

[root@localhost runc]# cat /etc/os-release 
NAME="CentOS Stream"
VERSION="9"
ID="centos"
ID_LIKE="rhel fedora"
VERSION_ID="9"
PLATFORM_ID="platform:el9"
PRETTY_NAME="CentOS Stream 9"
ANSI_COLOR="0;31"
LOGO="fedora-logo-icon"
CPE_NAME="cpe:/o:centos:centos:9"
HOME_URL="https://centos.org/"
BUG_REPORT_URL="https://bugzilla.redhat.com/"
REDHAT_SUPPORT_PRODUCT="Red Hat Enterprise Linux 9"
REDHAT_SUPPORT_PRODUCT_VERSION="CentOS Stream"

Host kernel information

[root@localhost runc]# uname -a
Linux localhost 5.14.0-282.el9.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Feb 23 14:35:36 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux

[kolyshkin]

And the reason is this PR systemd/systemd#23299 which adds support for cpu.idle. It adds a special value idle for CPUWeight parameter. In case CPUWeight="idle" is set, cpu.idle is set to 1, otherwise it is set to 0.

So, our test case which checks that setting --cpu-period (which is translated to CPUWeight) does not affect cpu.idle fails.

[kolyshkin]

This was added to systemd v252. Currently F37 is shipped with systemd v251, and CentOS 9 comes with systemd v252 -- thus we don't see the failure on Fedora.

[kolyshkin]

Kernel API (not documented elsewhere): https://lore.kernel.org/all/[email protected]/