☰

如何避免委派单元的 cgroup.subtree_control 在服务重启时被重置

tag-icon • tag-icon linux systemd cgroups
如何避免委派单元的 cgroup.subtree_control 在服务重启时被重置

在systemd v250环境中，当我重新启动服务时，它cgroup.subtree_control被重置。

如果我修改 的值cgroup.subtree_control，那么当我重新启动服务时，systemd 将尝试覆盖 的值cgroup.subtree_control。例如，如果我cpu向 subtree_control 添加一个控制器，则当服务重新启动时，systemd 会将其从 subtree_control 中删除。如果我此时创建了一个子目录，并且重启时子组中还有进程，那么启动就会失败。错误如下：

Control process exited, code=exited, status=219/CGROUP
...
Unit process 543222  (xxx) remains running after unit stoped.
....

重写 subtree_control 时这似乎失败了。

我想cgroup.subtree_control在启用委托时自行管理。

我希望 systemd 不要修改它，不要cgroup.subtree_control在服务重新启动时重置委托服务。

相关文件

单元文件：

[Unit]
Description=DelegateTest

[Service]
Type=simple
TimeoutSec=60s
KillMode=process
ExecStartPre=/bin/bash /test/start_pre.sh
ExecStart=/bin/bash /test/loader.sh
ExecStop=/bin/kill $MAINPID
ExecReload=/bin/kill -HUP $MAINPID
Delegate=yes

/test/start_pre.sh：

echo "start_pre"

/测试/loader.sh

echo "executing"
CGROUP=$(cat /proc/$$/cgroup)
CGROUP_PATH=/sys/fs/cgroup${CGROUP#*::}
if [[ ! -d ${CGROUP_PATH}/job ]]; then
    mkdir ${CGROUP_PATH}/job
fi
echo $$ > ${CGROUP_PATH}/job/cgroup.procs
echo "+cpu" > ${CGROUP_PATH}/cgroup.subtree_control
ping 127.0.0.1 > /dev/null

重现此问题的步骤

  1. systemctl 启动 DelegateTest.service
  2. systemctl status DelegateTest.service
Loaded: loaded (/usr/lib/systemd/system/DelegateTest.service; static)
     Active: active (running) since Thu 2022-07-21 09:58:31 CST; 1s ago
    Process: 541635 ExecStartPre=/bin/bash /test/start_pre.sh (code=exited, status=0/SUCCESS)
   Main PID: 541636 (bash)
      Tasks: 2 (limit: 23196)
     Memory: 660.0K
        CPU: 11ms
     CGroup: /system.slice/DelegateTest.service
             └─job
               ├─541636 /bin/bash /test/loader.sh
               └─541639 ping 127.0.0.1
  1. systemctl stop DelegateTest.service
  2. systemctl status DelegateTest.service
Loaded: loaded (/usr/lib/systemd/system/DelegateTest.service; static)
     Active: inactive (dead) since Thu 2022-07-21 09:58:36 CST; 981ms ago
    Process: 541635 ExecStartPre=/bin/bash /test/start_pre.sh (code=exited, status=0/SUCCESS)
    Process: 541636 ExecStart=/bin/bash /test/loader.sh (code=killed, signal=TERM)
    Process: 541644 ExecStop=/bin/kill $MAINPID (code=exited, status=0/SUCCESS)
   Main PID: 541636 (code=killed, signal=TERM)
      Tasks: 1 (limit: 23196)
     Memory: 300.0K
        CPU: 13ms
     CGroup: /system.slice/DelegateTest.service
             └─job
               └─541639 ping 127.0.0.1
  1. systemctl 启动 DelegateTest.service
Job for DelegateTest.service failed because the control process exited with error code.
See "systemctl status DelegateTest.service" and "journalctl -xeu DelegateTest.service" for details.
  1. systemctl status DelegateTest.service
× DelegateTest.service - DelegateTest
     Loaded: loaded (/usr/lib/systemd/system/DelegateTest.service; static)
     Active: failed (Result: exit-code) since Thu 2022-07-21 09:58:53 CST; 1s ago
    Process: 541649 ExecStartPre=/bin/bash /test/start_pre.sh (code=exited, status=219/CGROUP)
      Tasks: 1 (limit: 23196)
     Memory: 300.0K
        CPU: 84us
     CGroup: /system.slice/DelegateTest.service
             └─job
               └─541639 ping 127.0.0.1

相关问题：

https://github.com/systemd/systemd/issues/24064 https://github.com/systemd/systemd/issues/20026 https://github.com/systemd/systemd/issues/18293 https://github.com/systemd/systemd/pull/9119 https://github.com/systemd/systemd/issues/8645 https://github.com/systemd/systemd/issues/18104

相关内容