Describe the bug
When following the steps to configure a self-hosted runner service the recommended way is to use the template, found here.

The file content is,

[Unit]
Description={{Description}}
After=network.target

[Service]
ExecStart={{RunnerRoot}}/runsvc.sh
User={{User}}
WorkingDirectory={{RunnerRoot}}
KillMode=process
KillSignal=SIGTERM
TimeoutStopSec=5min

[Install]
WantedBy=multi-user.target

When used in tandem with the Restart=always option or when manually restarting the service, child processes are still hanging around. Which causes multiple processes to pickup a job under the same runner and it will crash.

The flow is:

service -> calls runsvc.sh -> calls RunnerService.js -> spawns children processes

As seen in the RunnerService.js file,

        console.log("Starting Runner listener with startup type: service");
        listener = childProcess.spawn(
          listenerExePath,
          ["run", "--startuptype", "service"],
          { env: process.env }
        );
      }

To Reproduce

1. Start a runner with a service
2. Restart the service
3. Witness multiple children processes still hanging around from the last process.

Expected behavior
No orphan processes hanging around after one of the (grand-)parent processes has been killed.

Since we're using systemd for lifecycle management of the processes, it should be handling all processes spawned by the service.

Runner Version and Platform

v2.325.0

RHEL8

Fix

We found that simply setting KillMode=control-group was enough to fix the issue, it is also what is recommended in the systemd docs.