Killing a child process and all of its children in Go

Felix Geisendörfer ·

Killing a child process and all of its children in Go

This blog post explores some common unix child process control use cases, and how to perform them in Go. I’m mostly sharing it to improve my own understanding of the subject, but hopefully this will be useful to others as well.

Problem: You are using os/exec to run a child process, but need to terminate it under certain conditions, e.g. after a timeout of 3 seconds.

Solution: Just call cmd.Process.Kill() when your condition triggers:

As you can see, the watch process is still alive and has been adopted by init (PPID = 1). The reason for this is that, under-the-hood, Go is is using kill(2) to send a KILL signal to the PID of the sh process, but not the watch process, turning it into an orphan. This is completely normal behavior for a unix programming language, but not desirable for many use cases.

Solution: In addition to sending a signal to a single PID, kill(2) also supports sending a signal to a Process Group by passing the process group id (PGID ) as a negative number. Child processes get the same PGID as their parents by default, so in order to avoid suicide, we’ll also need to ask Go to create a new process group for our child by setting the Setpgid field of syscall.SysProcAttr. Putting it all together, we get this:

Internally this will cause Go to call setpgid(2) between fork(2) and execve(2), to assign the child process a new PGID identical to its PID. This allows us to kill all processes in the process group by sending a KILL to -PID of the process, which is the same as -PGID. Assuming that the child process did not use setpgid(2) when spawning its own child, this should kill the child along with all of its children on any *Nix systems.

Further reading: Process control in unix a is a complex topic, so I recommend the resources below to learn more.