其他goroutine中的恐慌无法停止子进程

Question

I need to run a long-running child process and kill it if I quit (for any reason) out of parent application.

Here is the code:

cmd := exec.Command("./long-process")

defer cmd.Process.Kill()

if err != nil {
    log.Fatal(err)
}

var fail io.ReadCloser
fail.Close()

The fail here produces obvious

panic: runtime error: invalid memory address or nil pointer dereference

It works as expected - the child process is killed.

But this happens in a goroutine:

cmd := exec.Command("./long-process")

defer cmd.Process.Kill()

if err != nil {
    log.Fatal(err)
}

go func() {
    var fail io.ReadCloser
    fail.Close()
}()

The panic still happens, but then it seems defer is not called and the child process is not killed.

Any way to go around this?

UPDATE I need a cross-platform solution (at least for Linux and FreeBSD)

Minimal example:

infinite-loop.sh

#!/bin/bash

while true; do
  sleep 1
done

Don't forget to

chmod +x infinite-loop.sh

test1.go (error checking left out for brevity):

package main

import (
    "time"
    "io"
    "os/exec"
    "runtime"
)

func main() {

    cmd := exec.Command("./infinite-loop.sh")

    cmd.Start()

    defer cmd.Process.Kill()

    go func() {
        time.Sleep(100 * time.Millisecond)
        var fail io.ReadCloser
        fail.Close()
    }()

    for {
        runtime.Gosched()
    }
}

Let's run

ps aux | grep infinite-loop.sh | grep -v grep | wc -l; \
go run test1.go; \
ps aux | grep infinite-loop.sh | grep -v grep | wc -l


     0 <--- !!


panic: runtime error: invalid memory address or nil pointer dereference
[signal 0xb code=0x1 addr=0x20 pc=0x2130]

goroutine 5 [running]:
main.main.func1()
.../multiline/test1.go:19 +0x30
created by main.main
.../multiline/test1.go:20 +0x9a

goroutine 1 [runnable]:
runtime.Gosched()
/usr/local/Cellar/go/1.5.1/libexec/src/runtime/proc.go:166 +0x14
main.main()
.../multiline/test1.go:23 +0x9f
exit status 2

     1 <--- !!

0 processes before and 1 after exit.

If you comment out goroutine code - it works fine.

Now we can kill it:

kill $(ps aux | grep infinite-loop.sh | grep -v grep | awk {'print $2'})

Answer 1

There's no cross-platform solution to automatically kill a child process.

On Linux, you can use the pdeathsig functionality:

cmd := exec.Command("./long-process")

cmd.SysProcAttr = &syscall.SysProcAttr{
    Pdeathsig: syscall.SIGTERM,
}

On other platforms, the child needs to determine when to exit on its own. One way is to monitor a pipe or socket FD given to it from the parent. You could also have a process manager of some sort monitor the processes and cleanup if something goes wrong.

In general though, panics should be rare and get fixed. If you do have areas of code that are prone to panic'ing, you can recover locally and call for the cleanup of child processes before exiting.