Data race when calling Child.Stop with a process that takes too long to stop

[averche]

Consul Template version

v0.32.0

Configuration

The following minimal example to demonstrates the race. The full code can be found at averche/test-go/consul-template-child-gorace:

// 0s : start the child process
// 2s : stop the child process
// 7s : the library attempts to kill the child process
func run() (int, error) {
	process, err := child.New(&child.NewInput{
		Command:     "./my-script.sh", // a script that disregards SIGTERM
		Stdout:      os.Stdout,
		KillSignal:  syscall.SIGTERM,
		KillTimeout: 5 * time.Second,
	})
	if err != nil {
		return -2, err
	}

	if err := process.Start(); err != nil {
		return -3, fmt.Errorf("could not start the process: %s", err)
	}

	select {
	case <-time.After(2 * time.Second):
		process.Stop()
		return SuccessfullyStoppedTheProcess, nil

	case exitCode := <-process.ExitCh():
		return exitCode, nil
	}
}

func TestRunOnce(t *testing.T) {
	c, err := run()
	if err != nil {
		t.Fatal(err)
	}
	if c != SuccessfullyStoppedTheProcess {
		t.Fatalf("unexpected return code: %d", c)
	}
}

Debug output

$ go test -run TestRunOnce --race

2023/05/25 13:55:39 [INFO] (child) spawning: ./my-script.sh
sleeping for 20s
sleeping for 19s
2023/05/25 13:55:41 [INFO] (child) stopping process
received SIGTERM; ignoring it
sleeping for 18s
sleeping for 17s
sleeping for 16s
sleeping for 15s
sleeping for 14s
==================
WARNING: DATA RACE
Write at 0x00c0000f20c0 by goroutine 6:
  github.com/hashicorp/consul-template/child.(*Child).kill.func1()
      /Users/avean/go/pkg/mod/github.com/hashicorp/consul-template@v0.32.0/child/child.go:439 +0x84
  runtime.deferreturn()
...

Previous read at 0x00c0000f20c0 by goroutine 9:
  github.com/hashicorp/consul-template/child.(*Child).kill.func2()
      /Users/avean/go/pkg/mod/github.com/hashicorp/consul-template@v0.32.0/child/child.go:457 +0x70

... 

The full output can be found @ averche/test-go/consul-template-child-gorace.

Expected behavior

We attempt to stop the script, after 5 seconds, the script is killed immediately.

Actual behavior

The script is killed correctly but a go race is detected in tests.

Steps to reproduce

1. git clone /~https://github.com/averche/test-go && cd test-go/consul-template-child-gorace/
2. go test -run TestRunOnce --race
3. See the "DATA RACE" output