Child process remains alive after parent terminates while being debugged

[dead-claudia]

• Version: 8.0.0
• Platform: macOS Sierra
• Subsystem: child_process

Steps:

1. Type the following code below into a module.
2. Run it in a terminal with no arguments.
3. Attach Chrome's Devtools debugger to the child process.
4. Kill the parent.

if (process.execArgv.includes("--inspect-brk") return
child_process.spawn(process.argv[0], [__filename, "--inspect-brk"], {
  stdio: "inherit",
})
.on("exit", () => { console.log("exit") })
.on("error", () => { console.log("error") })

Expected:

The child should've been killed along with the parent, with the debugger disconnected.

Observed:

The child remains alive with the debugger at the first line waiting.

[dead-claudia]

Yes, the script recursively calls itself, but that's irrelevant to the repro (it's just easier to write/use).

[dead-claudia]

Also, it may be the underlying cause of #10158 and friends.

[dead-claudia]

My current workaround is using process.on("exit", () => child.kill()).

[bnoordhuis]

The child should've been killed along with the parent, with the debugger disconnected.

Why would you expect that? Node doesn't kill child processes when the parent process exits, debugger or no debugger. (The kernel does on ^C but that's a different mechanism.)

[dead-claudia]

@bnoordhuis

Why would you expect that? Node doesn't kill child processes when the parent process exits, debugger or no debugger. (The kernel does on ^C but that's a different mechanism.)

I edited the original bug to clarify killing the parent is what I meant. (I know ^C does that iff there are no SIGTERM handlers attached, so I removed that particular instruction.)

But here's the behavior I'm expecting:

1. A SIGTERM is sent to the parent.
2. The default action of termination + sending a SIGHUP to the child is performed.
3. No handler is registered for SIGHUP. (i.e. no process.on("SIGHUP", () => { ... }))
4. The default action of termination is performed.

---

The issue is likely that of the SIGHUP handler being registered and not killing the process upon reception.

[bnoordhuis]

Still sounds like the wrong expectation to me. When SIGTERM terminates the parent, children get reparented to the grandparent. That's not node-specific, it's traditional UNIX behavior.

You mention SIGHUP but that is normally only sent when the controlling tty goes away (nominally, when the user logs out.) Again, not node-specific.

[dead-claudia]

children get reparented to the grandparent

I competely missed that part (which really changes my expectations here). So I'll go ahead and close this.