Windows: npm run e2e does not properly clean up resources

[shazron]

Windows 11 x86, node-18, Windows Terminal with Powershell 7.

There are 2 node processes that are left running.

If you run npm run e2e once, everything will be fine, then the tests exit.
But if you run it again, it will complain about Error: listen EADDRINUSE: address already in use :::35729 (for livereload).
Something is not being cleaned up in Windows, but it is fine on macOS.

node:events:491
      throw er; // Unhandled 'error' event
      ^

Error: listen EADDRINUSE: address already in use :::35729
    at Server.setupListenHandle [as _listen2] (node:net:1740:16)
    at listenInCluster (node:net:1788:12)
    at Server.listen (node:net:1876:7)
    at Server.listen (C:\Users\shazr\Documents\git\work\adobe\aio-cli-plugin-app-dev\node_modules\livereload\lib\livereload.js:83:28)
    at exports.createServer (C:\Users\shazr\Documents\git\work\adobe\aio-cli-plugin-app-dev\node_modules\livereload\lib\livereload.js:256:14)
    at runDev (C:\Users\shazr\Documents\git\work\adobe\aio-cli-plugin-app-dev\src\lib\run-dev.js:109:41)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at async Dev.runOneExtensionPoint (C:\Users\shazr\Documents\git\work\adobe\aio-cli-plugin-app-dev\src\commands\app\dev\index.js:170:56)
    at async Dev.run (C:\Users\shazr\Documents\git\work\adobe\aio-cli-plugin-app-dev\src\commands\app\dev\index.js:59:7)
    at async Dev._run (C:\Users\shazr\Documents\git\work\adobe\aio-cli-plugin-app-dev\node_modules\@oclif\core\lib\command.js:117:22)
Emitted 'error' event on Server instance at:
    at Server.onError (C:\Users\shazr\Documents\git\work\adobe\aio-cli-plugin-app-dev\node_modules\livereload\lib\livereload.js:102:19)
    at WebSocketServer.emit (node:events:513:28)
    at Server.emit (node:events:513:28)
    at emitErrorNT (node:net:1767:8)
    at process.processTicksAndRejections (node:internal/process/task_queues:82:21) {
  code: 'EADDRINUSE',
  errno: -4091,
  syscall: 'listen',
  address: '::',
  port: 35729
}

Workaround

1. Open Task Manager
2. Search for node
3. End task for any node processes you find

[aiojbot]

JIRA issue created: https://jira.corp.adobe.com/browse/ACNA-2988

[shazron]

The aio app dev command is fine when run stand-alone, it's when its called via npm run e2e do the two node processes linger - so it must be cleanup of the exec calls.

[shazron]

Looks like it may be an execa/node issue in Windows (still investigating): sindresorhus/execa#433

[shazron]

One of the node processes is actually a background process - a child process of CCXProcess (part of CC Deskop), red herring. The remaining orphan node process is our server process not being killed.

[shazron]

Try to use spawn directly (see this) - when I get the child pid from the execa call, it does not exist when I try to kill it later. execa launches a shell (in this case cmd.exe), even when shell: false, windowsHide: false, and in turn, calls the node script, which probably is the process we need, not the cmd.exe process.