Next.js production build doesn't prevent process exiting due to unhandled exceptions

Bug report

Describe the bug

Next.js process exits with code 1 when API route throws exception in production mode.

A common mistake that could easily lay dormant causes Next.js to exit the PRODUCTION server when it eventually gets run. Yes, I understand pragmatically you shouldn’t be masking response errors, but I don’t feel very confident using Next.js in production, when a simple race condition that’s almost never triggered can bring down the server.

To Reproduce

$ npx create-next-app production-process-exits

// pages/api/hello.js

export default (req, res) => {
  res.end().write()
}

Or

// pages/api/hello.js

export default (req, res) => {
  setTimeout(() => res.write(), 1000)
  res.end()
}

$ yarn build
$ yarn start

Visit http://localhost:3000/api/hello, see process exit with the following stack trace:

events.js:287
      throw er; // Unhandled 'error' event
      ^

Error [ERR_STREAM_WRITE_AFTER_END]: write after end
    at write_ (_http_outgoing.js:637:17)
    at ServerResponse.write (_http_outgoing.js:629:15)
    at ServerResponse.write (/home/max/production-process-exits/node_modules/next/dist/compiled/compression/index.js:1:148632)
    at Timeout.eval [as _onTimeout] (webpack-internal:///./pages/api/hello.js:4:24)
    at listOnTimeout (internal/timers.js:549:17)
    at processTimers (internal/timers.js:492:7)
Emitted 'error' event on ServerResponse instance at:
    at writeAfterEndNT (_http_outgoing.js:692:7)
    at processTicksAndRejections (internal/process/task_queues.js:85:21) {
  code: 'ERR_STREAM_WRITE_AFTER_END'
}
error Command failed with exit code 1.

Expected behavior

Error should be output without exiting the process.

System information

• OS: Ubuntu 20.04
• Version of Next.js: 9.5.3
• Version of Node.js: 12.16.3

Additional Context

Short of wrapping my entire API in a try/catch block, I am using the following workaround at the entrypoint of my app:

// pages/_app.js

if (process.env.NODE_ENV === 'production') {
  process.on('uncaughtException', (error) => {
    console.error(error.stack)
    // Don't run process.exit(1)
  })
}

function MyApp({ Component, pageProps }) {
  return <Component {...pageProps} />
}

export default MyApp

[HaNdTriX]

Let it crash!

From my point of view this is the wanted behaviour.

Keeping the node process running on uncaught exceptions may lead to new errors (e.g. Memory Leaks, CPU peaks, etc.) that may be very hard to debug.

Your production setup should handle these crashes. Check out the following links if you want to know how:

• forever
• Docker restart policy
• pm2 watch
• kubernetes restart

[timneutkens]

This has been fixed a while ago, we now process.exit(0) at the end of the build process when there are no errors.