TypeError: fetch failed due to ECONNREFUSED cannot be caught and handled

[Radiergummi]

Bug Description

I'm working with an API that is a bit unstable; in some cases, the server might go down entirely, and I need to account for that, log an error, and retry later. Alas, if the server is down Node encounters a refused connection, which in turn triggers a fetch failed error in Undici, and that - for some reason - kills the entire process.

Reproducible By

Take a look at this example (the Open Notify API used here isn't available via HTTPS, triggering the same error):

async function loadPosition() {
    const response = await fetch('https://api.open-notify.org/iss-now.json');
    const data = await response.json();

    if (!response.ok || !data.iss_position || data.message !== 'success') {
        throw new Error('Invalid response');
    }

    const {latitude, longitude} = data.iss_position || {};

    return {latitude, longitude};
}

setTimeout(async () => {
    try {
        const {latitude, longitude} = await loadPosition();
        console.log(`The ISS is at ${latitude}/${longitude}!`)
    } catch (error) {
        console.error('Could not fetch ISS position data');
    } finally {
        setTimeout(loadPosition, 5000);
    }
}, 5000)

Expected Behavior

If a connection error occurs, I'd expect it to be caught in the catch block, reported on the console, and be done with; the newly scheduled timeout in the finally block should be run after just fine.

Logs & Screenshots

Instead, this causes the process to exit immediately:

Could not fetch ISS position data
 node:internal/deps/undici/undici:13737
       Error.captureStackTrace(err, this);
             ^
 
 TypeError: fetch failed
     at node:internal/deps/undici/undici:13737:13
     at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
     at Timeout.loadPosition (/index.ts:53:22) {
   [cause]: Error: connect ECONNREFUSED 138.68.39.196:443
       at TCPConnectWrap.afterConnect [as oncomplete] (node:net:1605:16) {
     errno: -111,
     code: 'ECONNREFUSED',
     syscall: 'connect',
     address: '138.68.39.196',
     port: 443
   }
 }
 
 Node.js v21.7.1

[Process exited with code 0]

Environment

MacOS 14.3.1
NodeJS 21.7.1

[metcoder95]

👋
The issue is rooted to the way your code is scheduling the retry within the timeout for retrying once more.
The first setTimeout has the correct statements for handling a situation of a failure in case of an error, but on the second timeout called within the finally scope, it just calls the loadPosition once more without any error handling; meaning that if it fails, it will be treated as an unhandled rejection leading to the process termnation.

A possible alternative can be:

async function loadPosition () {
  const response = await fetch('https://api.open-notify.org/iss-now.json')
  const data = await response.json()

  if (!response.ok || !data.iss_position || data.message !== 'success') {
    throw new Error('Invalid response')
  }

  const { latitude, longitude } = data.iss_position || {}

  return { latitude, longitude }
}

const callback = async () => {
  try {
    const { latitude, longitude } = await loadPosition()
    console.log(`The ISS is at ${latitude}/${longitude}!`)
  } catch (error) {
    console.error('Could not fetch ISS position data')
  } finally {
    setTimeout(callback, 5000)
  }
}

setTimeout(callback, 5000)

It produces:

Could not fetch ISS position data
Could not fetch ISS position data
Could not fetch ISS position data

[Radiergummi]

Ohh, this is embarrassing. Of course. Thanks a bunch.