Setup

Our environment for reproducing:

• pnpm 8.15.6, probably will work with npm just fine
• node v20.10.0

1. Run pnpm install
2. Copy .env.example to .env and fill out a newrelic api key and app name.
3. Then run pnpm start
4. Finally curl http://localhost:3000 or visit that in your browser. Do this several times, you should get an http 500.
5. Check newrelic APM and see an error similar to connect ECONNREFUSED <IPV4>. Clicking on this error will show a stack trace similar to:

Error: connect ECONNREFUSED 127.0.0.1:12345
    at TCPConnectWrap.afterConnect [as oncomplete] (node:net:1595:16) 
    at TCPConnectWrap.callbackTrampoline (node:internal/async_hooks:130:17)

Notice that there's no reference to this codebase. The error is clearly because of our fetch, but we properly wrapped it within a try/catch block!

Ideally this shouldn't get reported to newrelic at all as it's been handled. In the case that I'm wrong and this is truly unhandled, I should see my code in the stack trace, so I can find where I made my mistake!

We've also included an example that uses node-fetch instead which does not show the same behaviour. You can test this against the /node-fetch endpoint.

Bonus Round

Why are http 500 errors handled differently for the two fetch cases?

1. Change return res.status(200).send({ to send a http 500 in both of the catch blocks of the handlers.
2. Run pnpm start
3. curl http://localhost:3000/
4. Note that there ISNT any HttpError 500 errors reported in newrelic.
5. curl http://localhost:3000/node-fetch
6. Note that there ARE HttpError 500 errors reported in newrelic.