Ignore frontend errors not originating from application script

[aduth]

🛠 Summary of changes

Updates logic of isTrackableError to ignore any errors which occur from the same host, but which aren't from JavaScript files.

Following #10027, data for filename shows that errors are originating from filenames such as https://secure.login.gov/. The hypothesis is that app webviews or browser extensions may be injecting scripts evaluated on the page and throwing errors, which we want to ignore.

📜 Testing Plan

1. echo "setTimeout(() => { throw new Error('example'); }, 100)" >> app/javascript/packages/analytics/index.ts
2. NODE_ENV=production yarn build
3. rails s
4. Go to http://localhost:3000
5. Validate NewRelic logging. Easiest way is to add a binding.pry at the notice_error call and check that it's triggered

[zachmargolis]

do you think it would make sense to add a similar check on the server? I guess we could always add later if it turns out to be too noisy

[aduth]

do you think it would make sense to add a similar check on the server? I guess we could always add later if it turns out to be too noisy

That's a good question. I think generally yes, in the sense that we shouldn't trust any "validation" to occur exclusively in the client.

But it also sparks an idea that we could do this exclusively server-side, on which I'm conflicted:

• Since this is a critical-path script, anything to reduce its size is a win. And doing it in one place is simpler to maintain.
• But if we moved all validation to the server, it would put additional strain in handling X more requests, which could otherwise be filtered on the client

I'd probably lean toward keeping this as-is, and following-up with a pull request to perform the same validation server-side as a fallback.

[zachmargolis]

makes sense to me