If you're fetching data and an error is thrown, the data continues fetching, and fetching...

[scf4]

We have a max size of 20mb set since the servers we fetch from can sometimes give us an endless amount of data (I mean 12 hours of useless 300GB data endless! 😄).

Once the size limit is breached, node-fetch successfully throws an error, the promise is rejected — but the HTTP request isn't stopped, so everything appears fine.

That is until you suddenly see huge spikes in inbound traffic and CPU, and eventually realize a node process has been running for 12 hours instead of 12 seconds!

Is this part of the spec, or a bug? It feels like unexpected behavior to me.

Thanks!

[bitinn]

Consider looking into these:

• Set a timeout or use AbortSignal to the same effect.
• In v3 release, we are more strict in piping network related errors, so hopefully this issue won't be repeated.
• Normally the underlying connection should timeout according to your OS socket setup, I haven't seen a situation where node-fetch keeps the connection alive (we default to keep-alive: close), so you might want to look into your agent config and your OS config.