[Bug?]: Seeing 500s due to ERR_STREAM_PREMATURE_CLOSE after upgrade to 7.x

[shivghai]

What's not working?

Never really saw ERR_STREAM_PREMATURE_CLOSE 500s in our logs ever before. After the upgrade to 7.x, on our deployed app, I see this issue constantly:

Error [ERR_STREAM_PREMATURE_CLOSE]: Premature close
    at Readable.onclose (node:internal/streams/end-of-stream:154:30)
    at Readable.emit (node:events:518:28)
    at emitCloseNT (node:internal/streams/destroy:147:10)
    at process.processTicksAndRejections (node:internal/process/task_queues:81:21)

Initially saw this issue (500s accompanied with this error) happen with our end to end tests, and then with some customers. Was eventually able to reproduce it, I kid you not, by just refreshing the page a few times. Can't repro this locally because presumably the stream interruption can't be reproduced without a certain amount of latency expected in an actual deployed scenario.

I don't think we should have to handle 500s due to this specific issue, though please advise if this is expected behavior. This seems to be a lower level issue being bubbled up as a 500.

How do we reproduce the bug?

Deploy app + refresh a page with some graphql queries (presumably) a bunch of times

What's your environment? (If it applies)

System:
    OS: macOS 13.6
    Shell: 5.9 - /bin/zsh
  Binaries:
    Node: 20.11.1 - /private/var/folders/8y/99byt5ms31d1z8jgjfmyvcg80000gn/T/xfs-f1930489/node
    Yarn: 3.2.1 - /private/var/folders/8y/99byt5ms31d1z8jgjfmyvcg80000gn/T/xfs-f1930489/yarn
  Databases:
    SQLite: 3.39.5 - /usr/bin/sqlite3
  Browsers:
    Chrome: 124.0.6367.62
    Safari: 16.6
  npmPackages:
    @redwoodjs/auth-auth0-setup: 7.4.2 => 7.4.2 
    @redwoodjs/cli-data-migrate: 7.4.2 => 7.4.2 
    @redwoodjs/core: 7.4.2 => 7.4.2

Are you interested in working on this?

• I'm interested in working on this

[Tobbe]

😬 This is going to be a tough one to figure out. Since we can't reliably reproduce this we're going to have to guess based on code changes. So realistically the only way for us to fix this is if you (or someone else who comes along in the future) can help narrow this down to a specific minor release, or, even better, a specific canary release.

@shivghai Could you please start by saying exactly what version you upgraded from and what version you first saw this issue on?

[shivghai]

Upgraded from 6.6.2 to 7.4.2. FYI, I saw the topic/streaming label added - we're not using any Realtime features (just in case that was the assumption).
Just to be clear, the 500s that we are seeing from the server are for GraphQL queries (presumably).

This is very reproducible for our deployed app (and I imagine it will be for others too). Just have to refresh any page with >1 queries very very quickly (do it about 20 times and maybe it's reproducible on your app?).

[shivghai]

I'll keep this issue open while I try to figure out which release it is related to. Shot in the dark would be it's because of Node 20

[Tobbe]

Upgraded from 6.6.2 to 7.4.2. FYI, I saw the topic/streaming label added - we're not using any Realtime features (just in case that was the assumption).
Just to be clear, the 500s that we are seeing from the server are for GraphQL queries (presumably).

Thanks for the clarification! I'm deep into React Server Components at the moment, so when I saw ERR_STREAM... my mind immediately jumped to React Streaming. As in streaming the frontend from server to client. But that was clearly not what you were talking about 😅