Potential bug: Uncaught TypeError: Failed to execute 'enqueue' on 'ReadableStreamDefaultController'

[dirkschumacher]

In a shinylive application that uses r-shinylive I observed the following client error on Chrome (MacOS) when doing a very long running computation that is also quite memory intensive.

Uncaught TypeError: Failed to execute 'enqueue' on 'ReadableStreamDefaultController': Cannot enqueue a chunk into a closed readable stream
    at MessagePort.

The code that produced the error is here

...
  if (msg.body) {
    streamController!.enqueue(filter(msg.body, response!));
  }
  if (!msg.more_body) {
    // All done
    streamController!.close();
    clientPort.close();
  }
...

It seems like it can happen that a streamController gets closed before all remaining messages have been enqueued. I am sure there are multiple ways to address it. I am also happy to contribute a fix of course (e.g. we could stop the event handler in case the streamContoller has been closed - but maybe the root issue is somewhere else). Also sorry for not providing a reproducible example - quite hard to do so, as this error only occasionally appears.

[dirkschumacher]

Ok quick update. The event was not processed out of order. The event handler received a big message after a few minutes with more_body:false. When trying to enqueue it the controller was already closed.

{type: 'http.response.body', body: Uint8Array(...), more_body: false}

I am continuing to investigate.

[dirkschumacher]

This happens, it seems, when a request takes too long to complete. Then the cancel() event is fired. A fix would be to react to cancel and only process message events if the controller has not been canceled yet.