response.Body stream behaviour

[nstott]

In fetch, the response body is a ReadableStream with the data being pumped into the stream via the start method here: https://github.com/denoland/deno/blob/master/cli/js/fetch.ts#L25-L38

This is fine, but it has certain behavioural implications. we enqueue all the data as quickly as we can read it. there's no queue mgmt. this could cause memory problems if we have slow consumers of the data from these queues.

As an alternate implementation, we could pull the data by implementing

pull(controller: ReadableStreamController): Promise<void> {

rather then the current pump implementation. this means we get to apply a high watermark and have more queue management options
https://developer.mozilla.org/en-US/docs/Web/API/ReadableStream/ReadableStream

I'm not sure which is preferred at this point

[ry]

Your alternative is preferable. As we stream data only a fixed amount of memory should ever be allocated - as is done with Deno.copy.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.