How do applications know a DataChannel's buffer capacity, so they can avoid filling it?

[taylor-b]

A well-behaving application that sends a large amount of data will use the "bufferedAmount" event and "bufferedAmountLowThreshold", to ensure it doesn't fill the DataChannel's buffer and cause the channel to be closed.

But how does the application know what to set "bufferedAmountLowThreshold" to? If the spec doesn't provide any guarantees about an implementation's buffer size, the application developer is forced to research how large different browsers' buffers are, pick a value low enough, and hope the browser buffers don't get smaller.

So I'd suggest either:

• Providing a bufferSize attribute (pretty straightforward)
• Guarantee something about the buffer size. For example, "it MUST be N MB or larger".
• Get rid of the "data channel closing on buffer filling" behavior, and just throw/return an error instead, like a non-blocking socket. This is a more radical change, but is anyone really attached to the current behavior? It seems to cause a fair amount of confusion. See discussion here: https://groups.google.com/forum/#!topic/discuss-webrtc/apmq9VLY9lc

[stefhak]

I think this part is modeled after the web socket behavior (https://html.spec.whatwg.org/multipage/comms.html#dom-websocket-send). We could probably do better, but isn't this "standard" behavior, e.g. you can't find out how big objects you can post using fetch or xhr?