Ping timeouts due to sequential frame processing

We've been investigating a problem with our Tornado server, where the WebSocket connection would drop if the handling of a single message would take too much time (SO Question).

When we finally found that the reason for our disconnections was that the _receive_frame (Link) function in the WebSocket class processes only one message at a time, we replaced the awaiting of the result:

if is_final_frame:
    handled_future = self._handle_message(opcode, data)
    if handled_future is not None:
        await handled_future

With

if is_final_frame:
    handled_future = self._handle_message(opcode, data)
    if handled_future is not None:
        asyncio.create_task(handled_future)

With that change, everything seems to work great and we no longer have disconnections.

My questions are:

1. Why is message handling not done in parallel to ping processing?
2. What is the proper way of using lengthy handlers, without causing WebSocket ping timeouts?

Start with tornado/websocket.py at WebSocket._receive_frame and compare the awaited _handle_message path with the asyncio.create_task workaround described here. Review the linked Stack Overflow report for the timeout scenario. Done means establishing the intended relationship between lengthy message handlers and ping processing, then documenting or implementing the project-approved resolution.