I'm finished.
Regular response and file uploading uses TCP_CORK to push all data in a single packet.
Streaming response and websocket use TCP_NODELAY to minimize latency.

Flow can be controlled by set_tcp_nodelay() and set_tcp_cork() calls. tcp_nodelay and tcp_cork states are mutually exclusive, case with neither nodelay nor cork is possible also (current aiohttp works exactly under this case).

I'll merge PR tomorrow. The only thing left is documentation.

Is stream uncorked after writing last byte of response while in keep-alive mode ?

Also, same technique should be applied to http client when it writes http request.

I remember interesting technique in streaming over http and using chunked data. This cover not only EvenrSource http://www.w3schools.com/html/html5_serversentevents.asp but also videostreaming.

I mean that in these cases each chunk of chunked response contains complete data that should be delivered to browserwith minimal delay. Even when delay between chunks are long. So, in current implemntation, user should uncork manually after sending chunk and cork again before sending chunk. Does not it ? If yes, please document that example.

Also for websockets. Library should cork before sending frame and uncork afer sending. Is it so ?

Sorry, but code slightly complicated for me.

This thread has been automatically locked since there has not been
any recent activity after it was closed. Please open a new issue for
related bugs.

If you feel like there's important points made in this discussion,
please include those exceprts into that new issue.