h2: study creation of a CLOSING state about haproxy HOT 1 OPEN

Quick update on this. First, the thread was brought there on the HTTP-WG mailing list:

https://lists.w3.org/Archives/Public/ietf-http-wg/2019JanMar/0056.html

Second among the ideas discussed (using PING+ACK, SETTINGS+ACK etc to measure the round trip time), all these only work with a cooperating client. A nasty client could very well refrain from ACKing and continue to send new streams or data for example. So the approach is nice to avoid truncating responses only.

For new streams and to be sure GOAWAY is taken into account, fixing a hard limit to last_stream_id + max_concurrent_streams + 1 guarantees a full round-trip was seen (GOAWAY is not ACKed), since this would imply that the closure that follows the GOAWAY was received to permit the last one to be sent. It could be used either alone, or in conjunction with RST_STREAM(REFUSED_STREAM) to figure when to really close. It also implicitly guarantees that the last data were received by the client. And it's not exclusive to the other ones above.

from haproxy.