- From: Martin Thomson <martin.thomson@gmail.com>
- Date: Mon, 21 Apr 2014 21:08:12 -0700
- To: Daniel Sommermann <dcsommer@fb.com>
- Cc: William Chan (ιζΊζ) <willchan@chromium.org>, HTTP Working Group <ietf-http-wg@w3.org>
On 21 April 2014 17:24, Daniel Sommermann <dcsommer@fb.com> wrote: > Actually, the extra ack GOAWAY only avoids this issue for high RPS proxy <-> > server connections. For low RPS connections, the server's hard shutdown > timeout would probably fire before all the extra MAX_CONCURRENT_STREAMS are > requested, so the same old race condition still applies. Our proxy/server > connections are extremely busy, so we haven't noticed this shortcoming of > the workaround. As long the server waits >1RTT after sending the first GOAWAY before identifying the *real* last request, isn't that enough? Sure, if your clients exceed MAX_CONCURRENT_STREAMS before the RTT is up, then you are pretty much guaranteed to be OK, but that's a clear sign that you are setting MAX_CONCURRENT_STREAMS too low.
Received on Tuesday, 22 April 2014 04:08:39 UTC