On Wed, 2020-05-27 at 21:10 +0200, Michael Osipov wrote: > Am 2020-05-27 um 19:13 schrieb Oleg Kalnichevski: > > On Wed, 2020-05-27 at 18:26 +0200, Michael Osipov wrote: > > > > > > > ... > > > > > > > > https://github.com/ok2c/httpcomponents-core/commit/732a99563e73529e0359625fda54351b843b2e34 > > > > > > This patch does not work (reliably): > > > > > > * I have never seen EarlyResponseException > > > * When run over HTTPS I see early responses somtimes, sometimes I > > > get > > > SSLException > > > * Via HTTP I never see an early response, but only > > > SocketException > > > > > > Buffers again? Note that curl reads the physical response in any > > > case. > > > > > > See output here: > > > http://home.apache.org/~michaelo/issues/early-response/try-1/ > > > > > > > I do not want to give up yet. Could you please try out this patch? > > > > https://github.com/ok2c/httpcomponents-core/commit/8d7b2e3d38b4d696d45cde16c52dba278be15438 > > Apologies, I made a mistake, since take 1 was 5.1-SNAPSHOT, I > assumed > that take2 has this version too, noticed too late that it is > 5.0.1-SNAPSHOT actually. > > I have redone the tests, and raised to 50 iterations. Write buffer > in > HttpEntity is 16 KiB. > > 732a99563e73529e0359625fda54351b843b2e34: > * Performs great on HTTP, fails really early > * Does not work with TLS 1.3 at all > > 8d7b2e3d38b4d696d45cde16c52dba278be15438: > * Works with HTTP, but takes a lot of payload to detect early > response > * HTTPS is a bit better, it takes around 400 kB. > > Output is here: > http://home.apache.org/~michaelo/issues/early-response/try-2/ > > Michael
I do not think it really matters if it is 0 kb or 400 kb as long as the server does not reset the connection and the request gets correctly retried. I will improve the patch further. Please bear with me. Oleg --------------------------------------------------------------------- To unsubscribe, e-mail: httpclient-users-unsubscr...@hc.apache.org For additional commands, e-mail: httpclient-users-h...@hc.apache.org