Hi Cyril, On Thu, Nov 01, 2012 at 12:18:04PM +0100, Cyril Bonté wrote: > Hi Willy, > > Le 01/11/2012 11:47, Willy Tarreau a écrit : > >On Thu, Nov 01, 2012 at 11:35:49AM +0100, Michael Seiferle wrote: > >>Hi Willy, > >> > >>I am sorry I have to reopen my inquiry: > >> > >>>http-pretend-keepalive + http-close > >> seemed to do the trick for GET Requests, yet when POSTing I still see > >> the attached errors on the actual server machine, and the client seems > >> to stall. > >>(if this exceeds community support please let us know) :-) > > > >It appears so much random and inconsistent that I think we're > >overlooking something. > > I'm making some tests, forging requests. I'm observing something weird > with haproxy-1.4 (not with haproxy-1.5). > Using "http-send-name-header Host", in some conditions, it appears that > haproxy sends garbage data (well, everything in the buffer I think). In > my tests, this results in sending 2 requests at a time.
Hmmm indeed since it needs to modify a request that was already scheduled for forwarding, so it has to play with output data. I would not be surprized if something goes wrong with post requests where data are already available in the buffer. Worse, incoming POST data can fill the buffer and prevent haproxy from rewriting too large a header. I think that I should amend the documentation to indicate that all server header values must be of the exact same length to avoid this issue :-/ > I try to make some more tests. Thanks, I'm working on another issue :-) Willy
