Hi Andreas, There is a configuration guide provided with old version of haproxy. It's not up to date, but delivers a very good overview of the product. IE: http://haproxy.1wt.eu/download/1.2/doc/architecture.txt
Concerning the second question, we call this feature "TCP buffering". There is no configuration variable, this is implemented withing HAProxy directly. In HAProxy 1.5 for sure, and latest HAProxy 1.4 (to be confirmed). Note this is your kernel who should decided the buffer size to open with the server, but it can be forced by HAProxy: the following parameter can be used to set the buffer size you allow the TCP socket to accept: tune.rcvbuf.server. I warn you to carefully read the manual, because such settings are very sensible to tune and can lead to run out of memory... Note as well this is a global parameter which applies to all backends in your HAProxy configuration. Baptiste On Tue, Aug 20, 2013 at 2:49 PM, Andreas Mock <[email protected]> wrote: > Hi all, > > as a total HAProxy newbie I have the following questions: > > 1) Is there a documentation explaining the way HAProxy is working? > There is the configuration manual, but I couldn't find a document > explaining the basic steps involved. It would be nice to have > a top level view on the way HAProxy is working with crossreference > to config variables determining several aspects of these process > steps. > > 2) I hope I understand it right: HAProxy is able (http-mode) to > buffer the output (response) from a backend server, so that the > server process is made free as soon as possible. A slow client > would therefore "only" consume ressources on the HAProxy. > How much data from the response can be buffered by HAProxy? > Is there a config variable? Is it tcp/ip buffering or > buffering in HAProxy userspace? > > Thank you in advance. > > Best regards > Andreas Mock > >
