Hello Maciej, On Wed, Sep 08, 2010 at 08:01:42PM +0200, Maciej Bednarz wrote: > Hello, > > I have a problem with haproxy configuration. First let me explain my > scenario. I have two hardware servers: > > Server A runs a HAPROXY 1.4 and is accessible by HTTP clients > Server B runs two instances of HTTP backend software (two sockets) and is > only accessible by HAPROXY > > When now "thousands" of "crazy" clients access HAPROXY (500 mixed non > keep-alive and keep-alive requests/second), all requests are forwarded 1:1 to > my backend. If I have on Server A 100.000 sockets in TIME_WAIT, the same > amount of sockets are also in the same state on Server B.
Why is that a problem ? 100000 TIME_WAITs are not much at all. Sites running at 20000 connections per second generally have over one million. TIME_WAITs are very cheap (just a few bytes). I could reach 4 millions once during a test, they have almost no measurable impact on performance. > What I need is some kind of rewirte / multiplexing / pooling mechanizm to > achive the following: > > HAPROXY keeps 100 keep-alive connections to the backend and uses them as a > pool. No new connections are opened. When now a client will access the > HAPROXY its request should be distributed to one of the allready existing > connections without opening a new connection to the backend. This mechanism (connection multiplexing) will be worked on once we have server-side keep-alive. There are several issues doing this. HTTP only allows you to send an idempotent request over a keep-alive connection, because if the connection closes while you're sending, you have to re-emit the request without knowing if it was processed or not. So basically you must not send POSTs nor any request looking like a dynamic request, which limits the feature to static files. I know that some vendors tend to do that on anything because it looks better. Just ask them what customers use that on your payment requests ;-) Anyway, I agree that multiplexing is important. In fact, in my opinion, server-side keep-alive can only be useful with multiplexing. Without multiplexing, server-side keep-alive is useless and can only increase the server's load and reduce its capacity. That's why I'd really like to implement this just after we have keep-alive. Regards, Willy
