jinge <altman87120@...> writes: > > Thanks Amyas! > > But why just connect to the Direct (transparent) connection so high.
Which number is "too high" for direct? Maybe I misunderstood the question there. If I have maxconn 20 and I am getting 20 connections/sec then I will quickly end up with a huge number of sockets in various closing states - do you know what the tcp time_wait fin_wait2 timeouts are on your system? There are problems with setting the connection closing timeouts too low. The issues related to tuning tcp timeouts are complex and not universal: http://serverfault.com/questions/329845/how-to-forcibly-close-a-socket-in-time-wait I would do one thing until you have a working system - don't have any stuff in the defaults section, just put every option that you want to use in the frontend or backend so it is easier to see what options that part of your system is using explicitly. Others might disagree. Another thing I would do is not not log normal, how do you know how long nginx is taking to service requests from an haproxy point of view, how do you know the final connection state of a connection for a request going through haproxy? Turning off log normal is probably ok after you are sure everything is running fine.
