We are trying to understand the queueing behavior in haproxy better, especially with regard to global queues, queue depth, and weighting upstream servers when using round-robin distribution.
A few questions: - How can we balance across upstream servers by using different weight assignments, based on *performance* - For each listen socket, is there a shared queue before individual maxconn queues are filled? Does maxconn per listen socket set the size of this queue? If global maxconn is 1000, listen socket maxconn is 100 and individual server maxconn are set to 10, and we have 2 app servers, what's the total number of requests that will be queued if the app servers are not able to catch up to the traffic? Blake
