We use it together with OSPF and ECMP as a haproxy cluster, and We use the master-worker multi process model, there are many front end(s) and backend server(s), the bigest haproxy cluster may up to 2000 frontend and more than 10000 bakend servers total.
my one question is that : one of the cluster has 5 haproxy node, every node runs 46 core (multi process), so health check is expensive for the haproxy and the bakend servers 5 * 46 = 230 every 3 sec, I know that every process has own memory and the frontend and bakend are in one box. But may I use only one health check process ,and all the process share the result of the health check, then there are only one check every 3 sec , how to archive this ? sisuation: We had tested between single process and multi prcesses, the cpu load is nearly always 100% when using single process model, then We change to master-worker model, this model can archive high QPS and more cpu usage billy xue
