Stats and multiprocessing + multithreading in same time

#1

Hello

Im using Haproxy with both nbproc and nbthread.
this is my configuration :

nbproc 6
nbthread 8
cpu-map auto:1/1-8 0-7
cpu-map auto:2/1-8 8-15
cpu-map auto:3/1-8 16-23
cpu-map auto:4/1-8 24-31
cpu-map auto:5/1-8 32-39
cpu-map auto:6/1-8 40-47

this command confirms I have 48 threads with 1 master
ps -efL | grep haproxy | grep -v exporter | grep -v grep | wc -l

the problem is statistics, I don’t have a valid statistics with the following configuration :

listen stats-Static
 bind *:4444
 stats uri /Static-stats
 bind-process 5
 stats enable
 stats scope front-http
 stats scope Static
 stats hide-version
 stats show-legends

I even trying this on stats but its not working

 bind-process 5/1-8
 bind-process 5/32-39

How can I get every threads in process statistics ?
and another question is healthcheck is per thread or per process ?

thank you

#2

Why would you do that? That makes your setup unnecessarily complex.

Stats are per-process, not per-thread. You bind stats to a process, not to a thread, and you see everything from that process.

Per-process.

#3

to gain maximum performance .

the problem is exactly here when i use this mapping everything would fine :
cpu-map auto:1/all 0
cpu-map auto:2/all 1
cpu-map auto:3/all 2
cpu-map auto:4/all 3
cpu-map auto:5/all 4
cpu-map auto:6/all 5

I have 60000 sess/s on my application, one frontend 8 backends with 28 servers. Im using 5 servers for Haproxy (2x E5-2650 and 256gb ram ) and its important for me the gain maximum performance.

that’s good because with that number of servers health check is a bit challenging for me it takes 20% CPU usage on my worker server and I need to figure this out.

what do you suggest?

#4

hi @arash , how are you?

after set “nbproc” and “cpu-map” on global section,do you config “stats socket” per process on global section and config “bind-process” on your “frontend” section?

#5

Hi , Thanks long time no see ,

Yes all of processes has its own stats socket .

if you use this mapping config everything works
cpu-map auto:1/all 0
cpu-map auto:2/all 1
cpu-map auto:3/all 2
cpu-map auto:4/all 3
cpu-map auto:5/all 4
cpu-map auto:6/all 5

but if you use mapping to a range you willbe got into problem .

#6

Doesn’t all this change with the Master/worker model in HAProxy 1.8.x? I’m not sure how it really works, but if not, can the worker report stats back to the master and then we can bind stats to the master some how?