When running HAProxy with more than one CPU socket, it is [recommended](https://www.haproxy.com/documentation/hapee/2-1r1/configuration/system-tuning/#pin-network-interrupts-to-cores} that HAProxy should run with CPU affinity hence nbproc=2 and each process with its NUMA cores:
:~# lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
CPU(s): 40
On-line CPU(s) list: 0-39
Thread(s) per core: 2
Core(s) per socket: 10
Socket(s): 2
NUMA node(s): 2
NUMA node0 CPU(s): 0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38
NUMA node1 CPU(s): 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39
When trying to run my HAProxy with the following config:
nbproc 2
nbthread 32
cpu-map auto:1/1-16 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
cpu-map auto:2/17-32 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
I get error:
config : cannot enable multiple processes if multiple threads are configured. Please use either nbproc or nbthread but not both.
from the source code, it is not possible to use nbproc>1 and nbthread>1:
2213 if (global.nbproc > 1 && global.nbthread > 1) {
2214 ha_alert("config : cannot enable multiple processes if multiple threads are configured. Please use either nbproc or nbthread but not both.\n");
2215 err_code |= ERR_ALERT | ERR_FATAL;
2216 goto out;
2217 }
from Haproxy nbthread config it states using nbthread > This is exclusive with “nbproc” .
I want to optimise Haproxy running on dual socket server and run it in 2 processes where each process is binded to it Numa cores and utilizing the CPU L3 cache. but it seems that it is now impossible in newer HAproxy version.
To get better performance I’m forced to use only one CPU and not both with:
nbproc 1
nbthread 16
cpu-map auto:1/1-16 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
How can I optimise HAproxy on 2 sockets CPU and bind each to its own Numa cores?
1 post - 1 participant