What were you trying to do that didn't work?

Noticed that irqbalance uses a whole core of CPU virtually all the time on a very large server.  The system is a 32 socket HPE system with Sapphire Rapids processors has 1920 cores/3840 LCPUs 64TB of RAM.

Analysis with strace has shown that irqbalance is reading /proc/interrupts 1K IO's at a time.  On this scale system, the size of /proc/interrupts file object is nearly 512MB.  It takes roughly 9 seconds to just read that much data on this server.  Irqbalance cycles every 10 seconds, by default, thus it continuously runs.

Looking into changing the default wakeup time but looking for feedback as to how we handle larger systems.  Do we need to advise customers to change the wake up interval ?  KBASE article ?  Something more automatic ?

What is the impact of this issue to you?

Just noticed its consuming a core.  Wasn't sure if changes to irq's through it in such a state could effect system stability

Please provide the package NVR for which the bug is seen:

irqbalance-1.9.2-3.el9.x86_64

How reproducible is this bug?:

All the time on a very larger server

Steps to reproduce

1. top
2.  
3.  

Expected results

Actual results