How to simulate cpu pressure metrics for DRS?

[yezifeng]

I'm interested in DRS (Distributed Resource Scheduler) and followed the patch to do some testing.

The patch: https://lists.proxmox.com/pipermail/pve-devel/2022-April/052780.html

It gets VM pressure metrics via cgroup v2 (like: cat /sys/fs/cgroup/qemu.slice/147.scope/cpu.pressure). Then I simulated the high CPU load through stress command, but I couldn't make the CPU pressure indicator change.

Anyone know how to properly simulate cpu pressure?

Help please.

 

[yezifeng]

New progress

Run stress command on node (40 CPUs). Get the correct cpu pressure value.:

root@pve150:~# stress -c 60

root@pve150:~# cat /proc/pressure/cpu
some avg10=28.69 avg60=6.40 avg300=1.38 total=136756221

But run stress command on vm (4 CPUs). The cpu pressure value remains the same.：

root@vmtest:~# stress -c 8

root@pve150:~# cat /sys/fs/cgroup/qemu.slice/147.scope/cpu.pressure
some avg10=0.00 avg60=0.00 avg300=0.00 total=415

Problems with cgroup v2?