take maximum processor Mhz

[gburanov]

Currently we take just first processor Mhz, and exit, assuming they are all the same.

We depend on "github.com/shoenig/go-m1cpu" to get the cpu data

On linux platform this either uses maximum frequency https://github.com/shoenig/gopsutil/blob/master/cpu/cpu_linux.go#L88 from cpuinfo_max_freq (that would be correct)

However some machines lack this file, and then we fallback to real frequency https://github.com/shoenig/gopsutil/blob/master/cpu/cpu_linux.go#L175-L179

Real frequency can be much lower than maximal, and taking just first processor value can give incorrect data.

For example on my machine

[georgy@kf06 ~]$ cat /proc/cpuinfo  | grep "cpu MHz"
cpu MHz		: 2200.000
cpu MHz		: 2685.956
cpu MHz		: 2200.000
cpu MHz		: 2252.721
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2700.009
cpu MHz		: 2200.000
cpu MHz		: 2592.342
cpu MHz		: 1906.243
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2042.605
cpu MHz		: 2200.000
cpu MHz		: 1423.054
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2552.760
cpu MHz		: 1645.222
cpu MHz		: 2591.675
cpu MHz		: 1858.395
cpu MHz		: 2699.999
cpu MHz		: 1231.719
cpu MHz		: 2688.291
cpu MHz		: 2236.831
cpu MHz		: 2200.000
cpu MHz		: 1244.091
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2631.418
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 2200.000
cpu MHz		: 1789.280
cpu MHz		: 2200.000
cpu MHz		: 1754.116
cpu MHz		: 2586.694
cpu MHz		: 1209.627
cpu MHz		: 2700.003

Would not taking the maximum value be better here?

[hashicorp-cla]

All committers have signed the CLA.

[shoenig]

Hi @gburanov! Indeed I think your patch is the correct thing to do in the short-term (in fact, I thought this is what we were doing already 🤦‍♂️).

In the longer term though we need to start accounting for P vs. E cores (which is what go-m1cpu does, but specifically for Apple M1/M2 chips only), which I think is going to require more changes to gopsutil, or another bit of custom code in Nomad.

And if this is for linux/arm64 there is also the idea of using dmidecode to extract the maximum CPU performance data: #4233