Perf-Alert: Plugin - CPU-usage is per core

[olivierlambert]

Hi!

I'm not sure to get it

[KPS]

@olivierlambert
The alert is already triggered, when ONE vCore is at >95%
IMHO, it should trigger, when the VM usage exceeds 95% on all its cores

[olivierlambert]

Just to double check I get it:

1. You configure the perf alert plugin to report VM usage at more than 95%
2. You have a VM with 16 cores having only one core at 100%, the rest is almost idle.
3. You do NOT want to be alerted in that case, only if ALL cores are beyond 95%

Is that correct?

[KPS]

@olivierlambert
Correct - nearly…
When the total usage is >95% inside the vm

[olivierlambert]

Not sure to get the difference between "ALL cores are beyond 95%" vs "total usage >95%", isn't the same thing?

[KPS]

@olivierlambert
Nearly, but no:

Take a 2-core system with

• core 1 at 100%
• core 2 at 91%

Total >95%, but not all cores >95%

[olivierlambert]

Ah yes, you mean the average total CPU usage then Got it, thanks!

edit: or the sum, like if you have 8 cores, 800% would be the max and the threshold will be 760% (equivalent to 95 per 100)

[olivierlambert]

@Bastien-Nollet can you check the current logic for CPU alerts? I thought initially it was like @KPS described

[Bastien Nollet]

@olivierlambert I'm not familiar with the perf-alert plugin but it seems that @KPS is right, see : https://github.com/vatesfr/xen-orchestra/blob/7882debe7adc7d4945da7e27d9825b1ffcf78ca9/packages/xo-server-perf-alert/src/index.js#L31

[olivierlambert]

Okay so it will be more logical to use the same logic than with the Load balancer, where we compute a total of all vCPUs