loadavg figures - percentages or not?

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

loadavg figures - percentages or not?

Jamie Burchell

Hi there

 

On the Monit documentation on “System resource tests” (https://mmonit.com/monit/documentation/monit.html#System-resource-tests) the “loadavg” figures look like they could be percentages rather than absolute values. Is that correct, or am I misreading?

 

Most of the examples I’ve seen elsewhere for using the “loadavg” checks appear to be using single figures (https://mmonit.com/wiki/Monit/ConfigurationExamples - System Services).

 

If I’ve misunderstood then I’ll probably want to calculate these values based on the number of processors – but not necessary if these are percentages?

 

TIA

Jamie


--
To unsubscribe:
https://lists.nongnu.org/mailman/listinfo/monit-general
Reply | Threaded
Open this post in threaded view
|

Re: loadavg figures - percentages or not?

martinp@tildeslash.com
Hi,

the loadavg is not percent - as the manual states, it is absolute value: number of processes in the run queue. The practical limit depends on the number of CPUs and the typical load pattern - a rule of thumb we use is 2 processes per CPU core. If the machine has for example 48 cores, the loadavg of 96 is usually acceptable. There could be also spikes which are common and you may want to suppress false alerts, the example shows setup where high loadavg values for several consecutive cycles are needed before the alert is triggered.

To make the configuration easier, i think we can introduce some kind of "per CPU core" load average test, so the configuration will work the same regardless of CPU cores count, something like:

if loadavg(1m) per core > 1.9 then alert

Best regards,
Martin



On 8 Apr 2019, at 15:38, Jamie Burchell <[hidden email]> wrote:

Hi there

 

On the Monit documentation on “System resource tests” (https://mmonit.com/monit/documentation/monit.html#System-resource-tests) the “loadavg” figures look like they could be percentages rather than absolute values. Is that correct, or am I misreading?

 

Most of the examples I’ve seen elsewhere for using the “loadavg” checks appear to be using single figures (https://mmonit.com/wiki/Monit/ConfigurationExamples - System Services).

 

If I’ve misunderstood then I’ll probably want to calculate these values based on the number of processors – but not necessary if these are percentages?

 

TIA
Jamie
-- 
To unsubscribe:
https://lists.nongnu.org/mailman/listinfo/monit-general


--
To unsubscribe:
https://lists.nongnu.org/mailman/listinfo/monit-general
Reply | Threaded
Open this post in threaded view
|

RE: loadavg figures - percentages or not?

Jamie Burchell

Hi

 

Thanks for your time and help. I’ve made some changes to my configuration (which we are able to provision with Ansible) and am experimenting with the results:

 

if loadavg (1min) > {{ ansible_processor_count * 2 }} for 4 cycles then alert

if loadavg (5min) > {{ ansible_processor_count * 1.5 }} for 4 cycles then alert

if loadavg (15min) > {{ ansible_processor_count }} for 4 cycles then alert


(processor_count seems to be what we need here as we are using vCPUs)

 

Kind regards,

Jamie

From: monit-general [mailto:[hidden email]=[hidden email]] On Behalf Of [hidden email]
Sent: 08 April 2019 21:54
To: This is the general mailing list for monit <[hidden email]>
Subject: Re: loadavg figures - percentages or not?

 

Hi,

 

the loadavg is not percent - as the manual states, it is absolute value: number of processes in the run queue. The practical limit depends on the number of CPUs and the typical load pattern - a rule of thumb we use is 2 processes per CPU core. If the machine has for example 48 cores, the loadavg of 96 is usually acceptable. There could be also spikes which are common and you may want to suppress false alerts, the example shows setup where high loadavg values for several consecutive cycles are needed before the alert is triggered.

 

To make the configuration easier, i think we can introduce some kind of "per CPU core" load average test, so the configuration will work the same regardless of CPU cores count, something like:

 

            if loadavg(1m) per core > 1.9 then alert

 

Best regards,

Martin

 

 



On 8 Apr 2019, at 15:38, Jamie Burchell <[hidden email]> wrote:

 

Hi there

 

On the Monit documentation on “System resource tests” (https://mmonit.com/monit/documentation/monit.html#System-resource-tests) the “loadavg” figures look like they could be percentages rather than absolute values. Is that correct, or am I misreading?

 

Most of the examples I’ve seen elsewhere for using the “loadavg” checks appear to be using single figures (https://mmonit.com/wiki/Monit/ConfigurationExamples - System Services).

 

If I’ve misunderstood then I’ll probably want to calculate these values based on the number of processors – but not necessary if these are percentages?

 

TIA

Jamie

-- 
To unsubscribe:
https://lists.nongnu.org/mailman/listinfo/monit-general

 


--
To unsubscribe:
https://lists.nongnu.org/mailman/listinfo/monit-general