|Subject:||Re: Monitoring load average relative to number of CPUs|
|Date:||Thu, 10 Apr 2014 14:28:28 -0400|
We get around this issue by generating our configs dynamically from puppet. Here's the ERB excerpt:
if loadavg (1min) > <%= scope.lookupvar('processorcount').to_i * 2 %> for 5 cycles then alert
if loadavg (5min) > <%= scope.lookupvar('processorcount').to_i * 1.5 %> for 5 cycles then alert
if loadavg (15min) > <%= scope.lookupvar('processorcount').to_i %> for 5 cycles then alert
Cheers - Callum.--
On 10/04/2014 20:16, Martin Pala wrote:
the load average value is currently absolute only. The rule of thumb usually was that load average of 2 per CPU core is acceptable, which translates to 2 runnable processes per CPU. Relative load average settings may be useful - load of 1 per each CPU would correspond to 100%, i.e. on 16 cores system load 16 = 100%, to test for load corresponding to 32 on such system the value will be 200%. The percentage would work regardless of CPU count (for example in the case of single CPU the load 100% will correspond to load average 1).
We can maybe add such option in the future.
On 10 Apr 2014, at 18:50, Daniel Levine <address@hidden> wrote:
Is it possible to have monit alert when load average exceeds the number of CPUs, i.e. without having to set a specific number? This would be convenient for virtual machines. The documentation seems a little vague on the subject though.
I did try setting load average as a percentage, but that resulted in a syntax error.
Thanks in advance!
DE mobile: +49 1573 245 1224
UK mobile: +44 7968 378 810
Desk: +44 131 471 7572
For avoidance of doubt: This email does not constitute permission to add me to your mailing list.
|[Prev in Thread]||Current Thread||[Next in Thread]|