Hi,
We've noticed an issue with Cisco SG500 switches and Observium, where if you enable stacking mode (the default on the switch), regardless of if it is in a stack or not, the CPU reports a 50% average. Our investigations show it runs at 95% + for a couple of minutes whilst Observium is polling it, then drops back down to <10% when the query finishes. Turning stacking mode off drops the average CPU to <10% and CPU does not ramp up and down.
We raised a case with Cisco TAC, who said this is "expected behaviour":
At every 5 min cycle, the CPU rises to 85- 95% temporarily and then drops down. The reason for this is, Observium is checking for interface status. When switch is in standalone mode the interface count database is the max switch capacity. Example 48 ports = 48 ports plus channel ports ~ 60 ports In stack mode the predefined port count in the data base is around 880 ports. While the switch is responding to SNMP get , it is using up resources and that is why the CPU utilization will be high. While I was testing it, I did not see or encountered any network outage or Web access issues. We have already made an enhancement request for this behaviour, to limit the port count to any stack configuration.
Has anyone else encountered anything similar?
Adam - I realise that Observium is only polling what the switch is reporting, but if Observium can ascertain how many switches are in the stack, would it be possible to limit polling to only query active/connected stack members?
Thanks,
Jamie