So our machines swap….swap alot. There are the typical nagios system checks, you can have the check_swap call, but at my company the < 20 % hits so often we needed to figure out a way to get warned about crippling proformance. We discovered that when a machine has CPU WAIT spikes, the proformance for the machine tanks, so I tried to find a check for specificly monitor the CPU WAIT. I should mention also that most/all of the boxes we run are VMs, and waiting on IO for swap makes a sad sysadmin.

I took a stab at creating a check myself and the following is a simple nagios check that you can run via NRPE. The PERCENTAGE line is changeable, I’d make sure your alarm reflects your error cases. We choose 10 % because that’s when we started seeing real prefomance issues, but as all sysadmins know any CPU WAIT is bad in general.

CPUWAIT=`top -b -n 1| grep 'wa,' | awk -F ',' {'print $5'} | cut -d . -f 1 | tr -d ' ' `

if [ ${CPUWAIT} -le ${PERCENTAGE} ]
   echo Your CPU WAIT is under ${PERCENTAGE}%
elif [ ${CPUWAIT} -ge ${PERCENTAGE} ]
   echo Your CPU WAIT is at ${CPUWAIT}%, your machine aint happy
exit $stateid

You are probably wondering “There are alot of great systems checks on Nagios Exchange,” but the site is so hard to use and useally much more than I need, like in this case.