Brief introduction
The recent run of the online business, my colleagues told me that zabbix no warning when the machine cpu usage is very high, idle already low, but the alarm cpu does not come out. Then I started to analyze the cpu triggers an alarm condition, the summary of the analysis process to do next.
First, I see the alarm on the machine trigger condition, trigger configuration is
{base_template:system.cpu.util[,user].count(#3,85,"ge")}>1
The flip-flop indicates that the user space occupied by three percentage over 85% of monitoring an alarm. Then I see how much that value is through the top order on the machine
After reading this chart I found with my perception a little different, and I understand cpus before each process should be occupied by the sum of the fishes do, in fact, not the case.
Look at these two foundation meanings:
Cpu (s): indicates the percentage of the space occupied by a user CPU.
% CPU: Indicates that the last update to the current occupancy percentage of CPU time.
The percentage of each logical look cpu occupied core
In fact Cpu (s) means that all the user processes average for the entire cpu, as a percentage of each different core occupied, so look average.
% CPU shows the percentage occupied by a nuclear process, rather than the entire cpu (16-core) percentage, and sometimes may be greater than 100, it is because the process enables multi-threaded occupies multiple cores, so sometimes we see the worth more than 100% of the time, but will not exceed the total number of cores * 100.
It can be seen, then, indeed monitoring the trigger will not be triggered, in order to monitor the actual data, alarms, so under the supervision of the flip-flop strategy needs to be adjusted, will trigger or add a condition on the trigger before
{base_template:system.cpu.util[,idle].avg(1m)}<15
It is to determine the percentage of free, i.e., less than 15% when the idle over 85% using an alarm notification.
ps: cpu pressure measurement to trigger small case
echo "scale=500; 4*a(1)" | bc -l -q