Calculate CPU utilization [Prometheus]

1.CPU Usage formula

We know
CPU The usage rate of = ( All non idle states CPU Total usage time )/( All States CPU The sum of time )
By the following formula ：
CPU Total usage time = All non idle states CPU Total usage time + All idle states CPU Total usage time
obtain ：
CPU The usage rate of = 1 - ( All idle states CPU Total usage time )/( All States CPU The sum of time )

2.prometheus Calculation CPU Usage rate

CPU Total usage time :node_cpu_seconds_total, It means all kinds of occupied time .

2.1 CPU Use your time 5m Incremental value

5m Incremental value = Current sampling point value - 5m Previous sampling point value

Input ： increase(node_cpu_seconds_total[5m])

2.2 CPU Free time 5m Incremental value

node_cpu_seconds_total Sample data of , It is based on the implementation sequence key/vlue Format , and key yes "mode" The sampling is recorded CPU The time of that state (user、sys、idle etc. )
Input ：increase(node_cpu_seconds_total{mode=“idle”}[5m])

2.3 Polymerized multicore CPU Free time 5m Incremental value

By sampling data , You can see CPU Free time 5m Incremental value , Is each cpu And are taken as a sampling value , And every virtual machine \ The physical machine , Most of them are multicore , So we need to aggregate the idle time of all cores .

Input ：sum(increase(node_cpu_seconds_total{mode=“idle”}[5m]))

It can be seen from the results ,sum The function is to put all CPU Audit time wants to add , No aggregation by host , Therefore, it is necessary to introduce =by (instance)= function , It will bring sum The summed values are split in the specified way ,instance It represents the machine name . If you don't write by (instance) You need to be in {} Which instance of the data is required .

Input ：sum(increase(node_cpu_seconds_total{mode=“idle”}[5m])) by (instance)

2.4 Aggregate single machine multicore CPU Total usage time 5m Incremental value

Input ：sum(increase(node_cpu_seconds_total[5m])) by (instance)

2.5 Calculate the average 5 minute CPU Usage rate

Input ：1- sum(increase(node_cpu_seconds_total{mode=“idle”}[5m])) by (instance)/sum(increase(node_cpu_seconds_total[5m])) by (instance)