Run a command while CPU is under 80%

Question

I'm trying to run a command every 10 minutes while usage of CPU is under 80%. I'm not used to bash scripting. I've found a way to get the total use of CPU ``top -b -n1 | grep "Cpu(s)" | awk '{print $2 + $4}'` and then I want to compare with a maximum of 80%. But I'm not sure why I cannot compare both. Here the script:

#!/bin/bash

MAX=80
CPU=`top -b -n1 | grep "Cpu(s)" | awk '{print $2 + $4}'`

while true
  do
    if $CPU < $MAX
      then
        echo $CPU  # This is an example
        CPU=`top -b -n1 | grep "Cpu(s)" | awk '{print $2 + $4}'`
        sleep 10m
      fi
    done

The script fails because interpret the numbers as a command... (cannot open 80: No such file) Why I cannot compre $MAX and $CPU? and how would you tackle this?

Thanks in advance.

Answer 1

< is used for redirections. If you need to compare integers:

if [ "$MAX" -gt "$CPU" ]

But $CPU can also be float and bash cannot operate on floats directly so you have to use bc:

if [ "$(bc -l <<< "${CPU} > ${MAX}")" -eq 1 ]

You should also use $(..) form instead of legacy backticks and you could assign CPU only once within while loop:

#!/bin/bash

MAX=80

while true
do
    CPU="$(top -b -n1 | grep "Cpu(s)" | awk '{print $2 + $4}')"

    if [ "$(bc -l <<< "${CPU} > ${MAX}")" -eq 1 ]
    then
        echo "$CPU"  # This is an example
        sleep 10m
    fi
done

Answer 2

top(1) is not the best way to solve this problem. It's a rather heavyweight measuring tool, scanning far more than you need. Besides, it's focused on instantaneous values, which isn't the right thing here.

Instead, you should use the system load averages because they're computed continuously over long periods in the kernel. This means they carry no extra cost, thus don't impact your measurement, and they aren't affected by instantaneous changes...like calling top(1)!

Here's your script rewritten in terms of load averages:

#!/bin/bash
HIGH_LOAD=6      # 80% of an 8-core system; see link above

while true
  do
    CURR_LOAD=$(cut -f2 -d' ' < /proc/loadavg)
    if [ $CURR_LOAD -lt $HIGH_LOAD ]
      then
        run-background-command
      fi
    sleep 10m
  done

Note that the sleep call is moved outside the if condition: you want to delay regardless. Maybe you want to add an else instead so the delay is different for the low and high load cases.

Also unlike your version, we don't repeat the CPU usage check inside the loop and outside.