check_docker_cpu.sh

check_docker_cpu.sh

#!/usr/bin/env bash

#############################################################
# Check the cpu consumption of a running docker container
# by reading cgroup information
# 
# Licence: LGPL
# Author: Christoph Wiechert <[email protected]>
#############################################################

# Set States
STATE_OK=0
STATE_WARNING=1
STATE_CRITICAL=2
STATE_UNKNOWN=3

SCRIPTNAME=$0
VERSION=1.0.0

TMPDIR=/run/check_docker_cpu

# Set help
print_help () {
  echo "Usage: $SCRIPTNAME -n <containerName> -i <containerID> [-w <warning>] [-c <critical>]"
  echo ""
  echo "This plugin checks cpu consumption for a running docker container."
  echo ""
  echo "Provide either a -n OR -i"
  echo "Warn and crit thresholds are relative values in percent."
  echo "Omit -w and -c to return OK for every value"
  echo ""
  echo "Options:"
  echo "  -h                   Prints this helpscreen"
  echo "  -v                   Prints the version"
  echo "  -n <containerName>   Name of the running docker container"
  echo "  -i <containerId>     CID"
  echo "  -w <warning>         Warning threshold in percentage"
  echo "  -c <critical>        Critical threshold in percentage"
  echo "  -ignoreCores         Ignore the vCPU number"
  echo ""
}

# Parse CLI args
while [[ $# > 0 ]]; do
  key=$1
  case "$key" in
    -h|--help)
      print_help
      exit 0
    ;;
    -v|--version)
      echo $VERSION
      exit 0
    ;;
    -ignoreCores)
    echo IGNORE
      IGNORE_CORES="1"
    ;;
    -n)
      CONTAINER_NAME="$2"
      shift
    ;;
    -i)
      CONTAINER_ID="$2"
      shift
    ;;
    -w)
      WARN="$2"
      shift
    ;;
    -c)
      CRIT="$2"
      shift
    ;;
    *)
      >&2 echo "Ignoring unknowen option $key"
    ;;
  esac
  shift # past argument or value
done

mkdir -p $TMPDIR
if [ ! -w $TMPDIR ]; then
 >&2 echo "Error: $TMPDIR not writeable"
 exit $STATE_UNKNOWN
fi

if [ "$CONTAINER_NAME" = "" ] && [ "$CONTAINER_ID" = "" ]; then
  >&2 echo "Error: Either a ContainerName (-n) or a ContainerID (-i) is mandatory."
  print_help
  exit $STATE_UNKNOWN
fi

# Resolve ContainerName to CID
if [ "$CONTAINER_ID" = "" ]; then
  CONTAINER_ID=$(docker ps | grep -E "$CONTAINER_NAME[ ]*$" | cut -d " " -f 1)
  if [ "$CONTAINER_ID" == "" ] ; then
    >&2 echo "Error: Could not find CID for $CONTAINER_NAME. Is the container running?"
    exit $STATE_UNKNOWN
  fi
fi

# Read the values
NOW=$(expr `date +"%s"` \* 1000000000 + `date +"%N"`)
USAGE=`cat /sys/fs/cgroup/cpu,cpuacct/system.slice/docker-$CONTAINER_ID*.scope/cpuacct.usage`

if [ "$USAGE" == "" ] ; then
  >&2 echo "Error: Could not read cgroup values for ContainerID $CONTAINER_ID. Is the container running?"
  exit $STATE_UNKNOWN
fi

# Load values from last run
[ -e "$TMPDIR/cid-$CONTAINER_ID" ] && source $TMPDIR/cid-$CONTAINER_ID

# Save current values
echo "LAST_NOW=$NOW" > $TMPDIR/cid-$CONTAINER_ID
echo "LAST_USAGE=$USAGE" >> $TMPDIR/cid-$CONTAINER_ID

if [ "$LAST_NOW" == "" ]; then
  # This is the first run for this container, we cant do anything
  echo "First run"
  exit $STATE_OK
fi

# Caluclation
ELAPSED_TIME=$(expr $NOW - $LAST_NOW)
DELTA=$(expr $USAGE - $LAST_USAGE)
PERCENT=$(expr $DELTA \* 100 / $ELAPSED_TIME)

if [ "$PERCENT" == "" ]; then
  >&2 echo "Error: Unknown error"
  exit $STATE_UNKNOWN
fi

MAX=100
if [ "$IGNORE_CORES" != "1" ]; then
  PERCENT=$(expr $PERCENT / `nproc`)
else
  MAX=$(expr `nproc` \* 100)
fi


echo "CPU usage: ${PERCENT}% | CpuUsage=${PERCENT}%;$WARN;$CRIT;0;$MAX"

[ "$DONTWARN" == "1" ] && exit $STATE_OK

[ $PERCENT -gt $CRIT ] && exit $STATE_CRITICAL
[ $PERCENT -gt $WARN ] && exit $STATE_WARNING

exit $STATE_OK