check_dockerapi_stats

#!/bin/bash

# See https://docs.docker.com/engine/api/v1.40/#operation/ContainerStats

JQ=`which jq`
CURL=`which curl`
AWK=`which awk`

# Nagios return codes
STATE_OK=0
STATE_WARNING=1
STATE_CRITICAL=2
STATE_UNKNOWN=3

# Plugin variable description
PROGNAME=$(basename $0)
RELEASE="Revision 1.0.1"
AUTHOR="(c) 2020 Paul Bargewell (paul.bargewell@opusvl.com)"

if [ ! -x $JQ ]; then
  echo "UNKNOWN: jq not found or is not executable"
  exit $STATE_UNKNOWN
fi

if [ ! -x $CURL ]; then
  echo "UNKNOWN: curl not found or is not executable"
  exit $STATE_UNKNOWN
fi

if [ ! -x $AWK ]; then
  echo "UNKNOWN: awk not found or is not executable"
  exit $STATE_UNKNOWN
fi

# Functions plugin usage
print_release() {
    echo "$RELEASE $AUTHOR"
}

print_usage() {
    echo ""
    echo "$PROGNAME $RELEASE - Checks Docker Container Status from Docker API for Icinga2"
    echo ""
    echo "Usage: $PROGNAME"
    echo ""
    echo "  -h               Show this page"
    echo ""
    echo "  -H | --host      Host name / IP of Docker API server (required)"
    echo ""
    echo "  -P | --port      Port Number of Docker API Service"
    echo ""
    echo "  --https      Use HTTPS requires cert, key and cacert set"
    echo ""
    echo "    --cert       Client certificate (pem)"
    echo ""
    echo "    --key        Client key (pem)"
    echo ""
    echo "    --cacert     CA cert (pem)"
    echo ""
    echo "  -C | --container Name of container to inspect"
    echo ""
    echo "Usage: $PROGNAME --help"
    echo ""
    exit 0
}

print_help() {
	print_usage
  echo ""
  echo "This plugin will check Docker container statuses using the Docker API"
  echo ""
	exit 0
}

# Defaults
PROTOCOL=http://
PORT=2376
EXCLUDE=
INCLUDE=

# Parse parameters
while [ $# -gt 0 ]; do
  case "$1" in
    -h | --help)
      print_help
      exit $STATE_OK
      ;;
    -v | --version)
      print_release
      exit $STATE_OK
      ;;
    -H | --host)
      shift
      HOST=$1
      ;;
    -P | --port)
      shift
      PORT=$1
      ;;
    -C | --container)
      shift
      CONTAINER=$1
      ;;
    --exclude)
      shift
      EXCLUDE=$1
      ;;
    --include)
      shift
      INCLUDE=$1
      ;;      
    --https)
      PROTOCOL=https://
      ;;            
    --cert)
      shift
      CERT=$1
      ;;
    --key)
      shift
      KEY=$1
      ;;
    --cacert)
      shift
      CACERT=$1
      ;;      
    *)
      echo "Unknown argument: $1"
      print_usage
      exit $STATE_UNKNOWN
      ;;
  esac
  shift
done

if [ -z "$HOST" ]; then
  echo "UNKNOWN: Host name must be specified"
  exit $STATE_UNKNOWN
fi

TLS_OPTIONS=

if [ "$PROTOCOL" == "https://" ]; then
  # Check all cert info is set
  if [[ -z "$CERT" || -z "$KEY" || -z "$CACERT" ]]; then
    echo "UNKNOWN: Missing client certificate, key or ca cert."
    exit $STATE_UNKNOWN
  else
    TLS_OPTIONS="--cert $CERT --key $KEY --cacert $CACERT"
  fi
fi

# Get the ID of the container for the next call to stats
ID=$(eval "${CURL} --silent ${PROTOCOL}${HOST}:${PORT}/containers/json?all=true $TLS_OPTIONS | ${JQ} -r '.[] | select(.Names[0] == \"/${CONTAINER}\") .Id'")

# Make a one time (not streaming) call to stats
JSON=`${CURL} --silent ${PROTOCOL}${HOST}:${PORT}/containers/${ID}/stats?stream=false $TLS_OPTIONS`

if [ -z "$JSON" ]; then
  echo "UNKNOWN: Unable to collect container data from the docker API"
  exit $STATE_UNKNOWN
fi

NAGIOS_DATA=

memory_stats_usage=`echo ${JSON} | ${JQ} .memory_stats.usage`
memory_stats_stats_cache=`echo ${JSON} | ${JQ} .memory_stats.stats.cache`
used_memory=($memory_stats_usage - $memory_stats_stats_cache)

available_memory=`echo ${JSON} | ${JQ} .memory_stats.limit`

memory_usage=`echo "${used_memory} ${available_memory}" | ${AWK} '{printf "%.2f", $1 / $2 * 100}'`
memory_usage_d=`echo "${used_memory} ${available_memory}" | ${AWK} '{printf "%d", $1 / $2 * 100}'`

cpu_stats_cpu_usage_total_usage=`echo ${JSON} | ${JQ} .cpu_stats.cpu_usage.total_usage`
precpu_stats_cpu_usage_total_usage=`echo ${JSON} | ${JQ} .precpu_stats.cpu_usage.total_usage`

cpu_delta=`echo "${cpu_stats_cpu_usage_total_usage} ${precpu_stats_cpu_usage_total_usage}" | ${AWK} '{printf "%d", $1 - $2}'`

cpu_stats_system_cpu_usage=`echo ${JSON} | ${JQ} .cpu_stats.system_cpu_usage`
precpu_stats_system_cpu_usage=`echo ${JSON} | ${JQ} .precpu_stats.system_cpu_usage`

system_cpu_delta=`echo "${cpu_stats_system_cpu_usage} ${precpu_stats_system_cpu_usage}" | ${AWK} '{printf "%d", $1 - $2}'`

number_cpus=`echo ${JSON} | ${JQ} .cpu_stats.online_cpus`

cpu_usage=`echo "${cpu_delta} ${system_cpu_delta} ${number_cpus}" | ${AWK} '{printf "%f", ($1 / $2) * 100}'`
cpu_usage_d=`echo "${cpu_delta} ${system_cpu_delta} ${number_cpus}" | ${AWK} '{printf "%d", ($1 / $2) * 100}'`

NAGIOS_DATA="${NAGIOS_DATA}cpu_usage=${cpu_usage}%, memory_usage=${memory_usage}%, number_cpus=${number_cpus}" 


# Use the integer (non-decimal) version for the comparison
if [ $cpu_usage_d -gt 90 ]; then
  echo "DOCKER CONTAINER CRITICAL: ${NAGIOS_DATA}"
  exit $STATE_CRITICAL
elif [ $cpu_usage_d -gt 50 ]; then
  echo "DOCKER CONTAINER WARNING: ${NAGIOS_DATA}"
  exit $STATE_WARNING
else
  echo "DOCKER CONTAINER OK: ${NAGIOS_DATA}"
  exit $STATE_OK
fi