icinga2-checks/check_systemd_service.sh

#!/bin/bash

function usage {
  echo "Usage:
  -u [UNIT name]
  -t Service is triggered by a timer or is a oneshot service. Is allowed to be inactive
  -d A service is allowed to not be enabled"
}

UNIT_NAME=""
IS_TIMER=false
IS_ALLOWED_DISABLED=false

# Parse command line arguments
while getopts "u:t" opt; do
  case ${opt} in
    u )
      UNIT_NAME=$OPTARG
      ;;
    t )
      IS_TIMER=true
      ;;
    t )
      IS_ALLOWED_DISABLED=true
      ;;
    \? )
      echo "Invalid option: $OPTARG" 1>&2
      exit 1
      ;;
    : )
      echo "Invalid option: $OPTARG requires an argument" 1>&2
      exit 1
      ;;
  esac
done
shift $((OPTIND -1))

# Check if UNIT name is provided
if [ -z "$UNIT_NAME" ]; then
    usage
    exit -1
fi

# Check if the unit is enabled
enabled=$(systemctl is-enabled "$UNIT_NAME")
if [ "$enabled" != "enabled" ] && [ "$IS_ALLOWED_DISABLED" = false ]; then
    echo "CRITICAL - $UNIT_NAME is not enabled"
    exit 2
fi

# Check if the unit is active
active=$(systemctl is-active "$UNIT_NAME")
substate=$(systemctl show "$UNIT_NAME" --property=SubState --value)
exit_code=$(systemctl show "$UNIT_NAME" --property=ExecMainStatus --value)

if [ "$active" = "activating" ]; then
    echo "WARNING - $UNIT_NAME is activating"
    exit 1
elif [ "$active" != "active" ] && [ "$IS_TIMER" = false ]; then
    echo "CRITICAL - $UNIT_NAME is not active"
    exit 2
elif [ "$IS_TIMER" = true ] && [ "$active" = "inactive" ] && [ "$exit_code" != "0" ]; then
    echo "CRITICAL - $UNIT_NAME is inactive with non-zero exit code"
    exit 2
fi

# Get service start time
start_time=$(systemctl show "$UNIT_NAME" --property=ExecMainStartTimestamp --value)
end_time=$(systemctl show "$UNIT_NAME" --property=ExecMainExitTimestamp --value)
current_time=$(date +%s)
start_time_diff=$((current_time - $(date -d "$start_time" +%s)))
end_time_diff=$((current_time - $(date -d "$end_time" +%s)))

if [ "$IS_TIMER" = true ] && { [ "$substate" = "exited" ] || [ "$active" = "inactive" ]; }; then
    echo "OK - $UNIT_NAME is active and exited (returned $exit_code) and enabled. Exited at $end_time. | downtime=${end_time_diff}s"
else
    echo "OK - $UNIT_NAME is active and enabled. Started at $start_time. | uptime=${start_time_diff}s"
fi
exit 0