#!/usr/bin/env bash
set -u

SERVICE_NAME="salt-minion"
STATE_FILE="/opt/.saltrunning"
WARN_AGE=5400
CRIT_AGE=7200

usage() {
  cat <<'USAGE'
Usage: check_salt_running.sh [-s service] [-f state_file] [-w warn_age] [-c crit_age]

Checks that the salt-minion service is active and that the state file has been
updated recently. If Salt is not installed or the unit is unavailable, the
check reports WARNING instead of CRITICAL.
USAGE
}

while getopts ":s:f:w:c:h" opt; do
  case "$opt" in
    s) SERVICE_NAME="$OPTARG" ;;
    f) STATE_FILE="$OPTARG" ;;
    w) WARN_AGE="$OPTARG" ;;
    c) CRIT_AGE="$OPTARG" ;;
    h)
      usage
      exit 0
      ;;
    \?)
      echo "UNKNOWN - invalid option: -$OPTARG"
      usage
      exit 3
      ;;
    :)
      echo "UNKNOWN - option -$OPTARG requires an argument"
      usage
      exit 3
      ;;
  esac
done

WARN_AGE_MINUTES=$(( WARN_AGE / 60 ))
CRIT_AGE_MINUTES=$(( CRIT_AGE / 60 ))

if ! command -v systemctl >/dev/null 2>&1; then
  echo "UNKNOWN - systemctl not available"
  exit 3
fi

service_load_state="$(systemctl show -p LoadState --value "$SERVICE_NAME" 2>/dev/null || true)"
service_load_state="${service_load_state#LoadState=}"
if [[ -z "$service_load_state" || "$service_load_state" == "not-found" ]]; then
  echo "WARNING - $SERVICE_NAME is not installed or the unit is unavailable"
  exit 1
fi

service_state="$(systemctl is-active "$SERVICE_NAME" 2>/dev/null || true)"
if [[ "$service_state" != "active" ]]; then
  echo "CRITICAL - $SERVICE_NAME is not active (state: ${service_state:-unknown})"
  exit 2
fi

if [[ ! -e "$STATE_FILE" ]]; then
  echo "CRITICAL - $STATE_FILE does not exist"
  exit 2
fi

if ! last_run="$(stat -c '%Y' "$STATE_FILE" 2>/dev/null)"; then
  echo "UNKNOWN - unable to read mtime for $STATE_FILE"
  exit 3
fi

now="$(date +%s)"
age=$(( now - last_run ))
age_minutes=$(( age / 60 ))

if (( age < 0 )); then
  echo "UNKNOWN - $STATE_FILE has a future timestamp"
  exit 3
fi

if (( age >= CRIT_AGE )); then
  echo "CRITICAL - Salt has not updated for ${age_minutes} minutes (threshold ${CRIT_AGE_MINUTES} minutes) | saltrunning_age=${age_minutes}m;${WARN_AGE_MINUTES};${CRIT_AGE_MINUTES};0;"
  exit 2
fi

if (( age >= WARN_AGE )); then
  echo "WARNING - Salt has not updated for ${age_minutes} minutes (threshold ${WARN_AGE_MINUTES} minutes) | saltrunning_age=${age_minutes}m;${WARN_AGE_MINUTES};${CRIT_AGE_MINUTES};0;"
  exit 1
fi

echo "OK - $SERVICE_NAME active, last update ${age_minutes} minutes ago | saltrunning_age=${age_minutes}m;${WARN_AGE_MINUTES};${CRIT_AGE_MINUTES};0;"
exit 0