#!/bin/bash
# Healthcheck key public services; notify ntfy on failure transitions.
set -euo pipefail

STATE_DIR=/var/lib/nomadro-health
NOTIFY=/usr/local/bin/notify-ntfy
mkdir -p "$STATE_DIR"

check() {
  local name="$1"
  local cmd="$2"
  local ok=0
  if eval "$cmd" >/dev/null 2>&1; then
    ok=1
  fi
  local state_file="$STATE_DIR/${name}.ok"
  local prev=1
  [[ -f "$state_file" ]] && prev=$(cat "$state_file")
  echo "$ok" >"$state_file"
  if [[ "$ok" -eq 0 && "$prev" -eq 1 ]]; then
    "$NOTIFY" "Service down: $name" "Host $(hostname) — $name failed health check at $(date -u +%Y-%m-%dT%H:%M:%SZ)" high "rotating_light,server" || true
  elif [[ "$ok" -eq 1 && "$prev" -eq 0 ]]; then
    "$NOTIFY" "Service recovered: $name" "Host $(hostname) — $name is healthy again" default "white_check_mark,server" || true
  fi
}

check caddy "systemctl is-active --quiet caddy"
check docker "systemctl is-active --quiet docker"
check nomadro "systemctl is-active --quiet nomadro"
check nomadro-router "systemctl is-active --quiet nomadro-router"
check wordpress "curl -fsS -o /dev/null http://127.0.0.1:8088/"
check umami "curl -fsS -o /dev/null http://127.0.0.1:3010/api/heartbeat"
check ntfy "curl -fsS http://127.0.0.1:2586/v1/health | grep -q healthy"
check dylive "curl -fsS -o /dev/null -H 'Host: dylive.nomadro.com' http://127.0.0.1:4015/"
check nomadro-live "curl -fsS -o /dev/null http://127.0.0.1:3015/"
check wx "curl -fsS -o /dev/null http://127.0.0.1:3000/"
check public-nomadro "curl -fsS -o /dev/null https://nomadro.com/"
check public-umami "curl -fsS -o /dev/null https://umami.nomadro.com/api/heartbeat"
check public-ntfy "curl -fsS https://ntfy.nomadro.com/v1/health | grep -q healthy"