#!/usr/bin/env bash
set -euo pipefail

# Watchdog B MVP tri-state checker for OpenClaw main runtime.
# Output (stdout): exactly one token: running | stalled | idle
#
# Heuristic (MVP):
# - If openclaw.pid exists and process is alive => running unless logs are stale.
# - If process alive but log file hasn't changed for STALL_AFTER_SECONDS => stalled.
# - Otherwise => idle.
#
# Future extension point:
# - Replace/augment log-freshness with real main-agent session/ledger signals.

PID_FILE_DEFAULT="${OPENCLAW_PID_FILE:-/home/chchang/.openclaw/workspace/host-runtime/openclaw.pid}"
LOG_FILE_DEFAULT="${OPENCLAW_LOG_FILE:-/home/chchang/.openclaw/workspace/logs/openclaw.log}"

STALL_AFTER_SECONDS="${STALL_AFTER_SECONDS:-1200}"  # 20 minutes default
NOW_EPOCH="$(date +%s)"

pid_file="$PID_FILE_DEFAULT"
log_file="$LOG_FILE_DEFAULT"

get_mtime_epoch() {
  # GNU stat: %Y; BSD stat: -f %m
  local path="$1"
  if stat -c %Y "$path" >/dev/null 2>&1; then
    stat -c %Y "$path"
  else
    stat -f %m "$path"
  fi
}

proc_alive() {
  local pid="$1"
  [[ -n "$pid" ]] || return 1
  [[ "$pid" =~ ^[0-9]+$ ]] || return 1
  kill -0 "$pid" >/dev/null 2>&1
}

# No pid file => idle
if [[ ! -f "$pid_file" ]]; then
  echo "idle"
  exit 0
fi

pid="$(tr -d ' \t\n\r' < "$pid_file" || true)"

# PID file exists but process not alive => idle
if ! proc_alive "$pid"; then
  echo "idle"
  exit 0
fi

# Process alive. If no log file, assume running (can't assess stall)
if [[ ! -f "$log_file" ]]; then
  echo "running"
  exit 0
fi

log_mtime="$(get_mtime_epoch "$log_file")"
age=$(( NOW_EPOCH - log_mtime ))

if (( age > STALL_AFTER_SECONDS )); then
  echo "stalled"
else
  echo "running"
fi