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

WORKSPACE="/home/openclaw/.openclaw/workspace"
LOG="${WORKSPACE}/logs/doctorbot.log"
STATE_DIR="${WORKSPACE}/memory"
STATE_FILE="${STATE_DIR}/doctorbot-state.json"
MEM_CHECK_SCRIPT="${WORKSPACE}/scripts/memory-integrity-check.sh"
MEM_RESTORE_SCRIPT="${WORKSPACE}/scripts/memory-restore-from-db.sh"

mkdir -p "${WORKSPACE}/logs" "${STATE_DIR}"

log(){
  echo "[$(date -u +'%F %T UTC')] $*" >> "$LOG"
}

read_fail_count(){
  if [[ -f "$STATE_FILE" ]]; then
    python3 - <<'PY' "$STATE_FILE"
import json,sys
p=sys.argv[1]
try:
  d=json.load(open(p,'r',encoding='utf-8'))
  print(int(d.get('fail_count',0)))
except Exception:
  print(0)
PY
  else
    echo 0
  fi
}

write_state(){
  local count="$1"
  cat > "$STATE_FILE" <<EOF
{
  "fail_count": $count,
  "updated_utc": "$(date -u +'%Y-%m-%dT%H:%M:%SZ')"
}
EOF
}

# Memory integrity guard + auto-restore from external DB.
if ! "$MEM_CHECK_SCRIPT" >/dev/null 2>&1; then
  log "warn: memory integrity check failed"
  if "$MEM_RESTORE_SCRIPT" >> "$LOG" 2>&1; then
    log "ok: memory restore completed after failed integrity check"
  else
    rc=$?
    log "error: memory restore failed (code=${rc})"
  fi
fi

healthy=0
if openclaw status >/dev/null 2>&1; then
  healthy=1
fi

fail_count=$(read_fail_count)

if [[ "$healthy" -eq 1 ]]; then
  if [[ "$fail_count" -ne 0 ]]; then
    log "ok: status healthy again, reset fail counter"
  fi
  write_state 0
  exit 0
fi

fail_count=$((fail_count+1))
write_state "$fail_count"
log "warn: health check failed (consecutive=${fail_count})"

# 1st failure: no action, just observe
if [[ "$fail_count" -lt 2 ]]; then
  exit 0
fi

# 2nd+ consecutive failure: try gentle recovery once per run
log "action: openclaw gateway restart"
openclaw gateway restart >/dev/null 2>&1 || true
sleep 4

if openclaw status >/dev/null 2>&1; then
  log "ok: recovered after gateway restart"
  write_state 0
  exit 0
fi

log "action: systemctl --user restart openclaw-gateway.service"
systemctl --user restart openclaw-gateway.service >/dev/null 2>&1 || true
sleep 4

if openclaw status >/dev/null 2>&1; then
  log "ok: recovered after user service restart"
  write_state 0
  exit 0
fi

log "error: unrecovered (consecutive=${fail_count})"
exit 0