#!/usr/bin/env bash
# monitor_rate_limits.sh
# Checks OpenClaw usage log for recent rate_limit / 429 entries and acts.

# Use user-local paths to avoid sudo permissions
MONITOR_DIR=/home/alex/clawd/monitor
LOG=${MONITOR_DIR}/usage.log
COUNT_FILE=${MONITOR_DIR}/openclaw_rate_limit_count
THRESHOLD=2
WINDOW_MIN=10
DISCORD_USER=231473537649475588

mkdir -p "$MONITOR_DIR"
: > "${MONITOR_DIR}/monitor_last" || true

# Primary source: existing local DB/collected metrics (preferred)
# Fallback: global /var/log/openclaw/usage.log if present
SRC_LOG=/var/log/openclaw/usage.log

# Helper to send Discord alert via OpenClaw CLI (if available)
send_discord_alert(){
  local msg="$1"
  # Try OpenClaw message send (best effort)
  if command -v openclaw >/dev/null 2>&1; then
    openclaw message send --channel discord --to "$DISCORD_USER" --message "$msg" >/dev/null 2>&1 || true
  fi
}

# Look for rate_limit/429 in collected metrics DB first
FOUND=0
if command -v sqlite3 >/dev/null 2>&1 && [ -f ~/clawd/data/usage.db ]; then
  lines=$(sqlite3 ~/clawd/data/usage.db "SELECT value,extra FROM metrics WHERE key LIKE '%logs%' ORDER BY ts DESC LIMIT 5;" 2>/dev/null || true)
  if echo "$lines" | grep -E "rate_limit|429" >/dev/null 2>&1; then
    FOUND=1
  fi
fi

# If not found in DB, search the system log (fallback)
if [ $FOUND -eq 0 ] && [ -f "$SRC_LOG" ]; then
  if grep -E "rate_limit|429" "$SRC_LOG" --binary-files=without-match >/dev/null 2>&1; then
    FOUND=1
    RECENTS=$(grep -E "rate_limit|429" "$SRC_LOG" | tail -n 50)
  fi
fi

# Also check local monitor log for quick indicators
if [ $FOUND -eq 0 ] && [ -f "$LOG" ]; then
  if grep -E "rate_limit|429" "$LOG" >/dev/null 2>&1; then
    FOUND=1
    RECENTS=$(grep -E "rate_limit|429" "$LOG" | tail -n 50)
  fi
fi

# If still not found, try journalctl for openclaw unit
if [ $FOUND -eq 0 ] && command -v journalctl >/dev/null 2>&1; then
  RECENTS=$(journalctl -u openclaw --no-pager --since "${WINDOW_MIN} minutes ago" | grep -E "rate_limit|429" || true)
  if [ -n "$RECENTS" ]; then
    FOUND=1
  fi
fi

if [ $FOUND -eq 1 ]; then
  prev=0
  [ -f "$COUNT_FILE" ] && prev=$(cat "$COUNT_FILE" 2>/dev/null || echo 0)
  cur=$((prev+1))
  echo "$cur" > "$COUNT_FILE"
  echo "$(date -Iseconds) Detected rate_limit/429 entries (count=$cur)." >> "$LOG"
  echo "$RECENTS" >> "$LOG"

  # provider-specific detection
  alert_msg="⚠️ Rate limit hit - switching to fallback"
  if echo "$RECENTS" | grep -qi "anthropic"; then
    alert_msg="⚠️ Anthropic rate limit hit - switching to fallback"
  elif echo "$RECENTS" | grep -Ei "gemini|google" >/dev/null 2>&1; then
    alert_msg="⚠️ Gemini rate limit hit"
  elif echo "$RECENTS" | grep -qi "openrouter"; then
    alert_msg="⚠️ OpenRouter rate limit hit"
  elif echo "$RECENTS" | grep -Ei "copilot|github-copilot" >/dev/null 2>&1; then
    alert_msg="⚠️ Copilot rate limit hit"
  fi

  if [ "$cur" -ge "$THRESHOLD" ]; then
    echo "Threshold reached: restarting OpenClaw gateway and logging alert." >> "$LOG"
    if command -v openclaw >/dev/null 2>&1; then
      openclaw gateway restart || echo "gateway restart failed" >> "$LOG"
    fi
    echo "$(date -Iseconds) ALERT: $alert_msg" >> "$LOG"
    send_discord_alert "$alert_msg"
    # reset counter
    echo "0" > "$COUNT_FILE"
  fi
else
  echo "No recent rate_limit/429 entries found." >> "$LOG"
  echo "0" > "$COUNT_FILE"
fi