#!/bin/bash

cd /root/rl-swarm || exit 1

SCRIPT="./run_rl_swarm.sh"
LOG="./rl_swarm.log"
RESTART_INTERVAL=3600  # 1 час = 3600 сек

while true; do
    echo ">>> Cleaning up old processes at $(date)" | tee -a "$LOG"
    pkill -9 -u "$(id -u)" python
    pkill -9 -f libp2p
    sleep 2

    echo ">>> Starting RL Swarm at $(date)" | tee -a "$LOG"
    bash "$SCRIPT" >> "$LOG" 2>&1 &
    PID=$!

    # Таймер перезапуска
    (
      sleep $RESTART_INTERVAL
      echo ">>> [Timer] $RESTART_INTERVAL sec passed. Restarting RL Swarm..." | tee -a "$LOG"
      kill "$PID" 2>/dev/null
    ) &

    # Лог-мониторинг на случай ошибок или выхода
    tail -n 0 -F "$LOG" | while read -r line; do
        echo "$line"

        echo "$line" | grep -q "EOFError"
        if [ $? -eq 0 ]; then
            echo ">>> [Crash] EOFError detected. Restarting..." | tee -a "$LOG"
            kill "$PID" 2>/dev/null
            break
        fi

        if ! kill -0 "$PID" 2>/dev/null; then
            echo ">>> [Exit] RL Swarm stopped. Restarting..." | tee -a "$LOG"
            break
        fi
    done

    echo ">>> Restarting in 5 seconds..." | tee -a "$LOG"
    sleep 5
done