Kiwy-Signage/start.sh

#!/bin/bash

# Kivy Signage Player Startup Script with Watchdog
# This script monitors and auto-restarts the player if it crashes

# Get the directory where this script is located
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"

# Configuration
MAX_RETRIES=999999  # Effectively unlimited retries
RESTART_DELAY=5     # Seconds to wait before restart
HEALTH_CHECK_INTERVAL=30  # Seconds between health checks
HEARTBEAT_FILE="$SCRIPT_DIR/.player_heartbeat"
STOP_FLAG_FILE="$SCRIPT_DIR/.player_stop_requested"
LOG_FILE="$SCRIPT_DIR/player_watchdog.log"

# Ensure log file is writable
if [ ! -w "$(dirname "$LOG_FILE")" ]; then
    LOG_FILE="/tmp/kivy-player-watchdog.log"
fi

# Function to log messages (MUST be defined before use)
log_message() {
    local msg="[$(date '+%Y-%m-%d %H:%M:%S')] $1"
    echo "$msg"
    # Try to write to log file, ignore errors if permission denied
    echo "$msg" >> "$LOG_FILE" 2>/dev/null || true
}

# Load user's environment from systemd user session (most reliable method)
# This ensures we get the proper DISPLAY/WAYLAND_DISPLAY and session variables
load_user_environment() {
    # Try to get environment from active user session via systemctl
    local user_env
    if command -v systemctl &>/dev/null; then
        user_env=$(systemctl --user show-environment 2>/dev/null)
        if [ -n "$user_env" ]; then
            # Extract display-related variables
            echo "$user_env" | grep -E '^(DISPLAY|WAYLAND_DISPLAY|XDG_RUNTIME_DIR|DBUS_SESSION_BUS_ADDRESS)=' | while read -r line; do
                export "$line"
            done
            log_message "Loaded user environment from systemctl --user"
            return 0
        fi
    fi
    
    # Fallback: detect display manually
    log_message "Falling back to manual display detection..."
    
    if [ -z "$DISPLAY" ] && [ -z "$WAYLAND_DISPLAY" ]; then
        # Try to detect Wayland display
        if [ -S "/run/user/$(id -u)/wayland-0" ]; then
            export WAYLAND_DISPLAY=wayland-0
            log_message "Detected Wayland display: $WAYLAND_DISPLAY"
        # Try to detect X11 display
        elif [ -S "/tmp/.X11-unix/X0" ]; then
            export DISPLAY=:0
            log_message "Detected X11 display: $DISPLAY"
        else
            # Wait for display to come up (useful for systemd or delayed starts)
            log_message "Waiting for display server to be ready (up to 30 seconds)..."
            for i in {1..30}; do
                sleep 1
                if [ -S "/run/user/$(id -u)/wayland-0" ] 2>/dev/null; then
                    export WAYLAND_DISPLAY=wayland-0
                    log_message "Display server detected on attempt $i: Wayland"
                    return 0
                elif [ -S "/tmp/.X11-unix/X0" ] 2>/dev/null; then
                    export DISPLAY=:0
                    log_message "Display server detected on attempt $i: X11"
                    return 0
                fi
            done
        fi
    fi
    
    # Verify we have a display now
    if [ -z "$DISPLAY" ] && [ -z "$WAYLAND_DISPLAY" ]; then
        log_message "WARNING: No display server detected. This may cause graphics issues."
        return 1
    fi
    
    log_message "Display environment ready: DISPLAY=$DISPLAY WAYLAND_DISPLAY=$WAYLAND_DISPLAY"
    return 0
}

# Load the user environment early
load_user_environment

# Function to check if player is healthy
check_health() {
    # Check if heartbeat file exists and is recent (within last 60 seconds)
    if [ -f "$HEARTBEAT_FILE" ]; then
        local last_update=$(stat -c %Y "$HEARTBEAT_FILE" 2>/dev/null || echo 0)
        local current_time=$(date +%s)
        local diff=$((current_time - last_update))
        
        if [ $diff -lt 60 ]; then
            return 0  # Healthy
        else
            log_message "⚠️  Player heartbeat stale (${diff}s old)"
            return 1  # Unhealthy
        fi
    else
        # If heartbeat file doesn't exist yet, assume player is starting
        return 0
    fi
}

# Cleanup function
cleanup() {
    log_message "🛑 Watchdog received stop signal"
    rm -f "$HEARTBEAT_FILE"
    rm -f "$STOP_FLAG_FILE"
    exit 0
}

# Trap signals for graceful shutdown
trap cleanup SIGINT SIGTERM

log_message "=========================================="
log_message "🚀 Kivy Signage Player Watchdog Started"
log_message "=========================================="
log_message "Project directory: $SCRIPT_DIR"
log_message "Max retries: Unlimited"
log_message "Restart delay: ${RESTART_DELAY}s"
log_message ""

# Remove old stop flag if exists (fresh start)
rm -f "$STOP_FLAG_FILE"

# Change to the project directory
cd "$SCRIPT_DIR"

# Check if virtual environment exists
if [ -d ".venv" ]; then
    log_message "✓ Virtual environment found"
    source .venv/bin/activate
else
    log_message "⚠️  Creating virtual environment..."
    python3 -m venv .venv
    source .venv/bin/activate
    log_message "📦 Installing dependencies..."
    pip3 install -r requirements.txt
    log_message "✓ Virtual environment ready"
fi

# Check if configuration exists
if [ ! -f "config/app_config.json" ]; then
    log_message "⚠️  WARNING: Configuration file not found!"
    log_message "Player may not function correctly without configuration"
fi

# Main watchdog loop
retry_count=0
while true; do
    retry_count=$((retry_count + 1))
    
    log_message ""
    log_message "=========================================="
    log_message "▶️  Starting player (attempt #${retry_count})"
    log_message "=========================================="
    
    # Clean old heartbeat
    rm -f "$HEARTBEAT_FILE"
    
    # Start the player
    cd "$SCRIPT_DIR/src"
    python3 main.py &
    PLAYER_PID=$!
    
    log_message "Player PID: $PLAYER_PID"
    
    # Monitor the player
    while true; do
        sleep $HEALTH_CHECK_INTERVAL
        
        # Check if process is still running
        if ! kill -0 $PLAYER_PID 2>/dev/null; then
            log_message "❌ Player process crashed or stopped (PID: $PLAYER_PID)"
            break
        fi
        
        # Check health via heartbeat
        if ! check_health; then
            log_message "❌ Player health check failed - may be frozen"
            kill $PLAYER_PID 2>/dev/null
            sleep 2
            kill -9 $PLAYER_PID 2>/dev/null
            break
        fi
        
        # Player is healthy, continue monitoring
    done
    
    # Player stopped or crashed
    # Check if user requested intentional exit
    if [ -f "$STOP_FLAG_FILE" ]; then
        log_message "✋ Stop flag detected - user requested exit via password"
        log_message "Watchdog will NOT restart the player"
        log_message "To restart, run ./start.sh again"
        rm -f "$HEARTBEAT_FILE"
        break
    fi
    
    log_message "⏳ Waiting ${RESTART_DELAY}s before restart..."
    sleep $RESTART_DELAY
    
    # Cleanup any zombie processes
    pkill -9 -f "python3 main.py" 2>/dev/null
    
done

log_message ""
log_message "=========================================="
log_message "Watchdog stopped"
log_message "=========================================="

# Deactivate virtual environment (this line is never reached in watchdog mode)
deactivate