labFusion/runners/fix-cache-issues.sh

#!/bin/bash

# Cache Troubleshooting and Fix Script for LabFusion CI/CD
# This script helps diagnose and fix common cache timeout issues

set -e

echo "🔧 LabFusion Cache Troubleshooting Script"
echo "=========================================="

# Function to check if running in Docker
check_docker() {
    if [ -f /.dockerenv ]; then
        echo "🐳 Running inside Docker container"
        return 0
    else
        echo "🖥️  Running on host system"
        return 1
    fi
}

# Function to check cache service status
check_cache_service() {
    echo "📊 Checking cache service status..."
    
    # Check if cache service is running
    if pgrep -f "act_runner" > /dev/null; then
        echo "✅ act_runner process found"
    else
        echo "❌ act_runner process not found"
        return 1
    fi
    
    # Check cache directory
    CACHE_DIR="${HOME}/.cache/actcache"
    if [ -d "$CACHE_DIR" ]; then
        echo "✅ Cache directory exists: $CACHE_DIR"
        echo "   Size: $(du -sh "$CACHE_DIR" 2>/dev/null || echo "Unknown")"
    else
        echo "⚠️  Cache directory not found: $CACHE_DIR"
        echo "   Creating cache directory..."
        mkdir -p "$CACHE_DIR"
    fi
}

# Function to test network connectivity
test_network() {
    echo "🌐 Testing network connectivity..."
    
    # Test basic connectivity
    if ping -c 1 8.8.8.8 > /dev/null 2>&1; then
        echo "✅ Internet connectivity OK"
    else
        echo "❌ Internet connectivity failed"
    fi
    
    # Test Docker daemon
    if docker info > /dev/null 2>&1; then
        echo "✅ Docker daemon accessible"
    else
        echo "❌ Docker daemon not accessible"
    fi
}

# Function to detect the correct host IP
detect_host_ip() {
    echo "🔍 Detecting host IP address..."
    
    # Try different methods to get the host IP
    if command -v ip > /dev/null 2>&1; then
        # Linux with ip command
        HOST_IP=$(ip route get 1.1.1.1 | awk '{print $7; exit}' 2>/dev/null)
    elif command -v hostname > /dev/null 2>&1; then
        # Try hostname -I (Linux)
        HOST_IP=$(hostname -I | awk '{print $1}' 2>/dev/null)
    elif command -v ifconfig > /dev/null 2>&1; then
        # Try ifconfig (macOS/BSD)
        HOST_IP=$(ifconfig | grep -Eo 'inet (addr:)?([0-9]*\.){3}[0-9]*' | grep -Eo '([0-9]*\.){3}[0-9]*' | grep -v '127.0.0.1' | head -1)
    else
        # Fallback to localhost
        HOST_IP="127.0.0.1"
    fi
    
    if [ -z "$HOST_IP" ] || [ "$HOST_IP" = "127.0.0.1" ]; then
        echo "⚠️  Could not detect host IP, using localhost"
        HOST_IP="127.0.0.1"
    else
        echo "✅ Detected host IP: $HOST_IP"
    fi
    
    echo "$HOST_IP"
}

# Function to fix common cache issues
fix_cache_issues() {
    echo "🔧 Applying cache fixes..."
    
    # Create cache directory with proper permissions
    CACHE_DIR="${HOME}/.cache/actcache"
    mkdir -p "$CACHE_DIR"
    chmod 755 "$CACHE_DIR"
    
    # Detect host IP
    HOST_IP=$(detect_host_ip)
    
    # Set proper environment variables
    export ACTIONS_CACHE_URL="http://${HOST_IP}:40047/"
    export ACTIONS_RUNTIME_URL="http://${HOST_IP}:40047/"
    
    echo "✅ Cache directory created and configured"
    echo "✅ Environment variables set with host IP: $HOST_IP"
}

# Function to restart cache service
restart_cache_service() {
    echo "🔄 Restarting cache service..."
    
    # Stop existing runners
    pkill -f "act_runner" || true
    sleep 2
    
    # Start with updated configuration
    if [ -f "config_docker.yaml" ]; then
        echo "✅ Using updated Docker configuration"
        nohup ./act_runner daemon --config config_docker.yaml > runner.log 2>&1 &
    elif [ -f "config_heavy.yaml" ]; then
        echo "✅ Using updated heavy configuration"
        nohup ./act_runner daemon --config config_heavy.yaml > runner.log 2>&1 &
    else
        echo "⚠️  Updated configuration not found, using default"
        nohup ./act_runner daemon > runner.log 2>&1 &
    fi
    
    sleep 5
    
    if pgrep -f "act_runner" > /dev/null; then
        echo "✅ Cache service restarted successfully"
    else
        echo "❌ Failed to restart cache service"
        return 1
    fi
}

# Function to test cache functionality
test_cache() {
    echo "🧪 Testing cache functionality..."
    
    # Create a test cache entry
    TEST_KEY="test-cache-$(date +%s)"
    TEST_VALUE="test-value-$(date +%s)"
    
    echo "   Creating test cache entry: $TEST_KEY"
    echo "$TEST_VALUE" > "/tmp/cache-test"
    
    # Try to restore (this will fail but we can check the error)
    echo "   Testing cache restore..."
    if curl -s "http://${HOST_IP}:40047/cache/$TEST_KEY" > /dev/null 2>&1; then
        echo "✅ Cache service responding"
    else
        echo "❌ Cache service not responding"
        echo "   This is expected if no cache entry exists"
    fi
    
    # Clean up
    rm -f "/tmp/cache-test"
}

# Main execution
main() {
    echo "Starting cache troubleshooting..."
    echo ""
    
    check_docker
    echo ""
    
    check_cache_service
    echo ""
    
    test_network
    echo ""
    
    fix_cache_issues
    echo ""
    
    restart_cache_service
    echo ""
    
    test_cache
    echo ""
    
    echo "🎉 Cache troubleshooting complete!"
    echo ""
    echo "Next steps:"
    echo "1. Check runner logs: tail -f runner.log"
    echo "2. Test a workflow to see if cache issues are resolved"
    echo "3. If issues persist, check Docker networking configuration"
    echo ""
    echo "For more help, see: https://gitea.com/gitea/act_runner/src/branch/main/docs/configuration.md"
}

# Run main function
main "$@"