llm-fusion-mcp/health-check.sh

#!/bin/bash
# LLM Fusion MCP - Health Check & Monitoring Script

set -e

# Configuration
SERVICE_NAME="llm-fusion-mcp"
HEALTH_ENDPOINT="http://localhost:8000/health"
TIMEOUT=10
CHECK_INTERVAL=30

# Colors
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m'

print_status() {
    echo -e "${BLUE}[$(date '+%Y-%m-%d %H:%M:%S')]${NC} $1"
}

print_success() {
    echo -e "${GREEN}[SUCCESS]${NC} $1"
}

print_warning() {
    echo -e "${YELLOW}[WARNING]${NC} $1"
}

print_error() {
    echo -e "${RED}[ERROR]${NC} $1"
}

# Health check function
check_health() {
    local endpoint=$1
    local response
    local http_code
    
    response=$(curl -s -w "HTTPSTATUS:%{http_code}" --max-time $TIMEOUT "$endpoint" 2>/dev/null || echo "HTTPSTATUS:000")
    http_code=$(echo "$response" | grep -o "HTTPSTATUS:[0-9]*" | cut -d: -f2)
    
    if [ "$http_code" = "200" ]; then
        return 0
    else
        return 1
    fi
}

# Docker container check
check_container() {
    if docker ps --filter "name=${SERVICE_NAME}" --filter "status=running" | grep -q "$SERVICE_NAME"; then
        return 0
    else
        return 1
    fi
}

# System resource check
check_resources() {
    local container_id
    container_id=$(docker ps -q --filter "name=${SERVICE_NAME}")
    
    if [ -n "$container_id" ]; then
        local stats
        stats=$(docker stats --no-stream --format "table {{.CPUPerc}}\t{{.MemUsage}}" "$container_id" 2>/dev/null | tail -n 1)
        
        if [ -n "$stats" ]; then
            local cpu_usage memory_usage
            cpu_usage=$(echo "$stats" | awk '{print $1}' | tr -d '%')
            memory_usage=$(echo "$stats" | awk '{print $2}')
            
            echo "CPU: ${cpu_usage}%, Memory: ${memory_usage}"
            
            # Alert if CPU > 80%
            if (( $(echo "$cpu_usage > 80" | bc -l) )); then
                print_warning "High CPU usage: ${cpu_usage}%"
            fi
        fi
    fi
}

# Provider connectivity check
check_providers() {
    local response
    response=$(curl -s --max-time $TIMEOUT "${HEALTH_ENDPOINT}/providers" 2>/dev/null || echo "{}")
    
    if echo "$response" | grep -q "\"success\":true"; then
        local provider_count
        provider_count=$(echo "$response" | grep -o "\"configured\":true" | wc -l)
        echo "Active providers: $provider_count"
    else
        print_warning "Provider health check failed"
    fi
}

# Main monitoring function
run_monitor() {
    print_status "Starting LLM Fusion MCP health monitoring..."
    
    while true; do
        echo ""
        print_status "=== Health Check Report ==="
        
        # Container status
        if check_container; then
            print_success "✅ Container is running"
            
            # Resource usage
            local resource_info
            resource_info=$(check_resources)
            if [ -n "$resource_info" ]; then
                print_status "📊 Resource usage: $resource_info"
            fi
            
        else
            print_error "❌ Container is not running"
            print_status "Attempting to restart..."
            docker-compose restart "$SERVICE_NAME" || print_error "Failed to restart container"
            sleep 10
            continue
        fi
        
        # Health endpoint check
        if check_health "$HEALTH_ENDPOINT"; then
            print_success "✅ Health endpoint responding"
        else
            print_error "❌ Health endpoint not responding"
        fi
        
        # Provider check
        print_status "🔍 Checking AI providers..."
        check_providers
        
        # Disk space check
        local disk_usage
        disk_usage=$(df -h . | tail -1 | awk '{print $5}' | tr -d '%')
        if [ "$disk_usage" -gt 85 ]; then
            print_warning "⚠️  Low disk space: ${disk_usage}% used"
        else
            print_status "💾 Disk usage: ${disk_usage}%"
        fi
        
        # Log file size check
        if [ -d "./logs" ]; then
            local log_size
            log_size=$(du -sh ./logs 2>/dev/null | cut -f1 || echo "N/A")
            print_status "📝 Log directory size: $log_size"
        fi
        
        print_status "Next check in ${CHECK_INTERVAL} seconds..."
        sleep $CHECK_INTERVAL
    done
}

# One-time health check
run_check() {
    print_status "Running one-time health check..."
    
    # Container check
    if check_container; then
        print_success "✅ Container Status: Running"
    else
        print_error "❌ Container Status: Not Running"
        return 1
    fi
    
    # Health endpoint
    if check_health "$HEALTH_ENDPOINT"; then
        print_success "✅ Health Endpoint: OK"
    else
        print_error "❌ Health Endpoint: Failed"
        return 1
    fi
    
    # Resource usage
    local resource_info
    resource_info=$(check_resources)
    if [ -n "$resource_info" ]; then
        print_status "📊 Resource Usage: $resource_info"
    fi
    
    # Provider check
    check_providers
    
    print_success "🎉 All checks passed!"
    return 0
}

# Usage information
show_usage() {
    echo "LLM Fusion MCP Health Check Script"
    echo ""
    echo "Usage: $0 [COMMAND]"
    echo ""
    echo "Commands:"
    echo "  check     Run one-time health check"
    echo "  monitor   Start continuous monitoring"
    echo "  help      Show this help message"
    echo ""
    echo "Environment Variables:"
    echo "  HEALTH_ENDPOINT    Health check URL (default: http://localhost:8000/health)"
    echo "  CHECK_INTERVAL     Monitoring interval in seconds (default: 30)"
    echo "  TIMEOUT           HTTP timeout in seconds (default: 10)"
}

# Main script logic
case "${1:-check}" in
    "monitor")
        run_monitor
        ;;
    "check")
        run_check
        ;;
    "help"|"-h"|"--help")
        show_usage
        ;;
    *)
        print_error "Unknown command: $1"
        show_usage
        exit 1
        ;;
esac