homelab/scripts/validate-node.sh

#!/bin/bash

# ==============================================================================
# STANDALONE NODE VALIDATION TOOL
# ==============================================================================
# Comprehensive health check utility for homelab nodes
# Can be run independently on any managed host (post-bootstrap or ad-hoc)
#
# Usage:
#   ./validate-node.sh [OPTIONS]
#
# Options:
#   --help          Show this help message
#   --json          Output results in JSON format
#   --critical-only Show only critical errors (exit code 1 if any found)
#   --verbose       Show detailed output for each check
#
# Exit Codes:
#   0 - All checks passed or warnings only
#   1 - Critical errors found
#   2 - Invalid usage
#
# ==============================================================================

set -euo pipefail

# --- SCRIPT METADATA ---

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
SCRIPT_NAME="$(basename "${BASH_SOURCE[0]}")"
VERSION="1.0.0"

# --- LOAD LIBRARIES ---

# shellcheck source=./lib/detection.sh
source "${SCRIPT_DIR}/lib/detection.sh"
# shellcheck source=./lib/validation.sh
source "${SCRIPT_DIR}/lib/validation.sh"
# shellcheck source=./lib/network.sh
source "${SCRIPT_DIR}/lib/network.sh"

# --- COMMAND LINE ARGUMENTS ---

SHOW_HELP=false
OUTPUT_JSON=false
CRITICAL_ONLY=false
VERBOSE=false

parse_arguments() {
    while [[ $# -gt 0 ]]; do
        case "$1" in
            --help|-h)
                SHOW_HELP=true
                shift
                ;;
            --json)
                OUTPUT_JSON=true
                shift
                ;;
            --critical-only)
                CRITICAL_ONLY=true
                shift
                ;;
            --verbose|-v)
                VERBOSE=true
                shift
                ;;
            *)
                echo "ERROR: Unknown option: $1" >&2
                echo "Use --help for usage information" >&2
                exit 2
                ;;
        esac
    done
}

show_help() {
    cat <<'EOF'
=============================================================================
STANDALONE NODE VALIDATION TOOL v1.0.0
=============================================================================

Comprehensive health check for homelab infrastructure nodes.
Can be run on any managed host to verify operational readiness.

USAGE:
  ./validate-node.sh [OPTIONS]

OPTIONS:
  --help          Show this help message
  --json          Output results in JSON format (for monitoring integration)
  --critical-only Show only critical errors (suppress warnings)
  --verbose       Show detailed output for each check

EXIT CODES:
  0 - All checks passed (or warnings only)
  1 - Critical errors found
  2 - Invalid usage

CHECKS PERFORMED:
  • Disk Space & Performance
  • Memory & Swap Configuration
  • Network Routes & Connectivity
  • Hostname Resolution
  • NFS Client Configuration
  • Docker Daemon Health
  • Proxmox API (if applicable)
  • SSH Security Configuration
  • Firewall Status
  • Time Synchronization

EXAMPLES:
  ./validate-node.sh
    Run all checks with standard output

  ./validate-node.sh --critical-only
    Show only critical errors (useful in scripts)

  ./validate-node.sh --json
    Output JSON for monitoring/alerting systems

  ./validate-node.sh --verbose
    Show detailed information for each check

INTEGRATION:
  JSON output can be consumed by monitoring systems (Prometheus, Grafana, etc.)
  Example: ./validate-node.sh --json | jq '.errors'

=============================================================================
EOF
}

# --- JSON OUTPUT FUNCTIONS ---

generate_json_report() {
    # Generate JSON output for monitoring integration
    
    local hostname=$(hostname)
    local timestamp=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
    
    cat <<EOF
{
  "hostname": "$hostname",
  "timestamp": "$timestamp",
  "validation": {
    "passed": $VALIDATION_PASSED,
    "warnings": $VALIDATION_WARNINGS,
    "errors": $VALIDATION_ERRORS
  },
  "status": "$([ $VALIDATION_ERRORS -eq 0 ] && echo "healthy" || echo "critical")",
  "hardware_type": "$(detect_hardware_type)",
  "os_family": "$(detect_os_family)"
}
EOF
}

# --- MAIN WORKFLOW ---

main() {
    # Parse CLI arguments
    parse_arguments "$@"
    
    if [ "$SHOW_HELP" == "true" ]; then
        show_help
        exit 0
    fi
    
    # Standard output header (unless JSON mode)
    if [ "$OUTPUT_JSON" == "false" ]; then
        echo "======================================="
        echo "NODE VALIDATION TOOL v${VERSION}"
        echo "Host: $(hostname)"
        echo "Time: $(date -u +"%Y-%m-%d %H:%M:%S UTC")"
        echo "======================================="
        echo ""
    fi
    
    # Run validation suite
    if [ "$VERBOSE" == "true" ]; then
        # Show detection summary in verbose mode
        [ "$OUTPUT_JSON" == "false" ] && print_detection_summary
        [ "$OUTPUT_JSON" == "false" ] && echo ""
    fi
    
    # Run comprehensive validation
    run_validation_suite
    
    # Generate output
    if [ "$OUTPUT_JSON" == "true" ]; then
        generate_json_report
    else
        echo ""
        echo "Validation completed."
        echo ""
        
        # Show summary (already printed by run_validation_suite)
        if [ $VALIDATION_ERRORS -gt 0 ]; then
            echo "⚠️  Critical issues found - manual intervention required"
            exit 1
        elif [ $VALIDATION_WARNINGS -gt 0 ]; then
            echo "ℹ️  Warnings present - review recommended"
            exit 0
        else
            echo "✅ All checks passed - node is healthy"
            exit 0
        fi
    fi
    
    # Exit code based on errors
    [ $VALIDATION_ERRORS -eq 0 ] && exit 0 || exit 1
}

# --- ENTRY POINT ---

main "$@"