NextEdu/scripts/health-check.sh

#!/bin/bash
# 健康检查脚本
# 用法: ./health-check.sh
# 检查应用、数据库、磁盘空间、备份新鲜度,输出 JSON 报告

set -u

show_help() {
  cat <<EOF
用法: $0 [选项]
系统健康检查脚本,输出 JSON 格式报告

选项:
  --app-url URL        应用健康检查 URL(默认 http://localhost:8015)
  --no-app             跳过应用健康检查
  --no-db              跳过数据库检查
  --no-disk            跳过磁盘空间检查
  --no-backup          跳过备份新鲜度检查
  --disk-threshold PCT 磁盘空间阈值百分比(默认 90)
  --backup-max-age HRS 备份最大年龄(小时,默认 24)
  --help, -h           显示帮助信息

环境变量:
  DATABASE_URL              数据库连接 URL
  HEALTH_CHECK_URL          应用健康检查 URL(默认 http://localhost:8015)
  BACKUP_DIR                备份目录(默认 ./backups)
  HEALTH_CHECK_DISK_THRESHOLD  磁盘阈值(默认 90)
  HEALTH_CHECK_BACKUP_MAX_AGE  备份最大年龄(小时,默认 24)

退出码:
  0  健康
  1  异常
EOF
}

# 解析参数
CHECK_APP=1
CHECK_DB=1
CHECK_DISK=1
CHECK_BACKUP=1
APP_URL=""
DISK_THRESHOLD=""
BACKUP_MAX_AGE=""

while [ $# -gt 0 ]; do
  case "$1" in
    --help|-h)
      show_help
      exit 0
      ;;
    --app-url)
      if [ $# -lt 2 ]; then
        echo "ERROR: --app-url requires an argument" >&2
        exit 1
      fi
      APP_URL="$2"
      shift 2
      ;;
    --no-app) CHECK_APP=0; shift ;;
    --no-db) CHECK_DB=0; shift ;;
    --no-disk) CHECK_DISK=0; shift ;;
    --no-backup) CHECK_BACKUP=0; shift ;;
    --disk-threshold)
      if [ $# -lt 2 ]; then
        echo "ERROR: --disk-threshold requires an argument" >&2
        exit 1
      fi
      DISK_THRESHOLD="$2"
      shift 2
      ;;
    --backup-max-age)
      if [ $# -lt 2 ]; then
        echo "ERROR: --backup-max-age requires an argument" >&2
        exit 1
      fi
      BACKUP_MAX_AGE="$2"
      shift 2
      ;;
    *)
      echo "ERROR: Unknown argument: $1" >&2
      exit 1
      ;;
  esac
done

# 配置
APP_URL="${APP_URL:-${HEALTH_CHECK_URL:-http://localhost:8015}}"
BACKUP_DIR="${BACKUP_DIR:-./backups}"
DISK_THRESHOLD="${DISK_THRESHOLD:-${HEALTH_CHECK_DISK_THRESHOLD:-90}}"
BACKUP_MAX_AGE="${BACKUP_MAX_AGE:-${HEALTH_CHECK_BACKUP_MAX_AGE:-24}}"
DATABASE_URL="${DATABASE_URL:-}"

# JSON 输出辅助函数
json_escape() {
  echo "$1" | sed -e 's/\\/\\\\/g' -e 's/"/\\"/g' -e 's/\t/\\t/g'
}

# 检查结果数组
RESULTS=""
OVERALL_STATUS="healthy"
CHECKS_PASSED=0
CHECKS_FAILED=0
CHECKS_WARNED=0

add_result() {
  local name="$1"
  local status="$2"
  local message="$3"
  local detail="${4:-}"

  local escaped_message escaped_detail
  escaped_message=$(json_escape "$message")
  escaped_detail=$(json_escape "$detail")

  local result_entry
  result_entry="    {\"name\": \"$name\", \"status\": \"$status\", \"message\": \"$escaped_message\""
  if [ -n "$detail" ]; then
    result_entry="$result_entry, \"detail\": \"$escaped_detail\""
  fi
  result_entry="$result_entry }"

  if [ -z "$RESULTS" ]; then
    RESULTS="$result_entry"
  else
    RESULTS="$RESULTS,
$result_entry"
  fi

  case "$status" in
    pass) CHECKS_PASSED=$((CHECKS_PASSED + 1)) ;;
    fail) CHECKS_FAILED=$((CHECKS_FAILED + 1)); OVERALL_STATUS="unhealthy" ;;
    warn) CHECKS_WARNED=$((CHECKS_WARNED + 1)); [ "$OVERALL_STATUS" = "healthy" ] && OVERALL_STATUS="degraded" ;;
  esac
}

# 1. 应用健康检查
if [ "$CHECK_APP" -eq 1 ]; then
  if command -v curl >/dev/null 2>&1; then
    HTTP_CODE=$(curl -sf -o /dev/null -w "%{http_code}" -m 10 "$APP_URL" 2>/dev/null || echo "000")
    if [ "$HTTP_CODE" = "200" ] || [ "$HTTP_CODE" = "302" ]; then
      add_result "app" "pass" "Application is healthy" "HTTP $HTTP_CODE from $APP_URL"
    elif [ "$HTTP_CODE" = "000" ]; then
      add_result "app" "fail" "Application is not reachable" "Cannot connect to $APP_URL"
    else
      add_result "app" "fail" "Application returned error" "HTTP $HTTP_CODE from $APP_URL"
    fi
  else
    add_result "app" "warn" "curl not available, skipping app check" ""
  fi
fi

# 2. 数据库连接检查
if [ "$CHECK_DB" -eq 1 ]; then
  if [ -z "$DATABASE_URL" ]; then
    add_result "database" "warn" "DATABASE_URL not set, skipping DB check" ""
  else
    # 解析 DATABASE_URL
    DB_USER=$(echo "$DATABASE_URL" | sed -n 's/.*:\/\/\([^:]*\):.*/\1/p')
    DB_PASS=$(echo "$DATABASE_URL" | sed -n 's/.*:\/\/[^:]*:\([^@]*\)@.*/\1/p')
    DB_HOST=$(echo "$DATABASE_URL" | sed -n 's/.*@\([^:]*\):.*/\1/p')
    DB_PORT=$(echo "$DATABASE_URL" | sed -n 's/.*:\([0-9]*\)\/.*/\1/p')
    DB_NAME=$(echo "$DATABASE_URL" | sed -n 's/.*\/\([^?]*\).*/\1/p')

    if command -v mysql >/dev/null 2>&1; then
      if mysql -h "$DB_HOST" -P "$DB_PORT" -u "$DB_USER" -p"$DB_PASS" \
          -e "SELECT 1;" 2>/dev/null; then
        # 获取连接信息
        DB_VERSION=$(mysql -h "$DB_HOST" -P "$DB_PORT" -u "$DB_USER" -p"$DB_PASS" \
            -e "SELECT VERSION();" -s -N 2>/dev/null || echo "unknown")
        add_result "database" "pass" "Database connection successful" "Host: $DB_HOST:$DB_PORT, DB: $DB_NAME, Version: $DB_VERSION"
      else
        add_result "database" "fail" "Database connection failed" "Cannot connect to $DB_HOST:$DB_PORT/$DB_NAME"
      fi
    else
      add_result "database" "warn" "mysql client not available, skipping DB check" ""
    fi
  fi
fi

# 3. 磁盘空间检查
if [ "$CHECK_DISK" -eq 1 ]; then
  # 获取根分区或当前目录所在分区的使用率
  DISK_INFO=$(df -h . 2>/dev/null | tail -1)
  if [ -n "$DISK_INFO" ]; then
    DISK_USE_PCT=$(echo "$DISK_INFO" | awk '{print $5}' | sed 's/%//')
    DISK_USE_HUMAN=$(echo "$DISK_INFO" | awk '{print $3}')
    DISK_TOTAL_HUMAN=$(echo "$DISK_INFO" | awk '{print $2}')
    DISK_AVAIL_HUMAN=$(echo "$DISK_INFO" | awk '{print $4}')
    DISK_MOUNT=$(echo "$DISK_INFO" | awk '{print $6}')

    if [ "$DISK_USE_PCT" -ge "$DISK_THRESHOLD" ]; then
      add_result "disk" "fail" "Disk space critical" "Usage: ${DISK_USE_PCT}% (threshold: ${DISK_THRESHOLD}%), Used: ${DISK_USE_HUMAN}/${DISK_TOTAL_HUMAN}, Available: ${DISK_AVAIL_HUMAN}, Mount: ${DISK_MOUNT}"
    elif [ "$DISK_USE_PCT" -ge $((DISK_THRESHOLD - 10)) ]; then
      add_result "disk" "warn" "Disk space warning" "Usage: ${DISK_USE_PCT}% (threshold: ${DISK_THRESHOLD}%), Used: ${DISK_USE_HUMAN}/${DISK_TOTAL_HUMAN}, Available: ${DISK_AVAIL_HUMAN}, Mount: ${DISK_MOUNT}"
    else
      add_result "disk" "pass" "Disk space OK" "Usage: ${DISK_USE_PCT}%, Used: ${DISK_USE_HUMAN}/${DISK_TOTAL_HUMAN}, Available: ${DISK_AVAIL_HUMAN}, Mount: ${DISK_MOUNT}"
    fi
  else
    add_result "disk" "warn" "Could not determine disk usage" ""
  fi
fi

# 4. 备份新鲜度检查
if [ "$CHECK_BACKUP" -eq 1 ]; then
  if [ -d "$BACKUP_DIR" ]; then
    LATEST_BACKUP=$(ls -t "$BACKUP_DIR"/db_backup_*.sql.gz 2>/dev/null | head -1)
    if [ -n "$LATEST_BACKUP" ]; then
      # 获取备份文件修改时间(秒)
      BACKUP_MTIME=$(stat -c%Y "$LATEST_BACKUP" 2>/dev/null || stat -f%m "$LATEST_BACKUP" 2>/dev/null)
      CURRENT_TIME=$(date +%s)
      BACKUP_AGE_HOURS=$(( (CURRENT_TIME - BACKUP_MTIME) / 3600 ))

      BACKUP_SIZE=$(stat -c%s "$LATEST_BACKUP" 2>/dev/null || stat -f%z "$LATEST_BACKUP" 2>/dev/null)
      BACKUP_SIZE_HUMAN=$(echo "$BACKUP_SIZE" | awk '{split("B KB MB GB TB",v);i=1;while($1>=1024&&i<5){$1/=1024;i++};printf "%.1f%s",$1,v[i]}')

      if [ "$BACKUP_AGE_HOURS" -gt "$BACKUP_MAX_AGE" ]; then
        add_result "backup" "fail" "Backup is stale" "Latest backup is ${BACKUP_AGE_HOURS}h old (max: ${BACKUP_MAX_AGE}h), File: $(basename "$LATEST_BACKUP"), Size: $BACKUP_SIZE_HUMAN"
      elif [ "$BACKUP_AGE_HOURS" -gt $((BACKUP_MAX_AGE / 2)) ]; then
        add_result "backup" "warn" "Backup getting old" "Latest backup is ${BACKUP_AGE_HOURS}h old (max: ${BACKUP_MAX_AGE}h), File: $(basename "$LATEST_BACKUP"), Size: $BACKUP_SIZE_HUMAN"
      else
        add_result "backup" "pass" "Backup is fresh" "Latest backup is ${BACKUP_AGE_HOURS}h old, File: $(basename "$LATEST_BACKUP"), Size: $BACKUP_SIZE_HUMAN"
      fi
    else
      add_result "backup" "fail" "No backup files found" "No db_backup_*.sql.gz files in $BACKUP_DIR"
    fi
  else
    add_result "backup" "warn" "Backup directory does not exist" "$BACKUP_DIR"
  fi
fi

# 输出 JSON 报告
TIMESTAMP=$(date -u +"%Y-%m-%dT%H:%M:%SZ")

cat <<EOF
{
  "timestamp": "$TIMESTAMP",
  "status": "$OVERALL_STATUS",
  "summary": {
    "total": $((CHECKS_PASSED + CHECKS_FAILED + CHECKS_WARNED)),
    "passed": $CHECKS_PASSED,
    "failed": $CHECKS_FAILED,
    "warned": $CHECKS_WARNED
  },
  "checks": [
$RESULTS
  ]
}
EOF

if [ "$OVERALL_STATUS" = "unhealthy" ]; then
  exit 1
fi
exit 0