实现AI API Tokens计费产品方案和Java技术实现设计(含源码分享)

liuian 2025-04-27 14:46 137 浏览

以下是为基于 Java Spring Boot 3 的AI Token计费系统设计的完整方案，包含产品架构与技术实现细节：

一、系统架构设计

1. 分层架构

客户端 → API网关 → 业务微服务（用户/计费/订单/监控） → 数据库/缓存
           │          │
           ├─ Spring Security 鉴权
           └─ Spring Cloud Sleuth 链路追踪

2. 技术栈组合

核心框架: Spring Boot 3.1 + Spring WebFlux (响应式支持)
安全框架: Spring Security 6 + OAuth2/JWT
数据存储:

MySQL 8 (事务性数据)
Redis 7 (分布式锁/缓存)
Elasticsearch 8 (日志分析)

监控体系:

Micrometer + Prometheus + Grafana
Spring Boot Actuator (健康检查)

消息队列: RabbitMQ/Kafka (异步扣费)

二、数据库设计（JPA Entity示例）

1. 用户实体

java代码，

@Entity
@Table(name = "ai_user")
public class User {
    @Id
    @GeneratedValue(strategy = GenerationType.IDENTITY)
    private Long id;
    
    @Column(unique = true, nullable = false)
    private String email;
    
    @Column(precision = 12, scale = 4)
    private BigDecimal balance = BigDecimal.ZERO;
    
    @Version
    private Long version; // 乐观锁
}

2. 服务费率配置

java代码，

@Entity
@Table(name = "service_config")
public class ServiceConfig {
    @Id
    private String serviceId;
    
    private BigDecimal tokenRate;
    
    @Enumerated(EnumType.STRING)
    private TokenCalcMethod calcMethod; // ENUM类型
}

public enum TokenCalcMethod {
    CHAR_COUNT, WORD_COUNT, IMAGE_RESOLUTION
}

3. 消费记录（审计日志）

java代码，

@Entity
@Table(name = "token_record")
public class TokenRecord {
    @Id
    @GeneratedValue(strategy = GenerationType.SEQUENCE)
    private Long recordId;
    
    @ManyToOne
    private User user;
    
    private Instant requestTime;
    
    private Integer tokens;
    
    @Column(precision = 10, scale = 4)
    private BigDecimal cost;
}

三、核心功能实现

1. Token计算拦截器（Spring AOP）

java代码，

@Aspect
@Component
public class TokenBillingAspect {

    @Autowired
    private BillingService billingService;

    @Around("@annotation(com.ai.billing.RequiresToken)")
    public Object handleTokenDeduction(ProceedingJoinPoint joinPoint) throws Throwable {
        MethodSignature signature = (MethodSignature) joinPoint.getSignature();
        RequiresToken annotation = signature.getMethod().getAnnotation(RequiresToken.class);
        
        Object result = joinPoint.proceed(); // 执行AI服务
        
        int tokens = calculateTokens(result, annotation.serviceType());
        billingService.deductTokens(
            SecurityContextHolder.getContext().getAuthentication().getName(),
            annotation.serviceType(),
            tokens
        );
        
        return result;
    }

    private int calculateTokens(Object result, ServiceType serviceType) {
        // 根据服务类型计算Token逻辑
    }
}

2. 原子化扣费服务

java代码，

@Service
@Transactional
public class BillingService {

    @Autowired
    private UserRepository userRepository;
    
    @Autowired
    private RedisLockRegistry redisLockRegistry;

    public void deductTokens(String userId, String serviceId, int tokens) {
        Lock lock = redisLockRegistry.obtain(userId); // 分布式锁
        try {
            if (lock.tryLock(1, TimeUnit.SECONDS)) {
                User user = userRepository.findByEmail(userId)
                    .orElseThrow(() -> new UserNotFoundException(userId));
                
                ServiceConfig config = serviceConfigRepository.findById(serviceId)
                    .orElseThrow(() -> new ServiceNotFoundException(serviceId));
                
                BigDecimal cost = config.getTokenRate().multiply(BigDecimal.valueOf(tokens));
                
                if (user.getBalance().compareTo(cost) < 0) {
                    throw new InsufficientBalanceException();
                }
                
                user.setBalance(user.getBalance().subtract(cost));
                userRepository.save(user);
                
                tokenRecordRepository.save(new TokenRecord(user, tokens, cost));
            }
        } finally {
            lock.unlock();
        }
    }
}

3. 响应式支付接口（WebFlux）

java代码，

@RestController
@RequestMapping("/api/payment")
public class PaymentController {

    @Autowired
    private PaymentService paymentService;

    @PostMapping("/recharge")
    public Mono<ResponseEntity<PaymentResponse>> recharge(
        @RequestBody PaymentRequest request,
        @AuthenticationPrincipal Jwt jwt
    ) {
        return paymentService.processPayment(jwt.getSubject(), request)
            .map(response -> ResponseEntity.ok().body(response))
            .onErrorResume(e -> Mono.just(ResponseEntity.badRequest().build()));
    }
}

四、安全与监控方案

1. 安全防护

yaml配置文件，

# application-security.yml
spring:
  security:
    oauth2:
      resourceserver:
        jwt:
          issuer-uri: https://auth.ai.com
    rules:
      - pattern: /api/admin/**
        access: hasRole('ADMIN')
      - pattern: /api/payment/**
        access: isAuthenticated()

2. Prometheus监控配置

java代码，

@Configuration
public class MetricsConfig {

    @Bean
    MeterRegistryCustomizer<MeterRegistry> metricsCommonTags() {
        return registry -> registry.config().commonTags(
            "application", "ai-billing-service"
        );
    }
    
    @Bean
    public TimedAspect timedAspect(MeterRegistry registry) {
        return new TimedAspect(registry); // 方法级耗时监控
    }
}

五、高并发优化策略

1. 性能增强方案

策略	实现方式	目标
异步扣费	使用@Async + RabbitMQ	削峰填谷
缓存优化	Caffeine本地缓存 + Redis二级缓存	减少DB压力
批量操作	JPA @Query批量更新	提升吞吐量
连接池优化	HikariCP参数调优	降低延迟

2. 弹性设计

java代码，

// 基于Resilience4j的熔断机制
@CircuitBreaker(name = "billingService", fallbackMethod = "fallbackDeduction")
@RateLimiter(name = "billingRateLimit")
@Retry(name = "retryBilling")
public void deductTokens(...) { ... }

六、扩展能力设计

混合计费插件

java代码，

public interface BillingStrategy {
    BigDecimal calculateCost(ServiceConfig config, int tokens);
}

@Component
@ConditionalOnProperty(name = "billing.mode", havingValue = "hybrid")
public class HybridBillingStrategy implements BillingStrategy {
    // 组合计费逻辑
}

沙盒环境支持

java代码，

@Profile("sandbox")
@Configuration
public class SandboxConfig {
    @Bean
    public BillingService mockBillingService() {
        return new MockBillingService(); // 免扣费实现
    }
}

OpenAPI文档

java代码，

@Configuration
public class OpenApiConfig {
    @Bean
    public OpenAPI aiBillingOpenAPI() {
        return new OpenAPI().info(new Info().title("AI Billing API"));
    }
}

该方案充分利用Spring Boot 3的特性：