Seata常规用法与配置实战指南

本文将深入讲解Seata的三种主要事务模式(AT/TCC/SAGA)的使用方法，以及与微服务框架的集成实践。

一、AT模式使用

1. 依赖引入

<dependency>
    <groupId>io.seata</groupId>
    <artifactId>seata-spring-boot-starter</artifactId>
    <version>最新版本</version>
</dependency>

2. 基础配置

seata:
  enabled: true
  application-id: order-service
  tx-service-group: my_tx_group  # 事务分组需与seata-server配置一致
  service:
    vgroup-mapping:
      my_tx_group: default       # 映射到seata-server集群
  config:
    type: nacos                  # 配置中心类型
  registry:
    type: nacos                  # 注册中心类型
  store:
    mode: db                     # 事务日志存储模式(db/file/redis)
    db:
      datasource: druid
      db-type: mysql
      url: jdbc:mysql://127.0.0.1:3306/seata
      user: root
      password: root

3. 注解使用

@Service
public class OrderService {
    
    @GlobalTransactional(timeoutMills = 300000, name = "createOrder")
    public void createOrder(Order order) {
        // 1. 扣减库存
        storageFeignClient.deduct(order.getProductId(), order.getCount());
        
        // 2. 创建订单
        orderMapper.insert(order);
        
        // 3. 扣减账户余额
        accountFeignClient.debit(order.getUserId(), order.getMoney());
    }
}

实践建议：

• 超时时间(timeoutMills)应根据业务链路的实际耗时合理设置
• 全局事务入口方法建议添加事务名称(name)便于问题排查
• 避免在全局事务方法中进行耗时操作（如文件IO）

二、TCC模式实现

1. 接口定义

@LocalTCC
public interface AccountTccService {

    @TwoPhaseBusinessAction(name = "debit", commitMethod = "confirm", rollbackMethod = "cancel")
    boolean tryDebit(@BusinessActionContextParameter(paramName = "userId") String userId,
                    @BusinessActionContextParameter(paramName = "money") BigDecimal money);
    
    boolean confirm(BusinessActionContext context);
    
    boolean cancel(BusinessActionContext context);
}

2. 业务实现

@Service
public class AccountTccServiceImpl implements AccountTccService {
    
    @Autowired
    private AccountFreezeMapper freezeMapper;
    
    @Override
    public boolean tryDebit(String userId, BigDecimal money) {
        // 检查账户余额
        Account account = accountMapper.selectById(userId);
        if (account.getBalance().compareTo(money) < 0) {
            throw new RuntimeException("余额不足");
        }
        
        // 冻结金额
        AccountFreeze freeze = new AccountFreeze();
        freeze.setUserId(userId);
        freeze.setFreezeMoney(money);
        freeze.setState(AccountFreezeState.TRY);
        freezeMapper.insert(freeze);
        return true;
    }
    
    @Override
    public boolean confirm(BusinessActionContext context) {
        // 扣减实际金额
        String userId = (String) context.getActionContext("userId");
        BigDecimal money = (BigDecimal) context.getActionContext("money");
        
        accountMapper.reduceBalance(userId, money);
        
        // 删除冻结记录
        freezeMapper.delete(new LambdaQueryWrapper<AccountFreeze>()
                .eq(AccountFreeze::getUserId, userId)
                .eq(AccountFreeze::getState, AccountFreezeState.TRY));
        return true;
    }
    
    @Override
    public boolean cancel(BusinessActionContext context) {
        // 释放冻结金额
        String userId = (String) context.getActionContext("userId");
        freezeMapper.delete(new LambdaQueryWrapper<AccountFreeze>()
                .eq(AccountFreeze::getUserId, userId)
                .eq(AccountFreeze::getState, AccountFreezeState.TRY));
        return true;
    }
}

实践建议：

• Try阶段应预留资源，Confirm/Cancel阶段必须实现幂等性
• 业务上下文(BusinessActionContext)可用于跨阶段参数传递
• 建议每个TCC服务单独建表记录资源预留状态

三、SAGA模式应用

1. 状态机定义(JSON)

{
  "name": "orderProcess",
  "steps": [
    {
      "name": "createOrder",
      "service": "orderService",
      "serviceMethod": "create",
      "compensateService": "orderService",
      "compensateMethod": "cancel"
    },
    {
      "name": "reserveInventory",
      "service": "inventoryService",
      "serviceMethod": "reserve",
      "compensateService": "inventoryService",
      "compensateMethod": "release"
    },
    {
      "name": "processPayment",
      "service": "paymentService",
      "serviceMethod": "pay",
      "compensateService": "paymentService",
      "compensateMethod": "refund"
    }
  ]
}

2. 状态机执行

StateMachineEngine stateMachineEngine = StateMachineEngineHolder.getStateMachineEngine();

Map<String, Object> startParams = new HashMap<>();
startParams.put("orderParams", orderDto);
startParams.put("inventoryParams", inventoryReq);

StateMachineInstance instance = stateMachineEngine.start(
        "orderProcess", 
        null, 
        startParams);

实践建议：

• 适用于业务流程长、需要人工干预的场景
• 补偿服务必须实现幂等且可重试
• 建议配合可视化工具设计状态机流程

四、与微服务框架集成

1. Spring Cloud集成

@FeignClient(name = "account-service")
public interface AccountFeignClient {
    
    @PostMapping("/account/debit")
    Boolean debit(@RequestParam("userId") String userId, 
                 @RequestParam("money") BigDecimal money);
}

// 自动传递XID的RestTemplate配置
@Bean
@LoadBalanced
public RestTemplate restTemplate() {
    RestTemplate restTemplate = new RestTemplate();
    restTemplate.setInterceptors(Collections.singletonList(
            new SeataRestTemplateInterceptor()));
    return restTemplate;
}

2. Dubbo集成

@Service(version = "1.0.0", interfaceClass = AccountService.class)
public class AccountServiceImpl implements AccountService {
    
    @Override
    @GlobalTransactional
    public boolean debit(String userId, BigDecimal money) {
        // 业务实现
    }
}

// 消费者调用
@Reference(version = "1.0.0")
private AccountService accountService;

实践建议：

• Feign/RestTemplate需确保Seata拦截器正确配置
• Dubbo服务建议使用1.4.2+版本以获得最佳兼容性
• 跨服务调用时避免使用@Transactional与@GlobalTransactional混用

五、模式选型建议

选型原则：

• AT模式：适合大多数简单场景，对业务侵入小
• TCC模式：需要强一致性且业务可补偿的场景
• SAGA模式：业务流程长且可接受最终一致性的场景

通过以上配置和示例，开发者可以快速在项目中集成Seata的分布式事务能力。建议根据实际业务需求选择合适的事务模式，并在测试环境充分验证事务行为。