Spring Alibaba Cloud 使用 Seata 实现分布式事务(二)之原理分析

1,823 阅读16分钟

在上一篇文章Spring Alibaba Cloud 使用 Seata 实现分布式事务,Nacos 作为 配置中心(一)进行实战演示,这篇主要讲原理解释及流程分析,最好大家可以结合代码进行阅读

一. Seata 事务分组

1.什么是事务分组?

大家看完之后很疑惑 我配置

vgroup_mapping.prex_tx_group = "default"

这个事务组有什么用?

A:事务分组是 Seata 的资源逻辑,类似于服务实例。

service {
  #vgroup->rgroup
  vgroup_mapping.prex_tx_group = "default"
  #only support single node
  default.grouplist = "127.0.0.1:8091"
  #degrade current not support
  enableDegrade = false
  #disable
  disable = false
  #unit ms,s,m,h,d represents milliseconds, seconds, minutes, hours, days, default permanent
  max.commit.retry.timeout = "-1"
  max.rollback.retry.timeout = "-1"
}

file.conf中,比如这种配置,这里的fsp_tx_group 就是一个事务分组。一个seata-server可以管理多个事务分组。

2.通过事务分组如何找到后端集群?

A:首先程序中配置了事务分组(GlobalTransactionScanner 构造方法的 txServiceGroup 参数),程序会通过用户配置的配置中心去寻找 service.vgroup_mapping. 事务分组配置项,取得配置项的值就是 TC 集群的名称。拿到集群名称程序通过一定的前后缀+集群名称去构造服务名,各配置中心的服务名实现不同。拿到服务名去相应的注册中心去拉取相应服务名的服务列表,获得后端真实的 TC 服务列表。

3.为什么这么设计,不直接取服务名?

A:这里多了一层获取事务分组到映射集群的配置。这样设计后,事务分组可以作为资源的逻辑隔离单位,当发生故障时可以快速 failover。

4.我有10个微服务,那我要分10个组吗 ?

A:分组的含义就是映射到一套集群,所以你可以配一个分组也可以配置多个。如果你其他的微服务有独立发起事务可以配置多个,如果只是作为服务调用方参与事务那么没必要配置多个。

5.不知道分组的目的是什么?那不管什么情况我始终就一个分组有没问题?

A:没问题,分组是用于资源的逻辑隔离,多租户的概念。

6.是不是一个事务中所有的微服务都必须是同一组才行?

A:没有这个要求的。但是不同的分组需要映射到同一个集群上。

7.你说的集群是指 TC 集群吗?现在 TC 如何集群,我看配置里都是 default。

A:那个名字可以自己取,如果用文件形式你可以写多个地址列表,多台 server 以 DB 方式存储通过 DB 共享数据。如果用注册中心就可以自己发现,注册的时候定义了集群名。

备注:log_status是1,1的是防御性的,是收到globalrollback回滚请求,但是不确定某个事务分支的本地事务是否已经执行完成了,这时事先插入一条branchid相同的数据,插入的假数据成功了,本地事务继续执行就会报主键冲突自动回滚。假如插入不成功说明表里有数据这个本地事务已经执行完成了,那么取出这条undolog数据做反向回滚操作

二. Seata分布式事务原理解释

1. 相关概念

  • XID:一个全局事务的唯一标识,由ip:port:sequence组成
  • Transaction Coordinator (TC): 事务协调器,维护全局事务的运行状态,负责协调并驱动全局事务的提交或回滚。
  • Transaction Manager ™: 控制全局事务的边界,负责开启一个全局事务,并最终发起全局提交或全局回滚的决议。
  • Resource Manager (RM): 控制分支事务,负责分支注册、状态汇报,并接收事务协调器的指令,驱动分支(本地)事务的提交和回滚。

2. 配置文件

seata的配置文件入口为registry.conf查看代码ConfigurationFactory得知可以使用seata.config.name进行自定义配置文件,没有配置则默认使用registry.conf

 static {
        String seataConfigName = System.getProperty("seata.config.name");
        if (null == seataConfigName) {
            seataConfigName = System.getenv("SEATA_CONFIG_NAME");
        }

        if (null == seataConfigName) {
            seataConfigName = "registry";
        }

        String envValue = System.getProperty("seataEnv");
        if (null == envValue) {
            envValue = System.getenv("SEATA_ENV");
        }

        CURRENT_FILE_INSTANCE = null == envValue ? new FileConfiguration(seataConfigName + ".conf") : new FileConfiguration(seataConfigName + "-" + envValue + ".conf");
        instance = null;
    }

在registry中可以指定具体配置的形式,这里使用默认的file形式。在file.conf中有3部分配置内容

  1. transport transport 部分的配置对应 NettyServerConfig 类,用于定义 Netty 相关的参数,TM、RM 与 seata-server 之间使用 Netty 进行通信。
transport {
  # tcp udt unix-domain-socket
  type = "TCP"
  #NIO NATIVE
  server = "NIO"
  #enable heartbeat
  heartbeat = true
  #thread factory for netty
  thread-factory {
    boss-thread-prefix = "NettyBoss"
    worker-thread-prefix = "NettyServerNIOWorker"
    server-executor-thread-prefix = "NettyServerBizHandler"
    share-boss-worker = false
    client-selector-thread-prefix = "NettyClientSelector"
    client-selector-thread-size = 1
    client-worker-thread-prefix = "NettyClientWorkerThread"
    # netty boss thread size,will not be used for UDT
    boss-thread-size = 1
    #auto default pin or 8
    worker-thread-size = 8
  }
  shutdown {
    # when destroy server, wait seconds
    wait = 3
  }
  serialization = "seata"
  compressor = "none"
}
  1. service
service {
  #vgroup->rgroup
  vgroup_mapping.prex_tx_group = "default"
  #only support single node
  #配置Client连接TC的地址
  default.grouplist = "127.0.0.1:8091"
  #degrade current not support
  enableDegrade = false
  #disable
  disable = false
  #unit ms,s,m,h,d represents milliseconds, seconds, minutes, hours, days, default permanent
  max.commit.retry.timeout = "-1"
  max.rollback.retry.timeout = "-1"
}
  1. client
client {
# RM接收TC的commit通知后缓冲上限
  async.commit.buffer.limit = 10000
  lock {
    retry.internal = 10
    retry.times = 30
  }
  report.retry.count = 5
  tm.commit.retry.count = 1
  tm.rollback.retry.count = 1
}

3. 启动Server

启动成功输出Server started

2019-11-27 12:53:59.089 INFO [main]io.seata.core.rpc.netty.AbstractRpcRemotingServer.start:156 -Server started ...

4. 启动Client

对于Spring boot项目,启动运行xxxApplication的main方法即可,seata的加载入口类位于 com.alibaba.cloud.seata.GlobalTransactionAutoConfiguration

@Configuration
@EnableConfigurationProperties({SeataProperties.class})
public class GlobalTransactionAutoConfiguration {
    private final ApplicationContext applicationContext;
    private final SeataProperties seataProperties;

    public GlobalTransactionAutoConfiguration(ApplicationContext applicationContext, SeataProperties seataProperties) {
        this.applicationContext = applicationContext;
        this.seataProperties = seataProperties;
    }

    @Bean
    public GlobalTransactionScanner globalTransactionScanner() {
        String applicationName = this.applicationContext.getEnvironment().getProperty("spring.application.name");
        String txServiceGroup = this.seataProperties.getTxServiceGroup();
        if (StringUtils.isEmpty(txServiceGroup)) {
            txServiceGroup = applicationName + "-seata-service-group";
            this.seataProperties.setTxServiceGroup(txServiceGroup);
        }

        return new GlobalTransactionScanner(applicationName, txServiceGroup);
    }
}

可以看到支持一个配置项SeataProperties,用于配置事务分组名称

spring.cloud.alibaba.seata.tx-service-group=prex_tx_group

如果不指定则用spring.application.name+ -seata-service-group生成一个名称,所以不指定spring.application.name启动会报错

@ConfigurationProperties("spring.cloud.alibaba.seata")
public class SeataProperties {
    private String txServiceGroup;

    public SeataProperties() {
    }

    public String getTxServiceGroup() {
        return this.txServiceGroup;
    }

    public void setTxServiceGroup(String txServiceGroup) {
        this.txServiceGroup = txServiceGroup;
    }
}

有了applicationId和txServiceGroup则创建io.seata.spring.annotation.GlobalTransactionScanner对象,主要看其中的initClient方法

private void initClient() {
        if (LOGGER.isInfoEnabled()) {
            LOGGER.info("Initializing Global Transaction Clients ... ");
        }

        if (!StringUtils.isNullOrEmpty(this.applicationId) && !StringUtils.isNullOrEmpty(this.txServiceGroup)) {
            //init TM
            TMClient.init(this.applicationId, this.txServiceGroup);
            if (LOGGER.isInfoEnabled()) {
                LOGGER.info("Transaction Manager Client is initialized. applicationId[" + this.applicationId + "] txServiceGroup[" + this.txServiceGroup + "]");
            }
            //init RM
            RMClient.init(this.applicationId, this.txServiceGroup);
            if (LOGGER.isInfoEnabled()) {
                LOGGER.info("Resource Manager is initialized. applicationId[" + this.applicationId + "] txServiceGroup[" + this.txServiceGroup + "]");
            }

            if (LOGGER.isInfoEnabled()) {
                LOGGER.info("Global Transaction Clients are initialized. ");
            }

            this.registerSpringShutdownHook();
        } else {
            throw new IllegalArgumentException("applicationId: " + this.applicationId + ", txServiceGroup: " + this.txServiceGroup);
        }
    }

可以看到初始化了TMClientRMClient,对于一个服务既可以是TM角色也可以是RM角色,至于什么时候是TM或者RM则要看在一次全局事务中@GlobalTransactional注解标注在哪。 Client创建的结果是与TC的一个Netty连接,所以在启动日志中可以看到两个Netty Channel,其中也标明了transactionRole分别为TMROLE和RMROLE

file
日志中可以看到创建连接后,发送了注册请求,然后得到了结果相应,RmRpcClient、TmRpcClient成功实例化。

5. TM处理流程

在本例中,TM的角色是nacos-seata-order-server,因为nacos-seata-order-server的createOrder方法标注了@GlobalTransactional

@Slf4j
@Service
public class OrderServiceImpl extends ServiceImpl<OrderMapper, Order> implements IOrderService {

    @Autowired
    private RemoteStorageService remoteStorageService;

    @Autowired
    private RemoteAccountService remoteAccountService;

    @GlobalTransactional(name = "prex-create-order",rollbackFor = Exception.class)
    @Override
    public void createOrder(Order order) {
        log.info("当前 XID: {}", RootContext.getXID());
        log.info("下单开始,用户:{},商品:{},数量:{},金额:{}", order.getUserId(), order.getProductId(), order.getCount(), order.getPayMoney());
        //创建订单
        order.setStatus(0);
        boolean save = save(order);
        log.info("保存订单{}", save ? "成功" : "失败");

        //远程调用库存服务扣减库存
        log.info("扣减库存开始");
        remoteStorageService.reduceCount(order.getProductId(), order.getCount());
        log.info("扣减库存结束");

        //远程调用账户服务扣减余额
        log.info("扣减余额开始");
        remoteAccountService.reduceBalance(order.getUserId(), order.getPayMoney());
        log.info("扣减余额结束");

        //修改订单状态为已完成
        log.info("修改订单状态开始");
        update(Wrappers.<Order>lambdaUpdate().set(Order::getStatus, 1).eq(Order::getUserId, order.getUserId()));
        log.info("修改订单状态结束");

        log.info("下单结束");
    }
}

我们只需要使用一个 @GlobalTransactional 注解在业务方法上就可以实现分布式事务

@GlobalTransactional
    public void purchase(String userId, String commodityCode, int orderCount) {
        ......
    }

不涉及到分布式事务的场景下,我们使用spring管理本地事务使用@Transactional注解,Seata使用的是@GlobalTransactional注解,那么我们就从@GlobalTransactional注解入手, 它是在GlobalTransactionalInterceptor中被拦截处理

file
file

最终执行的是TransactionalTemplate方法

public class TransactionalTemplate {

    public Object execute(TransactionalExecutor business) throws Throwable {
    // 1. 获取当前全局事务实例或创建新的实例
        GlobalTransaction tx = GlobalTransactionContext.getCurrentOrCreate();
        //1.1 获取事务信息
        TransactionInfo txInfo = business.getTransactionInfo();
        if (txInfo == null) {
            throw new ShouldNeverHappenException("transactionInfo does not exist");
        } else {
            Object var5;
            try {
            // 2. 开启全局事务
                this.beginTransaction(txInfo, tx);
                Object rs = null;

                try {
                // 调用业务服务
                    rs = business.execute();
                } catch (Throwable var9) {
                //3.回滚所需的业务异常。 this.completeTransactionAfterThrowing(txInfo, tx, var9);
                    throw var9;
                }
                //  没有异常 提交事务
                this.commitTransaction(tx);
                var5 = rs;
            } finally {
                // 清除
                this.triggerAfterCompletion();
                this.cleanUp();
            }

            return var5;
        }
    }}

这个方法好像是整个全局分布式事务的流程

  1. 开始分布式事务 这里最终会调用DefaultGlobalTransaction的begin方法
    file

 public void begin(int timeout, String name) throws TransactionException {
        // a 
        //此处的角色判断有关键的作用
        //表明当前是全局事务的发起者(Launcher)还是参与者(Participant)
        //如果在分布式事务的下游系统方法中也加上GlobalTransactional注解
        //那么它的角色就是Participant,即会忽略后面的begin就退出了
        //而判断是发起者(Launcher)还是参与者(Participant)是根据当前上下文是否已存在XID来判断
        //没有XID的就是Launcher,已经存在XID的就是Participant
        if (this.role != GlobalTransactionRole.Launcher) {
            this.check();
            if (LOGGER.isDebugEnabled()) {
                LOGGER.debug("Ignore Begin(): just involved in global transaction [" + this.xid + "]");
            }

        } else if (this.xid != null) {
            throw new IllegalStateException();
        } else if (RootContext.getXID() != null) {
            throw new IllegalStateException();
        } else {
        //  b 利用TmRpcClient 之前建立好的channel给tc发送请求,获取获取TC返回的全局事务XID
            this.xid = this.transactionManager.begin((String)null, (String)null, name, timeout);
            this.status = GlobalStatus.Begin;
            // c
            RootContext.bind(this.xid);
            if (LOGGER.isInfoEnabled()) {
                LOGGER.info("Begin new global transaction [" + this.xid + "]");
            }

        }
    }
  • a.判断是不是分布式事务的发起者,GlobalTransactionRole.Launcher就是事务发起者角色,如果不是就直接return
  • b.这个方法主要利用TmRpcClient 之前建立好的channel给tc发送请求,获取全局事务id
  • c.将获取到到全局事务id放到seata上下文中
  1. 处理方法本身的业务逻辑
  2. 处理业务逻辑的时候报错了,则进行事务回滚,并抛出异常

DefaultTransactionManager负责TM与TC通讯,发送begin、commit、rollback指令

至此拿到TC返回的XID一个全局事务就开启了,日志中也反应了上述流程

file

全局事务创建后,就开始执行business.execute(),即业务代码orderService.createOrder(order);;进入RM处理流程

RM处理流程

@RestController
@RequestMapping("/account")
public class AccountController {

    @Autowired
    private IAccountService accountService;


    @PostMapping("/reduceBalance")
    @ResponseBody
    public R reduceBalance(@RequestParam("userId") Integer userId, @RequestParam("money") BigDecimal money) throws Exception {
        return R.builder().success(accountService.reduceBalance(userId, money)).msg("扣款成功").build();
    }
}

account的接口和service方法并未出现seata相关的代码和注解,那么它是如何加入到这次全局事务中的呢,答案是ConnectionProxy中,这也是前面说为什么必须要使用DataSourceProxy的原因,通过DataSourceProxy才能在业务代码的事务提交时,seata通过这个切入点,来给TC发送rm的处理结果

由于业务代码本身的事务提交被ConnectionProxy代理,所以在提交本地事务时,实际执行的是ConnectionProxy的commit方法

public class ConnectionProxy extends AbstractConnectionProxy {
    public void commit() throws SQLException {
        try {
            LOCK_RETRY_POLICY.execute(() -> {
                this.doCommit();
                return null;
            });
        } catch (SQLException var2) {
            throw var2;
        } catch (Exception var3) {
            throw new SQLException(var3);
        }
    }

    private void doCommit() throws SQLException {
    //如果当前是全局事务,则执行全局事务的提交
    //判断是不是全局事务,就是看当前上下文是否存在XID
        if (this.context.inGlobalTransaction()) {
            this.processGlobalTransactionCommit();
        } else if (this.context.isGlobalLockRequire()) {
            this.processLocalCommitWithGlobalLocks();
        } else {
            this.targetConnection.commit();
        }

    }
    
     private void processGlobalTransactionCommit() throws SQLException {
        try {
        //首先是向TC注册RM,拿到TC分配的branchId
            this.register();
        } catch (TransactionException var2) {
            this.recognizeLockKeyConflictException(var2, this.context.buildLockKeys());
        }

        try {
            if (this.context.hasUndoLog()) {
               //写入undolog UndoLogManagerFactory.getUndoLogManager(this.getDbType()).flushUndoLogs(this);
            }
            //提交本地事务,可以看到写入undolog和业务数据是在同一个本地事务中
            this.targetConnection.commit();
        } catch (Throwable var3) {
            LOGGER.error("process connectionProxy commit error: {}", var3.getMessage(), var3);
            //向TC发送rm的事务处理失败的通知
            this.report(false);
            throw new SQLException(var3);
        }
        //向TC发送rm的事务处理成功的通知
        this.report(true);
        this.context.reset();
    }
    
    // 注册RM,构建request通过netty向TC发送指令
    //将返回的branchId存在上下文中
    private void register() throws TransactionException {
        Long branchId = DefaultResourceManager.get().branchRegister(BranchType.AT, this.getDataSourceProxy().getResourceId(), (String)null, this.context.getXid(), (String)null, this.context.buildLockKeys());
        this.context.setBranchId(branchId);
    }
}
  1. 获取business-service传来的XID
  2. 绑定XID到当前上下文中
  3. 执行业务逻辑sql
  4. 向TC创建本次RM的Netty连接
  5. 向TC发送分支事务的相关信息
  6. 获得TC返回的branchId
  7. 记录Undo Log数据 向TC发送本次事务PhaseOne阶段的处理结果
  8. 从当前上下文中解绑XID

其中第1步和第9步,是在FescarHandlerInterceptor中完成的,该类并不属于seata,而是spring-cloud-alibaba-seata中对feign、rest支持的实现。bind和unbind XID到上下文中。到这里RM完成了PhaseOne阶段的工作,接着看PhaseTwo阶段的处理逻辑

6. 事务提交

由于这次请求是正常流程无异常的,所以分支事务会正常commit。 在storage-service启动时创建了与TC通讯的Netty连接,TC在获取各RM的汇报结果后,就会给各RM发送commit或rollback的指令

file

从日志中可以看到

  1. 收到XID=130.252.9.221:8091:2028558491,branchId=2028558493的commit通知
  2. 执行commit动作
  3. 将commit结果发送给TC,branchStatus为PhaseTwo_Committed 具体看下执行commit的过程,在AbstractRMHandler类的doBranchCommit方法之前是接收TC消息包装处理路由的过程
io.seata.rm.AbstractRMHandler
//拿到通知的xid、branchId等关键参数
//然后调用RM的branchCommit
protected void doBranchCommit(BranchCommitRequest request, BranchCommitResponse response) throws TransactionException {
        String xid = request.getXid();
        long branchId = request.getBranchId();
        String resourceId = request.getResourceId();
        String applicationData = request.getApplicationData();
        if (LOGGER.isInfoEnabled()) {
            LOGGER.info("Branch committing: " + xid + " " + branchId + " " + resourceId + " " + applicationData);
        }

        BranchStatus status = this.getResourceManager().branchCommit(request.getBranchType(), xid, branchId, resourceId, applicationData);
        response.setXid(xid);
        response.setBranchId(branchId);
        response.setBranchStatus(status);
        if (LOGGER.isInfoEnabled()) {
            LOGGER.info("Branch commit result: " + status);
        }

    }

最终会将branceCommit的请求调用到AsyncWorker的branchCommit方法。AsyncWorker的处理方式是seata架构的一个关键部分,大部分事务都是会正常提交的,所以在PhaseOne阶段就已经结束了,这样就可以将锁最快的释放。PhaseTwo阶段接收commit的指令后,异步处理即可。将PhaseTwo的时间消耗排除在一次分布式事务之外。

io.seata.rm.datasource.AsyncWorker
public class AsyncWorker implements ResourceManagerInbound {
    private static final Logger LOGGER = LoggerFactory.getLogger(AsyncWorker.class);
    private static final int DEFAULT_RESOURCE_SIZE = 16;
    private static final int UNDOLOG_DELETE_LIMIT_SIZE = 1000;
    private static int ASYNC_COMMIT_BUFFER_LIMIT = ConfigurationFactory.getInstance().getInt("client.async.commit.buffer.limit", 10000);
    private static final BlockingQueue<AsyncWorker.Phase2Context> ASYNC_COMMIT_BUFFER;
    private static ScheduledExecutorService timerExecutor;

    public AsyncWorker() {
    }

    public BranchStatus branchCommit(BranchType branchType, String xid, long branchId, String resourceId, String applicationData) throws TransactionException {
        if (!ASYNC_COMMIT_BUFFER.offer(new AsyncWorker.Phase2Context(branchType, xid, branchId, resourceId, applicationData))) {
            LOGGER.warn("Async commit buffer is FULL. Rejected branch [" + branchId + "/" + xid + "] will be handled by housekeeping later.");
        }

        return BranchStatus.PhaseTwo_Committed;
    }

    //通过一个定时任务消费list中的待提交XID
    public synchronized void init() {
        LOGGER.info("Async Commit Buffer Limit: " + ASYNC_COMMIT_BUFFER_LIMIT);
        timerExecutor = new ScheduledThreadPoolExecutor(1, new NamedThreadFactory("AsyncWorker", 1, true));
        timerExecutor.scheduleAtFixedRate(new Runnable() {
            public void run() {
                try {
                    AsyncWorker.this.doBranchCommits();
                } catch (Throwable var2) {
                    AsyncWorker.LOGGER.info("Failed at async committing ... " + var2.getMessage());
                }

            }
        }, 10L, 1000L, TimeUnit.MILLISECONDS);
    }

    private void doBranchCommits() {
        if (ASYNC_COMMIT_BUFFER.size() != 0) {
            HashMap mappedContexts;
            AsyncWorker.Phase2Context commitContext;
            Object contextsGroupedByResourceId;
            for(mappedContexts = new HashMap(16); !ASYNC_COMMIT_BUFFER.isEmpty(); ((List)contextsGroupedByResourceId).add(commitContext)) {
                commitContext = (AsyncWorker.Phase2Context)ASYNC_COMMIT_BUFFER.poll();
                contextsGroupedByResourceId = (List)mappedContexts.get(commitContext.resourceId);
                if (contextsGroupedByResourceId == null) {
                    contextsGroupedByResourceId = new ArrayList();
                    mappedContexts.put(commitContext.resourceId, contextsGroupedByResourceId);
                }
            }

            Iterator var28 = mappedContexts.entrySet().iterator();

            while(true) {
         //一次定时任务取出ASYNC_COMMIT_BUFFER中的所有待办数据
        //以resourceId作为key分组待办数据,resourceId就是一个数据库的连接url
        //在前面的日志中可以看到,目的是为了覆盖应用的多数据源问题
                if (var28.hasNext()) {
                    Entry<String, List<AsyncWorker.Phase2Context>> entry = (Entry)var28.next();
                    Connection conn = null;

                    try {
                        DataSourceProxy dataSourceProxy;
                        try {       //根据resourceId获取数据源以及连接
                            DataSourceManager resourceManager = (DataSourceManager)DefaultResourceManager.get().getResourceManager(BranchType.AT);
                            dataSourceProxy = resourceManager.get((String)entry.getKey());
                            if (dataSourceProxy == null) {
                                throw new ShouldNeverHappenException("Failed to find resource on " + (String)entry.getKey());
                            }

                            conn = dataSourceProxy.getPlainConnection();
                        } catch (SQLException var26) {
                            LOGGER.warn("Failed to get connection for async committing on " + (String)entry.getKey(), var26);
                            continue;
                        }

                        List<AsyncWorker.Phase2Context> contextsGroupedByResourceId = (List)entry.getValue();
                        Set<String> xids = new LinkedHashSet(1000);
                        Set<Long> branchIds = new LinkedHashSet(1000);
                        Iterator var9 = contextsGroupedByResourceId.iterator();

                        while(var9.hasNext()) {
                            AsyncWorker.Phase2Context commitContext = (AsyncWorker.Phase2Context)var9.next();
                            xids.add(commitContext.xid);
                            branchIds.add(commitContext.branchId);
                            int maxSize = xids.size() > branchIds.size() ? xids.size() : branchIds.size();
                            if (maxSize == 1000) {
                                try {
                                    UndoLogManagerFactory.getUndoLogManager(dataSourceProxy.getDbType()).batchDeleteUndoLog(xids, branchIds, conn);
                                } catch (Exception var25) {
                                    LOGGER.warn("Failed to batch delete undo log [" + branchIds + "/" + xids + "]", var25);
                                }

                                xids.clear();
                                branchIds.clear();
                            }
                        }

                        if (!CollectionUtils.isEmpty(xids) && !CollectionUtils.isEmpty(branchIds)) {
                            try {
                               
                               //执行undolog的处理,即删除xid、branchId对应的记录 UndoLogManagerFactory.getUndoLogManager(dataSourceProxy.getDbType()).batchDeleteUndoLog(xids, branchIds, conn);
                            } catch (Exception var24) {
                                LOGGER.warn("Failed to batch delete undo log [" + branchIds + "/" + xids + "]", var24);
                            }
                            continue;
                        }
                    } finally {
                        if (conn != null) {
                            try {
                                conn.close();
                            } catch (SQLException var23) {
                                LOGGER.warn("Failed to close JDBC resource while deleting undo_log ", var23);
                            }
                        }

                    }

                    return;
                }

                return;
            }
        }
    }

    public BranchStatus branchRollback(BranchType branchType, String xid, long branchId, String resourceId, String applicationData) throws TransactionException {
        throw new NotSupportYetException();
    }

    static {
        ASYNC_COMMIT_BUFFER = new LinkedBlockingQueue(ASYNC_COMMIT_BUFFER_LIMIT);
    }

}

所以对于commit动作的处理,RM只需删除xid、branchId对应的undolog既可

7. 事务回滚

对于rollback场景的触发有两种情况:

  1. 分支事务处理异常,即io.seata.rm.datasource.ConnectionProxy中report(false)的情况
  2. TM捕获到下游系统上抛的异常,即发起全局事务标有@GlobalTransactional注解的方法捕获到的异常。在前面io.seata.tm.api.TransactionalTemplate类的execute模版方法中,对business.execute()的调用进行了catch,catch后会调用rollback,由TM通知TC对应XID需要回滚事务
public void rollback() throws TransactionException {
        //只有Launcher能发起这个rollback
        if (this.role == GlobalTransactionRole.Participant) {
        // Participant has no responsibility of committing
            if (LOGGER.isDebugEnabled()) {
                LOGGER.debug("Ignore Rollback(): just involved in global transaction [" + this.xid + "]");
            }

        } else if (this.xid == null) {
            throw new IllegalStateException();
        } else {
            int retry = ROLLBACK_RETRY_COUNT;

            try {
                while(retry > 0) {
                    try {
                        this.status = this.transactionManager.rollback(this.xid);
                        break;
                    } catch (Throwable var6) {
                        LOGGER.error("Failed to report global rollback [{}],Retry Countdown: {}, reason: {}", new Object[]{this.getXid(), retry, var6.getMessage()});
                        --retry;
                        if (retry == 0) {
                            throw new TransactionException("Failed to report global rollback", var6);
                        }
                    }
                }
            } finally {
                if (RootContext.getXID() != null && this.xid.equals(RootContext.getXID())) {
                    RootContext.unbind();
                }

            }

            if (LOGGER.isInfoEnabled()) {
                LOGGER.info("[" + this.xid + "] rollback status:" + this.status);
            }

        }
    }

TC汇总后向参与者发送rollback指令,RM在AbstractRMHandler类的doBranchRollback方法中接收这个rollback的通知

protected void doBranchRollback(BranchRollbackRequest request, BranchRollbackResponse response) throws TransactionException {
        String xid = request.getXid();
        long branchId = request.getBranchId();
        String resourceId = request.getResourceId();
        String applicationData = request.getApplicationData();
        if (LOGGER.isInfoEnabled()) {
            LOGGER.info("Branch Rollbacking: " + xid + " " + branchId + " " + resourceId);
        }

        BranchStatus status = this.getResourceManager().branchRollback(request.getBranchType(), xid, branchId, resourceId, applicationData);
        response.setXid(xid);
        response.setBranchId(branchId);
        response.setBranchStatus(status);
        if (LOGGER.isInfoEnabled()) {
            LOGGER.info("Branch Rollbacked result: " + status);
        }

    }

然后将rollback请求传递到io.seata.rm.datasource.DataSourceManager类的branchRollback方法

public BranchStatus branchRollback(BranchType branchType, String xid, long branchId, String resourceId, String applicationData) throws TransactionException {
        //根据resourceId获取对应的数据源
        DataSourceProxy dataSourceProxy = this.get(resourceId);
        if (dataSourceProxy == null) {
            throw new ShouldNeverHappenException();
        } else {
            try {
                UndoLogManagerFactory.getUndoLogManager(dataSourceProxy.getDbType()).undo(dataSourceProxy, xid, branchId);
            } catch (TransactionException var9) {
                if (LOGGER.isInfoEnabled()) {
                    LOGGER.info("branchRollback failed reason [{}]", var9.getMessage());
                }

                if (var9.getCode() == TransactionExceptionCode.BranchRollbackFailed_Unretriable) {
                    return BranchStatus.PhaseTwo_RollbackFailed_Unretryable;
                }

                return BranchStatus.PhaseTwo_RollbackFailed_Retryable;
            }

            return BranchStatus.PhaseTwo_Rollbacked;
        }
    }

最终会执行io.seata.rm.datasource.undo.UndoLogManager类的undo方法,这个undo,其实就是通过undolog来反向生成一个回滚sql,然后执行这个回滚sql来达到rollback的效果

public void undo(DataSourceProxy dataSourceProxy, String xid, long branchId) throws TransactionException {
        Connection conn = null;
        ResultSet rs = null;
        PreparedStatement selectPST = null;
        boolean originalAutoCommit = true;

        while(true) {
            try {
                conn = dataSourceProxy.getPlainConnection();
                if (originalAutoCommit =
                conn.getAutoCommit()) {
                // 整个撤消过程应在本地事务中运行
                    conn.setAutoCommit(false);
                }

                selectPST = conn.prepareStatement(SELECT_UNDO_LOG_SQL);
                selectPST.setLong(1, branchId);
                selectPST.setString(2, xid);
                rs = selectPST.executeQuery();
                boolean exists = false;

                while(rs.next()) {
                    exists = true;
                    int state = rs.getInt("log_status");
                    if (!canUndo(state)) {
                        if (LOGGER.isInfoEnabled()) {
                            LOGGER.info("xid {} branch {}, ignore {} undo_log", new Object[]{xid, branchId, state});
                        }

                        return;
                    }

                    String contextString = rs.getString("context");
                    Map<String, String> context = this.parseContext(contextString);
                    Blob b = rs.getBlob("rollback_info");
                    byte[] rollbackInfo = BlobUtils.blob2Bytes(b);
                    String serializer = context == null ? null : (String)context.get("serializer");
                    UndoLogParser parser = serializer == null ? UndoLogParserFactory.getInstance() : UndoLogParserFactory.getInstance(serializer);
                    BranchUndoLog branchUndoLog = parser.decode(rollbackInfo);

                    try {
                        setCurrentSerializer(parser.getName());
                        List<SQLUndoLog> sqlUndoLogs = branchUndoLog.getSqlUndoLogs();
                        if (sqlUndoLogs.size() > 1) {
                            Collections.reverse(sqlUndoLogs);
                        }

                        Iterator var19 = sqlUndoLogs.iterator();

                        while(var19.hasNext()) {
                            SQLUndoLog sqlUndoLog = (SQLUndoLog)var19.next();
                            TableMeta tableMeta = TableMetaCacheFactory.getTableMetaCache(dataSourceProxy).getTableMeta(dataSourceProxy, sqlUndoLog.getTableName());
                            sqlUndoLog.setTableMeta(tableMeta);
                            AbstractUndoExecutor undoExecutor = UndoExecutorFactory.getUndoExecutor(dataSourceProxy.getDbType(), sqlUndoLog);
                            undoExecutor.executeOn(conn);
                        }
                    } finally {
                        removeCurrentSerializer();
                    }
                }

                if (exists) {
                    this.deleteUndoLog(xid, branchId, conn);
                    conn.commit();
                    if (LOGGER.isInfoEnabled()) {
                        LOGGER.info("xid {} branch {}, undo_log deleted with {}", new Object[]{xid, branchId, AbstractUndoLogManager.State.GlobalFinished.name()});
                        break;
                    }
                } else {
                    this.insertUndoLogWithGlobalFinished(xid, branchId, UndoLogParserFactory.getInstance(), conn);
                    conn.commit();
                    if (LOGGER.isInfoEnabled()) {
                        LOGGER.info("xid {} branch {}, undo_log added with {}", new Object[]{xid, branchId, AbstractUndoLogManager.State.GlobalFinished.name()});
                        break;
                    }
                }

                return;
            } catch (SQLIntegrityConstraintViolationException var44) {
                if (LOGGER.isInfoEnabled()) {
                    LOGGER.info("xid {} branch {}, undo_log inserted, retry rollback", xid, branchId);
                }
            } catch (Throwable var45) {
                if (conn != null) {
                    try {
                        conn.rollback();
                    } catch (SQLException var42) {
                        LOGGER.warn("Failed to close JDBC resource while undo ... ", var42);
                    }
                }

                throw new BranchTransactionException(TransactionExceptionCode.BranchRollbackFailed_Retriable, String.format("Branch session rollback failed and try again later xid = %s branchId = %s %s", xid, branchId, var45.getMessage()), var45);
            } finally {
                try {
                    if (rs != null) {
                        rs.close();
                    }

                    if (selectPST != null) {
                        selectPST.close();
                    }

                    if (conn != null) {
                        if (originalAutoCommit) {
                            conn.setAutoCommit(true);
                        }

                        conn.close();
                    }
                } catch (SQLException var41) {
                    LOGGER.warn("Failed to close JDBC resource while undo ... ", var41);
                }

            }
        }

    }
  1. 根据xid和branchId查找PhaseOne阶段提交的undolog
  2. 如果找到了就根据undolog中记录的数据生成回放sql并执行,即还原PhaseOne阶段修改的数据
  3. 第2步处理完后,删除该条undolog数据
  4. 如果第1步没有找到对应的undolog,就插入一条状态为GlobalFinished的undolog. 出现没找到的原因可能是PhaseOne阶段的本地事务异常了,导致没有正常写入。因为xid和branchId是唯一索引,所以第4步的插入,可以防止PhaseOne阶段后续又写入成功,那么PhaseOne阶段就会异常,这样业务数据也是没有提交成功的,数据最终是回滚了的效果

参考资料:
t.im/kzf3
t.im/kzf5

开源项目推荐

作者的开源项目推荐:

关注公众号回复开源项目即可获取