Seata AT模式TransactionHook被删除探究
作者:梦想实现家_Z
前言
兄弟们,刚刚又给seata社区修了一个BUG,有用户提了issue反应TransactionHook在某些情况下不会被调用:

相关issue链接:github.com/seata/seata…,该用户在issue中已经指出了相关问题所在:

下面我们来看一下到底是什么原因导致了上述BUG的产生。
问题定位
根据用户的反馈,我们找到目标源码io.seata.tm.api.TransactionalTemplate#execute():
try {
// 开启分布式事务,获取XID
beginTransaction(txInfo, tx);
Object rs;
try {
// 执行业务代码
rs = business.execute();
} catch (Throwable ex) {
// 3. 处理异常,准备回滚.
completeTransactionAfterThrowing(txInfo, tx, ex);
throw ex;
}
// 4. 提交事务.
commitTransaction(tx, txInfo);
return rs;
} finally {
//5. 回收现场
resumeGlobalLockConfig(previousConfig);
triggerAfterCompletion();
cleanUp();
}
问题代码就出在cleanUp()中,我们来看一下里面做了什么操作,最终我们定位到:
public final class TransactionHookManager {
private static final ThreadLocal<List<TransactionHook>> LOCAL_HOOKS = new ThreadLocal<>();
// 注册TransactionHook
public static void registerHook(TransactionHook transactionHook) {
if (transactionHook == null) {
throw new NullPointerException("transactionHook must not be null");
}
List<TransactionHook> transactionHooks = LOCAL_HOOKS.get();
if (transactionHooks == null) {
LOCAL_HOOKS.set(new ArrayList<>());
}
LOCAL_HOOKS.get().add(transactionHook);
}
// 移除当前线程上所有TransactionHook
public static void clear() {
LOCAL_HOOKS.remove();
}
}
由上面的源码可知,cleanUp()操作时把当前线程中的所有TransactionHook都清除掉了。也就是说,假如事务A和事务B共用同一个线程,当事务B处理完毕后,调用了cleanUp()回收现场时,把该线程当中存储的所有TransactionHook全部清除掉了,导致事务A的生命周期中找不到该事务对应的TransactionHook,从而产生了BUG。
如何解决
通过与seata社区的大佬不断地沟通,最终敲定以下方案:
1.改造TransactionHookManager.LOCAL_HOOKS,把数据类型改成ThreadLocal<Map<String, List<TransactionHook>>>,Map中的key对应分布式事务XID;
2.针对当前上下文中没有XID,那么key就为null,因为HashMap允许key为null;
3.当用户查询指定XID下的hook时,连同key为null对应的hook也一起返回;
- 第一步比较好理解,因为事务A和事务B对应的
TransactionHook没有被区分出来,所以造成了清理事务B的TransactionHook时连同事务A的TransactionHook一起被清除,那么我们修改数据结构来区分事务A和事务B的TransactionHook,以便清理的时候不会造成误删;
第二步为什么要针对没有XID的时候也要能设置TransactionHook,因为有这么一段代码:
private void beginTransaction(TransactionInfo txInfo, GlobalTransaction tx) throws TransactionalExecutor.ExecutionException {
try {
// 执行triggerBeforeBegin()
triggerBeforeBegin();
// 注册分布式事务,生成XID
tx.begin(txInfo.getTimeOut(), txInfo.getName());
// 执行triggerAfterBegin()
triggerAfterBegin();
} catch (TransactionException txe) {
throw new TransactionalExecutor.ExecutionException(tx, txe,
TransactionalExecutor.Code.BeginFailure);
}
}
上面的代码会产生一个问题,因为我们的TransactionHook依赖于XID,但是triggerBeforeBegin()执行的时候还没有产生XID,所以为了能够在没有XID的时候也能够让TransactionHook生效,我们要有一个虚值key来临时设置TransactionHook;
第三步的设计时为了在第二步的基础上,当事务开启后获取XID后,要保证XID获取前注册的TransactionHook也要生效,我们在通过XID查询TransactionHook时要把虚值key对应的TransactionHook也一起返回;
注意事项
在实际代码修改中,发现triggerAfterCommit()、triggerAfterRollback()、triggerAfterCompletion()在被调用时始终拿不到对应的TransactionHook,最终debug下来发现在调用这三个方法前,上下文中的XID被解绑了,导致拿到的XID为空。代码类似下面这样:
try {
// 调用triggerBeforeCommit()
triggerBeforeCommit();
// 提交事务,清除XID
tx.commit();
if (Arrays.asList(GlobalStatus.TimeoutRollbacking, GlobalStatus.TimeoutRollbacked).contains(tx.getLocalStatus())) {
throw new TransactionalExecutor.ExecutionException(tx,
new TimeoutException(String.format("Global transaction[%s] is timeout and will be rollback[TC].", tx.getXid())),
TransactionalExecutor.Code.TimeoutRollback);
}
// 调用triggerAfterCommit()
triggerAfterCommit();
} catch (TransactionException txe) {
// 4.1 Failed to commit
throw new TransactionalExecutor.ExecutionException(tx, txe,
TransactionalExecutor.Code.CommitFailure);
}
不过经过我的一番查找,发现GlobalTransaction中是包含XID属性的,所以果断从GlobalTransaction对象中取XID传进来。
修改后的代码如下:
try {
// 调用triggerBeforeCommit()
triggerBeforeCommit();
// 提交事务,清除XID
tx.commit();
if (Arrays.asList(GlobalStatus.TimeoutRollbacking, GlobalStatus.TimeoutRollbacked).contains(tx.getLocalStatus())) {
throw new TransactionalExecutor.ExecutionException(tx,
new TimeoutException(String.format("Global transaction[%s] is timeout and will be rollback[TC].", tx.getXid())),
TransactionalExecutor.Code.TimeoutRollback);
}
// 调用triggerAfterCommit()
triggerAfterCommit(tx.getXid());
} catch (TransactionException txe) {
// 4.1 Failed to commit
throw new TransactionalExecutor.ExecutionException(tx, txe,
TransactionalExecutor.Code.CommitFailure);
}
改造后的TransactionHookManager
public final class TransactionHookManager {
private TransactionHookManager() {
}
private static final ThreadLocal<Map<String, List<TransactionHook>>> LOCAL_HOOKS = new ThreadLocal<>();
/**
* get the current hooks
*
* @return TransactionHook list
*/
public static List<TransactionHook> getHooks() {
String xid = RootContext.getXID();
return getHooks(xid);
}
/**
* get hooks by xid
*
* @param xid
* @return TransactionHook list
*/
public static List<TransactionHook> getHooks(String xid) {
Map<String, List<TransactionHook>> hooksMap = LOCAL_HOOKS.get();
if (hooksMap == null || hooksMap.isEmpty()) {
return Collections.emptyList();
}
List<TransactionHook> hooks = new ArrayList<>();
List<TransactionHook> localHooks = hooksMap.get(xid);
if (StringUtils.isNotBlank(xid)) {
List<TransactionHook> virtualHooks = hooksMap.get(null);
if (virtualHooks != null && !virtualHooks.isEmpty()) {
hooks.addAll(virtualHooks);
}
}
if (localHooks != null && !localHooks.isEmpty()) {
hooks.addAll(localHooks);
}
if (hooks.isEmpty()) {
return Collections.emptyList();
}
return Collections.unmodifiableList(hooks);
}
/**
* add new hook
*
* @param transactionHook transactionHook
*/
public static void registerHook(TransactionHook transactionHook) {
if (transactionHook == null) {
throw new NullPointerException("transactionHook must not be null");
}
Map<String, List<TransactionHook>> hooksMap = LOCAL_HOOKS.get();
if (hooksMap == null) {
hooksMap = new HashMap<>();
LOCAL_HOOKS.set(hooksMap);
}
String xid = RootContext.getXID();
List<TransactionHook> hooks = hooksMap.get(xid);
if (hooks == null) {
hooks = new ArrayList<>();
hooksMap.put(xid, hooks);
}
hooks.add(transactionHook);
}
/**
* clear hooks by xid
*
* @param xid
*/
public static void clear(String xid) {
Map<String, List<TransactionHook>> hooksMap = LOCAL_HOOKS.get();
if (hooksMap == null || hooksMap.isEmpty()) {
return;
}
hooksMap.remove(xid);
if (StringUtils.isNotBlank(xid)) {
hooksMap.remove(null);
}
}
}以上就是Seata AT模式TransactionHook被删除探究的详细内容,更多关于Seata AT删除TransactionHook的资料请关注脚本之家其它相关文章!
