.NET分布式事务未提交造成6107错误或系统被挂起的问题分析定位

问题描述：

系统中多个功能不定期出现“Unable to get error message (6107) (0).”错误，即分布式事务超时，但报出错误的部分功能根本没有使用分布式事务。

原因分析：

推测是存在分布式事务未提交的情况，回到线程池后被复用造成的，例如：

系统中A功能存在分布式事务未提交的问题，处理A功能的线程执行完成后回到线程池，B功能复用了处理A功能的线程，A功能未提交的分布式事务将B功能的代码也包含进去了，依次类推可能还有C、D等功能，直到A功能开启的分布式事务超时报出6107错误，而报错的B或C、D等功能本身是没有问题的。

Demo验证：

   1:      ThreadPool.SetMinThreads(1, 1);

   2:      ThreadPool.SetMaxThreads(1, 1);

   3:   

   4:      ThreadPool.QueueUserWorkItem((x) =>

   5:      {

   6:          Console.WriteLine("33: " + Thread.CurrentThread.ManagedThreadId);

   7:          TransactionScope ts = new TransactionScope(TransactionScopeOption.Required, new TimeSpan(0, 0, 10));

   8:          using (OracleConnection conn = new OracleConnection(connstr))

   9:          {

  10:              conn.Open();

  11:              var cmd = conn.CreateCommand();

  12:              cmd.CommandText = @"update workitem set name = 'yyxx' where workitemid = '64365dc4-41d7-4057-af68-012d552660e2'";

  13:              cmd.ExecuteNonQuery();

  14:              conn.Close();

  15:          }

  16:          Thread.Sleep(100);

  17:          Console.WriteLine("33: end");

  18:      });

  19:              

  20:      ThreadPool.QueueUserWorkItem((x) => {

  21:          Console.WriteLine("44: " + Thread.CurrentThread.ManagedThreadId);

  22:          try

  23:          {

  24:              while (true)

  25:              {

  26:                  using (OracleConnection conn = new OracleConnection(connstr))

  27:                  {

  28:                      conn.Open();

  29:                      var cmd = conn.CreateCommand();

  30:                      cmd.CommandText = @"update workitem set name = 'xxyy' where workitemid = '68265741-4433-42e9-b7e7-63ae1ce8bbc8'";

  31:                      cmd.ExecuteNonQuery();

  32:                      conn.Close();

  33:                  }

  34:                  Thread.Sleep(100);

  35:              }

  36:          }

  37:          catch (Exception e)

  38:          {

  39:              Console.WriteLine(e);

  40:          }

  41:          Console.WriteLine("44: end");

  42:      });

解决方法：

使用HttpModule 分别在beginRequest和endRequest中判断是否已包含了事务上下文

如果beginRequest中没有事务，endRequest时有未提交事务，则说明当前请求对应的是问题源头；

如果beginRequest和endRequest都有，说明当前请求是线程复用后的受害者。

示例如下：

   1:          protected void context_EndRequest(object sender, EventArgs e)

   2:          {

   3:              Debug.WriteLine(Thread.CurrentThread.ManagedThreadId);

   4:              Type t = typeof(Transaction);

   5:              FieldInfo field = t.GetField("complete", BindingFlags.Instance | BindingFlags.NonPublic);

   6:              if (Transaction.Current != null)

   7:              {

   8:                  bool isComplete = (bool) field.GetValue(Transaction.Current);

   9:                  if (!isComplete)

  10:                  {

  11:                      Debug.WriteLine(

  12:                          string.Format(@"Creation Time: {0}\r\nDistributedIdentifier:{1}\r\nLocalIdentifier:{2}",

  13:                              Transaction.Current.TransactionInformation.CreationTime,

  14:                              Transaction.Current.TransactionInformation.DistributedIdentifier,

  15:                              Transaction.Current.TransactionInformation.LocalIdentifier));

  16:                      //Transaction.Current.Rollback();

  17:                      //Transaction.Current.Dispose();

  18:                  }

  19:              }

  20:          }