Thanks for your reply,Sahil Malik.
I think the only one MSDTC is used(which is in application
server),MSDTC
communicate with RMs of different databases.
Here is my code below:
public interface IDistributeTransaction
{
int StartTransaction();
int CommitTransaction();
int RollBackTranstion();
int ExecuteSql(string connectionstring,string SqlStr,out string ErrStr);
}
public class AppDistributeTrans:MarshalByRefObject,IDistributeTransaction
{
//It stores DistributeTransaction Instance;
private static ArrayList m_DistributeTransList;
//The exact Instance which created when the first time remote machine
called
.
private DistributeTransaction m_DistributeTrans
public AppDistributeTrans()
{
//Create or find the proper Instance of DistributeTransaction,each
Instance have a different ID,so I can find it.
............
}
#region IDistributeTransaction ³ÉÔ±
int ShangHaiERP.Corp.IDistributeTransaction.StartTransaction()
{
return m_DistributeTrans.StartTransaction();
}
int ShangHaiERP.Corp.IDistributeTransaction.RollBackTranstion()
{
int rt = m_DistributeTrans.RollBackTranstion();
m_DistributeTrans = null;
return rt;
}
int ShangHaiERP.Corp.IDistributeTransaction.CommitTransaction()
{
int rt = 0;
string err = "";
try
{
rt = m_DistributeTrans.CommitTransaction();
}
catch(Exception ex)
{
err = ex.Message + ex.StackTrace + ex.Source;
}
m_DistributeTrans = null;
return rt;
}
int ShangHaiERP.Corp.IDistributeTransaction.ExecuteSql(string
connectionstring,string SqlStr,out string ErrStr)
{
return m_DistributeTrans.ExecuteSql(connectionstr,SqlStr,out
ErrStr);
}
#endregion
}
[ Transaction( TransactionOption.RequiresNew) ]
[ ObjectPooling(true, 5, 10) ]
public class DistributeTransaction:ServicedComponent
{
.......
public DistributeTransaction()
{
}
#region IDistributeTransaction
//Just do nothing,I don't want to autocomplete,programmer can use it just
like ADO.Net's transaction.
public int StartTransaction()
{
AddLogEx(1,m_LogFileName,"StartTrans" +
ContextUtil.TransactionId.ToString());
return 0;
}
public int RollBackTranstion()
{
AddLogEx(1,m_LogFileName,"RollBack" +
ContextUtil.TransactionId.ToString());
ContextUtil.SetAbort();
return 0;
}
public int CommitTransaction()
{
AddLogEx(1,m_LogFileName,ContextUtil.TransactionId.ToString());
ContextUtil.SetComplete();
return 0;
}
public int ExecuteSql(string connectionstr, string SqlStr, out string
ErrStr)
{
ErrStr = "";
try
{
AddLogEx(1,m_LogFileName,"Exe1" +
ContextUtil.TransactionId.ToString());
OleDbConnection cn = new OleDbConnection(connectionstr);
OleDbCommand cm = new OleDbCommand(SqlStr);
cm.Connection = cn;
cn.Open();
cm.ExecuteNonQuery();
cn.Close();
}
catch(OleDbException ex)
{
AddLogEx(1,m_LogFileName,ex.Message);
return DistributeTransaction.ERROR_SQL;
}
catch(Exception ex)
{
AddLogEx(1,m_LogFileName,ex.Message);
return DistributeTransaction.ERROR_OTHER;
}
return 0;
}
private bool AddLogEx(int LogLevel,string UserName,string LogStr)
{
string FileName;
string TmpStr,LogFileDir;
StreamWriter Sw;
LogFileDir = m_LogPath;
Directory.CreateDirectory(LogFileDir);
switch (LogLevel)
{
case 0:
FileName = LogFileDir + UserName + "_DBLOG";
break;
case 1:
FileName = LogFileDir + UserName + "_LOG";
break;
default:
goto case 1;
}
try
{
FileInfo Fi = new FileInfo(FileName);
if (Fi.Exists == false)
{
FileStream Fs = new FileStream(FileName,FileMode.Create);
Fs.Close();
}
if (Fi.Length > this.m_FileMaxSize)
{
Fi.CopyTo(FileName + "_" +
DateTime.Now.ToString("yyyyMMddHHmm"),true);
Fi.Delete();
}
}
catch
{
}
Sw = new StreamWriter(FileName,true);
try
{
TmpStr = DateTime.Now.ToString("yyyy-MM-dd HH:mm:ss # ") + LogStr;
Sw.WriteLine(TmpStr);
Sw.Close();
}
catch (Exception e)
{
TmpStr = e.Message;
return false;
}
return true;
}
#endregion
}
"Sahil Malik" <
[email protected]> дÈëÓʼþ
Oh my, so we have remoting, msdtc on the two appdomains possibly two
machines involving remoting, and we have two databases involving a
distributed transaction between them.
My first kneejerk reaction is to try and find a simpler solution to what
you
are trying to acheive. MSDTC transactions by default use
IsolationLevel.Serializable between databases, in Sql Server 2005 and
.NET
2.0 combination, this is slightly better in the sense that the isolation
level is promoted to serializable once the transaction enlists a
second
database. But still it ends up becoming Serializable which is a complete
pig.
The second reason for my recommendation to change your architecture is
the
heavy duty traffic you will generate due to two MSDTC instances on two
machines.
The third reason is the pain you will have to go through as far as
deployment of your application is concerned.
And the final reason is System.Transactions in .NET 2.0 will subsume
transactional capabilities of System.EnterpriseServices, but maybe you
are
writing the below code for .NET 2.0 as nothing in your post seems to
contradict that. Anyway, the above 3 reasons still hold good.
If you decide to change the architecture a bit, please let us know
what
exactly are you trying to acheive and we could suggest something. Can the
two databases talk to each other? viz. can you setup a remote login
between
them and do a BEGIN DISTRIBUTED TRANSACTION ... or ... use linked tables
to
acheive the same?
But if for some reason you have compelling reasons to use MSDTC, I would
need to see how exactly is your class that implements
IDistributeTransaction
and m_DistributeTrans is an instance of, ... how exactly is that
enlisting
itself the transactions.
- Sahil Malik
http://dotnetjunkies.com/weblog/sahilmalik
http://blogs.apress.com/authors.php?author=Sahil Malik
I want to update MultiDatabase with Microsoft's MSDTC.Just like ADO.Net
provided,
I designed an interface below:
public interface IDistributeTransaction
{
int StartTransaction();
int CommitTransaction();
int RollBackTranstion();
int ExecuteSql(string connectionstring,string SqlStr,out string
ErrStr);
}
I called the Interface by Remoting.
int rt = m_DistributeTrans.StartTransaction();
if(rt == 0)//sucess
{
rt = ExecuteSql(....);//Update A database;
if(rt == 0)//sucess
{
rt = m_DistributeTrans.ExecuteSql(....);//Update B database
}
if(rt == 0)//sucess
m_DistributeTrans.CommitTransaction();//Commit;
else //failed
DistributeTrans.RollBackTranstion();//Rollback;
}
But when Rollback,the A database is updated sucessful. The
ContextUtil.TransactionId is the same during the whole Transaction. Why
this
happened?
