System.Transactions: timed out transactions crash host process

[UltimaDeveloper]

Hi. I am using Devart dotConnect provider for Oracle in Direct mode (version 6.80) in conjunction with System.Transactions classes. When network connection to Oracle is lost or Oracle session is killed (from an external application), I sometimes get the unhandled exception which crashes my process.

Apparently System.Transactions is doing something behind the scenes like rolling back timed out transactions every now and then. But dotConnect throws an exception when Rollback is called and the connection is in invalid state (i.e, session is killed or tcp connection to Oracle server is closed):

Message: ORA-03113: end-of-file on communication channel
at Devart.Data.Oracle.c2.a(Byte[] A_0, Int32 A_1, Int32 A_2)
at Devart.Data.Oracle.c2.a(Byte& A_0)
at Devart.Data.Oracle.cw.b(Byte[] A_0, Int32 A_1, Int32 A_2)
at Devart.Data.Oracle.cw.f()
at Devart.Data.Oracle.cp.c()
at Devart.Data.Oracle.as.c()
at Devart.Data.Oracle.OracleInternalConnection.Rollback()
at Devart.Common.r.b(Enlistment A_0)
at System.Transactions.VolatileEnlistmentAborting.EnterState(InternalEnlistment enlistment)
at System.Transactions.TransactionStateAborted.EnterState(InternalTransaction tx)
at System.Transactions.EnlistableStates.Timeout(InternalTransaction tx)
at System.Transactions.Bucket.TimeoutTransactions()
at System.Transactions.BucketSet.TimeoutTransactions()
at System.Transactions.TransactionTable.ThreadTimer(Object state)
at System.Threading.ExecutionContext.RunInternal(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
at System.Threading.ExecutionContext.Run(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
at System.Threading.TimerQueueTimer.CallCallback()
at System.Threading.TimerQueueTimer.Fire()
at System.Threading.TimerQueue.FireNextTimers()

We have no control over System.Transactions behavior, so we can't handle exceptions dotConnect is throwing on a worker thread taken from the thread pool. How do we prevent the application from crashing in such a condition?

[wscalf]

Not sure if this is relevant, but are you disposing your TransactionScope (ie: 'usinging' it?) That should, based on my understanding of it, be when it rolls back the transaction, and you would probably want to do that before closing your connection, disposing your DbContext, etc.

[UltimaDeveloper]

Thanks wscalf for your suggestion, I think it might be relevant and I've already tried to investigate it. I've found that all TransactionScope instances are disposed of properly (via «using» statements).

[Pinturiccio]

We have reproduced the issue. We will investigate it and notify you about the results as soon as possible.

[UltimaDeveloper]

Thanks a lot Pinturiccio!
I'm looking forward to hearing from you.

[UltimaDeveloper]

Hi Pinturiccio!

Any news on this one?

[Pinturiccio]

We are investigating the issue, but we can't tell any timeframe at the moment.

[UltimaDeveloper]

Hi Pinturiccio!

we can't tell any timeframe at the moment.

Why not?
Did you assign the job of resolving this issue to a developer?
Can you tell me at least rough approximation when to expect the bugfix?
Weeks? Months?

[Pinturiccio]

Our development team is investigating the issue. We haven't found the solution yet and we don't know when the issue will be solved.

[UltimaDeveloper]

Thanks for your reply.
Can you suggest a workaround?
What can we do to save our server process from crashing while your team is investigating the issue?

Also, we just encountered a bit different situation.
Temporary connection problems caused the process crash with a different stack trace:

in Devart.Data.Oracle.DirectUtils.b(Int32 A_0)
in Devart.Data.Oracle.as.c()
in Devart.Data.Oracle.OracleInternalConnection.Rollback()
in Devart.Common.r.c(SinglePhaseEnlistment A_0)
in System.Transactions.DurableEnlistmentAborting.EnterState(InternalEnlistment enlistment)
in System.Transactions.TransactionStateAborted.EnterState(InternalTransaction tx)
in System.Transactions.EnlistableStates.Timeout(InternalTransaction tx)
in System.Transactions.Bucket.TimeoutTransactions()
in System.Transactions.BucketSet.TimeoutTransactions()
in System.Transactions.TransactionTable.ThreadTimer(Object state)
in System.Threading.ExecutionContext.RunInternal(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
in System.Threading.ExecutionContext.Run(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
in System.Threading.TimerQueueTimer.CallCallback()
in System.Threading.TimerQueueTimer.Fire()
in System.Threading.TimerQueue.FireNextTimers()

P.S. This server is running with Devart.Data.Oracle.dll version 6.3.0.185

[UltimaDeveloper]

And here is another one (Devart.Data.Oracle.dll version 6.80.350.0):

Devart.Data.Oracle.OracleException, Message: "ORA-01012: not logged on"

in Devart.Data.Oracle.at.d()
in Devart.Data.Oracle.cp.c()
in Devart.Data.Oracle.as.c()
in Devart.Data.Oracle.OracleInternalConnection.Rollback()
in Devart.Common.r.c(SinglePhaseEnlistment A_0)
in System.Transactions.DurableEnlistmentAborting.EnterState(InternalEnlistment enlistment)
in System.Transactions.TransactionStateAborted.EnterState(InternalTransaction tx)
in System.Transactions.EnlistableStates.Timeout(InternalTransaction tx)
in System.Transactions.Bucket.TimeoutTransactions()
in System.Transactions.BucketSet.TimeoutTransactions()
in System.Transactions.TransactionTable.ThreadTimer(Object state)
in System.Threading.ExecutionContext.RunInternal(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
in System.Threading.ExecutionContext.Run(ExecutionContext executionContext, ContextCallback callback, Object state, Boolean preserveSyncCtx)
in System.Threading.TimerQueueTimer.CallCallback()
in System.Threading.TimerQueueTimer.Fire()
in System.Threading.TimerQueue.FireNextTimers()

Looks like any exception thrown on a worker thread during System.Transactions.BucketSet.TimeoutTransactions... is unhandled and leads to the termination of the host process.

[Pinturiccio]

We've made some corrections concerning this issue in our new build. When the new build is released, please check whether the issue is solved. We will post here when the corresponding build of dotConnect for Oracle is available for download.

[UltimaDeveloper]

Thanks a lot Pinturiccio!
Looking forward to your posting.

[slaxman]

We are running into the same issue with version 7.2.122.0. Is there a fix?

[Pinturiccio]

We can't reproduce the issue with an uncatchable exception from the background thread, that causes the application termination. We have fixed the issue in dotConnect for Oracle 7.7.257. Try using the latest version of dotConnect for Oracle. You can download dotConnect for Oracle Trial from our site http://www.devart.com/dotconnect/oracle/download.html and test your application.