|
But how can we write a standard test case for this issue , the bug happens if we have network issue or packet loss only at a specific time/line but once it happens a complete thread , its session , associated connection and transaction is in a stale state which can't be recovered from.
JdbcTransaction.java
1. protected void doBegin() {
2. try {
3. if ( managedConnection != null )
4. { throw new TransactionException( "Already have an associated managed connection" ); }
5. managedConnection = transactionCoordinator().getJdbcCoordinator().getLogicalConnection().getConnection();
6. wasInitiallyAutoCommit = managedConnection.getAutoCommit();
7. LOG.debugv( "initial autocommit status:
8. {0}
9. ", wasInitiallyAutoCommit );
10. if ( wasInitiallyAutoCommit )
11. { LOG.debug( "disabling autocommit" ); managedConnection.setAutoCommit( false ); }
12. }
13. catch( SQLException e )
14. { throw new TransactionException( "JDBC begin transaction failed: ", e ); // here we need to set managedConnection=null }
15. isDriver = transactionCoordinator().takeOwnership();
16 .}
before this method is triggered from session.beginTransaction() -> SessionImpl -> AbstractTransactionImpl.begin() ->JdbcTransaction.doBegin()
the tranasction is in NOT_ACTIVE state with no managedConnection value
but in case line #5 is successful , we have a connection set to managedConnection and then we have issue at line #11 (managedConnection.setAutoCommit( false )) the we get the exception thrown but the managedConnection is already to some value and is not null So next time we we call beginTransaction , because this is session per thread , getCurrentSession will give the same session and use the same transaction where we will get an error at line #4 as teh condition managedConnection != null is true.
|
