[DBCP-569] Fix test random failure on TestSynchronizationOrder.testInterposedSynchronization

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 2.9.0
• Flags: Patch
• Assignee: Unassigned
• Reporter: Florent Guillaume
• Created: 04/Jan/21 00:44
• Updated: 13/Jan/21 17:45
• Resolved: 13/Jan/21 17:29

Description

TestSynchronizationOrder.testInterposedSynchronization may fail randomly depending on GC behavior, because the test involves a Transaction object held only be weak references, which may be cleared earlier that the test expects:

java.sql.SQLException: Unable to enlist connection because the transaction has been garbage collected
	at org.apache.commons.dbcp2.managed.TransactionContext.getTransaction(TransactionContext.java:186)
	at org.apache.commons.dbcp2.managed.TransactionContext.setSharedConnection(TransactionContext.java:105)
	at org.apache.commons.dbcp2.managed.ManagedConnection.updateTransactionStatus(ManagedConnection.java:310)
	at org.apache.commons.dbcp2.managed.ManagedConnection.<init>(ManagedConnection.java:89)
	at org.apache.commons.dbcp2.managed.ManagedDataSource.getConnection(ManagedDataSource.java:64)
	at org.apache.commons.dbcp2.managed.TestSynchronizationOrder.testInterposedSynchronization(TestSynchronizationOrder.java:118)

The TransactionContext.transactionRef and the TransactionRegistry.caches are holding onto the Transaction (acquired in TransactionRegistry.getActiveTransactionContext()) only using weak refs.

However in TestSynchronizationOrder the fake TransactionManager returns a Transaction (that's an instance of an anonymous class) but nobody holds a strong reference to that. The fake TransactionManager should create the fake Transaction at begin() time, and hold onto it using a strong reference until commit() time.

Issue Links

• links to: Pull request #84