Atomikos Forum

Hi,

We have been observing this memory leak issue in our application which is running within a tcserver. We are using Hibernate 4.3.6 final and Spring 4.1.1 Final along with Atomikos 3.7 for XA transactions and Oracle as the DB

We have configured OracleXADataSource as the XA datasource and are using the following JTA related properties for configuring hibernate:

    hibernate.current_session_context_class=jta              hibernate.transaction.factory_class=org.hibernate.engine.transaction.internal.jta.CMTTransactionFactory
hibernate.connection.release_mode="AFTER_TRANSACTION

Also we are using com.atomikos.icatch.jta.UserTransactionManager as the transaction manager and com.atomikos.icatch.jta.UserTransactionImp as the transaction implementation class.

We are experiencing a constant increase in heap memory possibly due to a leak in the atomikos classes as revealed by the profiler BELOW

1. Repeated heap dumps reveal that there are many hashtable entries getting created in the SiblingMapper table referred to from the JdbcTransactionaResource class.
While the shallow heap for the Siblingmapper instance is 48 the retained heap (consisting of the hashtable entries) seems to be 854 MB. This is the same result across multiple heap dumps and is being reported as a suspect by the profiler.

2. Also there are some hashMap entries which seem to be referred from the FileLogStream which are persisting accross multiple heap dumps.

3. Apart from the above two (HashTable and HashMap instances), the live memory histograms within the profiler also keeps on showing instances of Atomikos classes SynchToFSM, AtomikosConnectionProxy and AtomikosConnectionProxy$JdbcRequeueSynchronization which are gradually increasing with time even after repeated Garbage Collections. So there seems to be some leaks from within these instances.


We would request you to look into the issue and let us know if there is any known issue or if there is any missing configuration with Atomikos which could be causing these leaks.