Unexpected problems with JBoss connection pooling - oracle

Question

I've spend the last days trying to locate the cause of some new problem during development that raised a few days ago... and I've not found it yet. But I've found a workaround. But let's start with the problem itself.
We are using JBoss EAP 6.1.0.GA (AS 7.2.0.Final-redhat-8) as our application server for a quite large enterprise project. The JPA layer is handled by Hibernate Core {4.2.0.Final-redhat-1} using oracle.jdbc.OracleDriver (Version 11.2) connecting Oracle 11.2.0.3.0.
A few weeks ago everything worked as expected and we had no database related problems. We were using the following datasource:
<datasource jta="true" jndi-name="java:/myDS" pool-name="myDS" enabled="true" use-java-context="true" use-ccm="true">
<connection-url>jdbc:oracle:thin:#192.168.0.93:1521:DEV</connection-url>
<driver>oracle</driver>
<transaction-isolation>TRANSACTION_READ_COMMITTED</transaction-isolation>
<pool>
<min-pool-size>1</min-pool-size>
<max-pool-size>20</max-pool-size>
<prefill>true</prefill>
<use-strict-min>false</use-strict-min>
<flush-strategy>FailingConnectionOnly</flush-strategy>
</pool>
<security>
<user-name>MY_DB</user-name>
<password>pass</password>
</security>
</datasource>
Most of the time we had 5-10 open connections with 1-3 in use (single development environment)... the pool held that level and worked just fine.
But with some unknown changes to our code that pool stopped working... didn't release it connections anymore... even did not re-use them at all! It took a few simple requests to fill the pool up to the maximum of 20 connections and JPA refused any new database queries.
We've spend several days to find the relevant changes to our code... without success!
Today I've discovered a workaround. We changes persistence.xml a little bit:
<persistence-unit name="myPU">
<provider>org.hibernate.ejb.HibernatePersistence</provider>
<jta-data-source>java:/myDS</jta-data-source>
<properties>
<property name="jboss.entity.manager.factory.jndi.name" value="java:/myDSMF" />
<property name="hibernate.dialect" value="org.hibernate.dialect.Oracle10gDialect" />
<property name="hibernate.transaction.manager_lookup_class" value="org.hibernate.transaction.JBossTransactionManagerLookup" />
<property name="hibernate.default_batch_fetch_size" value="1000" />
<property name="hibernate.jdbc.batch_size" value="0" />
<property name="hibernate.connection.release_mode" value="after_statement" />
<!-- <property name="hibernate.connection.release_mode" value="after_transaction" /> -->
<property name="hibernate.connection.SetBigStringTryClob" value="true" />
</properties>
</persistence-unit>
Changing hibernate.connection.release_mode from after_transaction to after_statement did the trick. But that setting has never been touched before. Now connections are released as expected and the pooling is usable again.
I don't get why after_transaction doesn't work anymore... because changes are committed. We see all these things in the database. And committing a transaction should end it - doesn't it?
Although I've found that simple workaround I'd really get to know the problem. I've no good feeling to delay that knowledge until production time. So any feedback is very well appreciated! Thanks!

Answer 1

You are using JTA . So after_transaction mode in never recommended for JTA transactions.
Here is the document from JBOSS site.
after_transaction - says to use
ConnectionReleaseMode.AFTER_TRANSACTION. This setting should not be
used in JTA environments. Also note that with
ConnectionReleaseMode.AFTER_TRANSACTION, if a session is considered to
be in auto-commit mode connections will be released as if the release
mode were AFTER_STATEMENT.
so you should either use auto or after_statement explicitly, to aggressively release the connection.
References
Connection Release Modes.