Hi,
in our application, we frequently (not always) observe
TransactionRequiredExceptions on startup:
2020-07-03T05:02:40,261 ERROR - Could not create or update entity in
database
javax.persistence.TransactionRequiredException:
Exception Description: No externally managed transaction is currently
active for this thread
at
org.eclipse.persistence.internal.jpa.transaction.JTATransactionWrapper.throwCheckTransactionFailedException(JTATransactionWrapper.java:96)
~[!/:?]
at
org.eclipse.persistence.internal.jpa.transaction.JTATransactionWrapper.checkForTransaction(JTATransactionWrapper.java:56)
~[!/:?]
at
org.eclipse.persistence.internal.jpa.transaction.JTATransactionWrapper.registerIfRequired(JTATransactionWrapper.java:133)
~[!/:?]
at
org.eclipse.persistence.internal.jpa.EntityManagerImpl.joinTransaction(EntityManagerImpl.java:2237)
~[!/:?]
at
org.apache.aries.jpa.support.impl.XAJpaTemplate.txExpr(XAJpaTemplate.java:60)
~[?:?]
at
org.apache.aries.jpa.support.impl.AbstractJpaTemplate.txExpr(AbstractJpaTemplate.java:43)
~[?:?]
We are running our application in Karaf 4.2.8, using the features jpa
2.7.2, eclipselink 2.7.5, transaction 2.0 as basis for the persistence
layer. Basically, the DAO methods are rather simple: the JpaTemplate
service is injected using the respective Declarative Services
annotation, and objects are then persisted with
jpaTemplate.tx(entityManager -> entityManager.persist(objectToPersist));
No TransactionType is defined, i.e. the default TransactionType.Required
will be used.
When starting the system from the "clean" state, i.e. with the Karaf
bundle cache not yet populated, we often - not always - receive the
exceptions mentioned above. When such exceptions occur, and Karaf is
restarted without deletion of the bundle cache, the exceptions will
always reappear. When restarting Karaf with deletion of the bundle
cache, sometimes the issue reappears, sometimes it doesn't.
The issue can always be reproduced by restarting the
pax-transx-tm-geronimo bundle from the Karaf console. The issue can
always be solved by restarting the
org.apache.aries.jpa.eclipselink.adapter bundle from the Karaf console.
When investigating further, it could be seen that in the "error state",
the XAJpaTemplate in org.apache.aries.jpa.support and the
EntityManagerImpl and JTATransactionWrapper in
org.eclipse.persistence.jpa hold different instances of the
GeronimoTransactionManager (the EntityManagerImpl and
JTATransactionWrapper still hold an old instance), while in the "good
state", all these classes hold the same instance of the
GeronimoTransactionManager. Obviously the
org.apache.aries.jpa.eclipselink.adapter bundle doesn't care of changes
of the transaction manager. Only when restarting this bundle, the
services for EclipseLink are recreated and thus also receive the updated
transaction manager. Depending on the startup order of bundles with the
same start level (all involved bundles have start level 80), this
situation often happens on initial startup. When the start order is
fixed such that it results in the "good state", it remains "good" unless
the bundle cache is deleted, because the bundle cache fixes the start
order. When restarting with "clean", the game starts from scratch.
Now the question: does anybody know a solution for this problem? How can
we grant that our application doesn't run into this situation on
startup? Or is this something which can only be corrected in the Aries
JPA Eclipselink Adapter code?
Thanks & Regards
Jochen
