]
Scott Marlow commented on WFLY-7108:
------------------------------------
WFLY-7075 adds an extension for disabling the test for whether the unsynchronized
persistence context is available when a synchronized (or JTA joined) persistence context
is needed.
WFLY-7108 addresses the identified bug mentioned in WFLY-7075. We will fix the identified
bug via a separate jira in case we decide to back port the bug fix but not the extension.
At this point, I think they are both useful but still prefer the separation.
UNSYNCHRONIZED extended persistence context associated but not joined
with JTA transaction and target component is persistence context of SYNCHRONIZED,
IllegalStateException should be thrown
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Key: WFLY-7108
URL:
https://issues.jboss.org/browse/WFLY-7108
Project: WildFly
Issue Type: Feature Request
Components: JPA / Hibernate
Affects Versions: 10.1.0.Final
Environment: *AppServer:* WildFly 10.1.0.Final
*WildFly-jpa:* wildfly-jpa-10.1.0.Final.jar
Reporter: Scott Marlow
Assignee: Scott Marlow
As we mentioned in
https://issues.jboss.org/browse/WFLY-6127 we found bug and made our
solution.
*The obtained defect:*
A defect related to the check of synchronization type (to satisfy JPA spec 2.1 section
7.6.4.1) was found in WildFly 10.1.0.Final.
The Method getSynchronizationType of class ExtendedEntityManager ALWAYS returns type of
synchronization as SYNCHRONIZED (jar file: wildfly-jpa-10.1.0.Final.jar)
*FIX:*
We made a fork of WildFly-jpa project at:
https://github.com/argustelecom/wildfly/tree/WFLY-6127
Our Fix commit:
https://github.com/wildfly/wildfly/commit/3bff5fde3cfc23f3999dc75c320029e...
_Changes_: The method getSynchronizationType returns declared synchronization type.
*Consequences:*
We use own realisation of Martin Fowler pattern "Unit of Work". We initialize
UNSYNCHRONIZED Extended Persistence Context and our UnitOfWork realisation manages the
synchronization with transaction.
Our Beans could be controlled by UnitOfWork, but also could be used as part of WebService
call.
It requires a declaration of synchronize persistence context.
We catch IllegalStateException after we fixed defect of synchronization type
determination, because we initialize UNSYNCHRONIZED persistence context, but we use
declare SYNCHRONIZED persistence context in our beans.
However, our UnitOfWork realisation control synchronization of persistence context and we
can synchronize context before synchronization type check.
*Our actions:*
We add ability to check synchronization type in the method
testForMixedSynchronizationTypes of class TransactionScopedEntityManager by
isJoinToTransaction method (i.e. the actual type of synchronization).
This ability realized by property "jboss.as.jpa.syncasjoin" in persistence.xml
file. Only if this property setted to true - we perform testForMixedSynchronizationTypes
by isJoinToTransaction method. We work as usual if this property not defined or setted to
false.
_Commit_:
https://github.com/wildfly/wildfly/commit/195a8a65a9fae006ad603e425f6a16d...
*Example for reproduction:*
I modified quickstart example: [^kitchensink-ear.rar]
MemberRepository begin Extended UNSYNCHRONIZED Persistence Context.
MemberFinder.find called from MemberRepository. MemberFinder declared
"SYNCHRONIZED", but MemberRepository declared UNSYNCHRONIZED.
MemberRepository also perform join Transaction.
Questions from [~smarlow]:
- whether you could instead use an application-managed entity manager instead (which is
similar to extended persistence context except the app controls it.
We decided to use Container-managed EntityManager, because
- Application-managed entity managers don’t automatically propagate the JTA transaction
context. With application-managed entity managers the persistence context is not
propagated to application components
- The container makes our job
We want to use the existing mechanism
*In Addition:*
Formally, this is out of JPA SPEC, but we have the following reasons:
- In the development process of the JPA specification got a question: "Should we
relax this if the PC has been joined to the transaction?".
But unfortunately, Linda DeMichiel decided to leave as current behavior because no
feedback was given.
(
https://java.net/projects/jpa-spec/lists/jsr338-experts/archive/2011-08/m... )
- We found JIRA task
https://java.net/jira/browse/JPA_SPEC-6 but it was closed because
"No feedback in favor of changing current approach"