I have an EJB, whose method (among other things) persists JPA entity. If the method throws an error, the transaction is rolled back and the entity is not persisted.
However, I do want that entity to be persisted regardless of any exceptions, that might occur in the EJB method.
I'm using WebSphere 7.0, EJB3.0, JPA 1.0 (OpenJPA that is in WAS), DB2, if it matters.
I tried setting #TransactionAttribute(TransactionAttributeType.NOT_SUPPORTED) on top of EJB; with that, entity is not persisted even if there is no exception. I also tried commiting transaction myself (em.getTransaction().commit()), but getTransaction() throws exception (because transactions are managed by container).
Use bean-managed transactions.
#Stateless
#TransactionManagement(TransactionManagementType.BEAN)
public class MyEJB {
#PersistenceContext(unitName="...")
private EntityManager _em;
#Resource
private UserTransaction _utx;
public void myEJBMethod() {
_utx.begin();
// Use _em
_utx.commit();
// Do other work that might throw an exception.
}
}
Alternatively, use TransactionAttributeType.REQUIRES_NEW as suggested by edalorzo.
I am not an expert on EJBs, but I have been dealing with JPA and transactions for a few days now.
I recently answered another question about how entities resided in a context, and how this works in Java EE applications, the context is linked with your JTA transaction.
You can see details of this answer by clicking here. I think it is useful to understand how to context works in order to comprehend the nature of problems like the one you describe.
If you do not provide transaction support, then there is nothing to persist from the container standpoint, and therefore, your changes to the context are transient.
Also you have to consider that once an exception occurs, your context becomes invalid, and the entities in it get detached. (There are a few exceptions to this, like NoResultException).
Thus, from that point on, if you want to commit something, you need a new JTA transaction, with a new fresh JPA context in order to be able to commit changes to the database.
As I said, I am not an expert in EJBs, but if your method fails due to exceptions and you still would like to retry the transaction again by re-invoking the method, then you could force a new transaction to be created every time the method is invoked and by this, you would create a new fresh JPA context.
On the other hand, if you want your modifications to the entities to be persisted, regardless of exceptions in the method, then you might like to consider moving the code that is updating the entities to a new EJB method defined to start a new transaction (TransactionAttributeType.REQUIRES_NEW) every time you invoke it.
By the time this second inner method finishes, your work over the transactions will be automatically flushed to the database, regardless of the outer method of you EJB failing.
Basically, you would be providing a new context for your entities, and linking such context to a new transaction, scoped to commit when the inner method completes.
The natural behavior in EJB containers, as far as I understand, is that ever method joins the already existing transaction, and this is what you might like to prevent, from my point of view.
Another alternative: if you want to control your context using a different transaction support then you might like to consider providing a resource-local based persistence unit and you can manually instantiate your entity manager and control transaction scope as you wish. But honestly, this does not sound like a good idea to me, at least no in the context of the problem that you described.
Related
I have a JPA entity that links to others -- something like this:
#Entity
class LinkRec implements Serializable {
...
#OneToOne
private OtherEntity otherTable;
...
}
So my logic eventually can delete this entity (calling the EntityManger.remove method), then I want to write to a log file what was done, including reference members of the otherTable object. Is this a permitted operation in JPA?
Is this a permitted operation in JPA?
Yes.
What JPA (underlying JPA provider) does when you invoke remove is just "mark" that the instance is expected to be deleted/removed. But even if the transaction is committed (and the instance deleted from the database) or not, the instance object remains the same. Any changes on its attributes depend on what you do.
Due to you mark the entity as removed you won't can refresh the instance's state from the database (call EntityManager.refersh method). You will get an IllegalArgumentException.
Be aware that, in other cases, you could screw up if you refresh the entity before loggin what you want.
I quote a text from the JPA specification (see Synchronization to the Database section) that could help you to understand the "JPA" behaivor
Synchronization to the database does not involve a refresh of any managed entities unless the refresh operation is explicitly invoked on those entities or cascaded to them as a result of the specification of the cascade=REFRESH or cascade=ALL annotation element value
The relevant line in the spec is:
After an entity has been removed, its state (except for generated state) will be that of the entity at the point at which the remove operation was called.
Since this is all I can find on the subject in the spec, I would say that it could vary from implementation to implementation. In my opinion, this makes what you are tying to do dangerous. It may work in one JPA implementation and not another, or work in one version and not in an upgrade.
If I had to guess on implementations, I would say that #OneToOne objects will probably work okay. Where I would worry is with things like #OneToMany. In the case of Hibernate for example: this collection may be hydrated and in memory, but it may also point to a proxy. If it is a proxy and you call the getter it will check with the database for the collection and fail to load it because the object is gone.
I'm experiencing a very strange behaviour with Transactions using play-2.2.2 with JPA and Eclipse-Link.
My Controller-Action is annotated with #Transactional like this:
#Transactional
public static Result submitOrder() {
// class does call private Methods which persist some entities (methods not annotated)
//...
Action is calling private methods to persist data (This should happen in the same transaction since no other Transaction is started).
During the Methods calls (at random locations) data gets written to db (inserts and updates). Debuging shows that the same (active) transaction is used before and after the write. EntityTransactionImpl:commit is never executed and transaction stays active until request is finished ( watched play.db.jpa.JPA.em().getTransaction() )
How is it possible that the data is written although transaction is still active?
It breakes the setRollbackOnly Mechanism since already written data isn't rolled back.
May there be any kind of Timeout that issue these writes.
Can you suggest any debug-entry-point to narrow down the problem (where can i start debuging the actual write-operations, if not in EntityTransactionImpl:commit)?
Dependencies in build.sbt
persistence.xml
The above described behaviour seemed very odd at first, but then i read about FlushMode and now it makes sense.
The FlushMode of eclipselink as well as hibernate is set to FlushModeType.AUTO
FlushModeType.AUTO automatically flushes Entities to the DB when it thinks it's neccessary. This can be because of an readable operation (Query) on a Persited (but not flushed) Entity but it also happened somehow randomly during my observations.
This breaks the rollback-on-failure mechanism, which I thought must be the standard behaviour of #Transactional.
To achive a propper rollback (on failure or if setRollbackOnly() is set), of all persisted but not flushed entities on transcaction commit, you have to explicitly set the FlushMode at the beginning of your Action.
JPA.em().setFlushMode(FlushModeType.COMMIT);
If you're using Eclipselink, you can also set the following property to make it default behaviour:
<property name="eclipselink.persistence-context.flush-mode" value="commit" />
Links which helped me understand:
Eclipselink Context Flushmode
what to use flush mode auto or commit
performance tuning hibernate
In my case of application managed transaction, I've to choose between:
Using one single EntityManager and calling clear() before each new transaction. Share the EntityManager using a ThreadLocal.
Creating a new EntityManager for each transaction.
I don't have much experience on JPA. My question is which one is better in terms of performance?
I would recommend creating a new EntityManager per transaction. This is the way JPA was designed. The EntityManager should not be an expensive object to create. (the EntityManagerFactory is very expensive though, so ensure you only have one of those).
The link provided by okwap is very helpfull. To make sure it will not slip through, and to follow the board rules, I put a copy here:
- an EntityManager contains a persistence context, that will track
everything read through it, so to avoid bloated memory, you should
acquire a new one, or clear it at some point
- if you read the same object through two different EntityManager you
will get different objects back, so will loose object identity, which
is something to consider
Based on that, I will add, that reading through two different EntityManager may even give objects with different content, if a database transaction was performed by someone else in the meantime. But if reading repeatedly through the same entitymanager, the 2nd read wil just get the objet from the entitymanager cache, so the newer state wil just not be visible.
In my application I need most objects fetched in detached mode (fetched with the find API).
I'm wondering if there is a way to ask a detached object from the JPA provider and save the extra call to detach() API.
In additional I would expect the object created in such mode to be less expensive since the JPA provider doesn't need to add it to the entity manager context.
Is there a way to achieve this with JPA APIs?
Is there a way to achieve such functionality with query results?
Specifically I'm using Eclipse Link so if there is a specific way to do it with this implementation it will be helpful as well.
You can fetch a detached entity without an extra call to detach() if you fetch it outside a transaction. If you are not using container-managed transactions, it's trivial, simply do not start a transaction.
If you are using CMT, you have to make sure the requesting object is not a transaction-enabled EJB:
if in an EJB, suspend the transaction by annotating the appropriate method with:#TransactionAttribute(TransactionAttributeType.NOT_SUPPORTED),
or
call the EntityManager from a POJO. You dont have to call it directly, it only impotrant that the query result will end in a non-EJB object.
AFAIK, there is no performance gain to be expected, since the query result will always be put in the current persistence context, however shortlived it may be.
EDIT: There is another possibility to get detached objects which does not depend on transaction demarcations: JPA constructor expressions:
List<DTO> dtos = em.createQuery("SELECT NEW com.example.DTO( o.title, o.version) FROM Entity o").getResultList();
The constructed type must have a constructor with all the relevant attributes. The objects in the list, entities or not, will always be created detached. However there is a small overhead of instantiating a new object.
I am using EclipseLink 2.3.0. I have a method that I am calling from a unit test (hence outside of a container, no JTA) that looks like this:
EntityManager em = /* get an entity manager */;
em.getTransaction().begin();
// make some changes
em.getTransaction().commit();
The changes were NOT being persisted to the database, and looked at this for a long time and finally realized that EntityManager.getTransaction() is actually returning a NEW EntityTransaction, rather than the same one in both calls. The effect is that the first call creates a new transaction and begins it, and the second call creates ANOTHER transaction and commits it. Because the first transaction was never committed, the changes are not saved. We verified this like this:
log.info(em.getTransaction().toString());
log.info(em.getTransaction().toString());
Which resulted in these log messages:
INFO: org.eclipse.persistence.internal.jpa.transaction.EntityTransactionImpl#1e34f445
INFO: org.eclipse.persistence.internal.jpa.transaction.EntityTransactionImpl#706a4d1a
The two different object ID's verifying that there are two different instances. Changing the code to this:
EntityManager em = /* get an entity manager */;
EntityTransaction tx = em.getTransaction();
tx.begin();
// make some changes
tx.commit();
... remedied the problem. Now when I run the code, I see the SQL statements generated to do the database work, and looking in the database, the data has been changed.
I were a bit surprised by this outcome, since I have seen numerous code examples online (for JPA generally and for EclipseLink specifically) that recommend the code we used for managing transactions. I searched far and wide for information specifically about this but have not found anything. So what's going on?
I looked in the JPA spec for something that specifies exactly what getTransaction() does and it was not specific if the transaction is new or the same. Is there a setting in persistence.xml that controls this? Is the behavior specific to each implementation of the JPA spec?
Thanks so much for any information or guidance.
Using getTransaction() does work in JPA and in EclipseLink (this is how our own tests work).
My guess is you are doing something else very odd.
Are you using Spring, or another layer?
Please include the entire code and persistence.xml for your test. Ensure that you are not using JTA in your persistence.xml.
The JPA spec (see paragraph 7.5.4) has explicit examples showing the use of getTransaction() to begin and commit the transaction. So your code should be fine.
Your test shows that you get two different objects, but that doesn't mean the same transaction is not used. Maybe the returned object is just some proxy to a single, real, transaction object.
Or maybe the transaction is committed or rollbacked inside the code hidden under // make some changes.
Have u tried to use persist before commit: ?
Employee employee = new Employee("Samuel", "Joseph", "Wurzelbacher");
em.getTransaction().begin();
em.persist(employee);
em.getTransaction().commit();