I have a JPA-persisted object model that contains a many-to-one relationship: an Account
has many Transactions
. A Transaction
has one
Using merge is risky and tricky, so it's a dirty workaround in your case. You need to remember at least that when you pass an entity object to merge, it stops being attached to the transaction and instead a new, now-attached entity is returned. This means that if anyone has the old entity object still in their possession, changes to it are silently ignored and thrown away on commit.
You are not showing the complete code here, so I cannot double-check your transaction pattern. One way to get to a situation like this is if you don't have a transaction active when executing the merge and persist. In that case persistence provider is expected to open a new transaction for every JPA operation you perform and immediately commit and close it before the call returns. If this is the case, the merge would be run in a first transaction and then after the merge method returns, the transaction is completed and closed and the returned entity is now detached. The persist below it would then open a second transaction, and trying to refer to an entity that is detached, giving an exception. Always wrap your code inside a transaction unless you know very well what you are doing.
Using container-managed transaction it would look something like this. Do note: this assumes the method is inside a session bean and called via Local or Remote interface.
@TransactionAttribute(TransactionAttributeType.REQUIRED)
public void storeAccount(Account account) {
...
if (account.getId()!=null) {
account = entityManager.merge(account);
}
Transaction transaction = new Transaction(account,"other stuff");
entityManager.persist(account);
}