Re: [Springframework-developer] Hibernate commit/rollback

[<tri...@tr...>]

Juergen,

The log shows that it is a MySQL database - do we know that the table is one
that support transactions?  If not then no matter what we or hibernate do is
going to help. 

Thomas

> There's a bug entry on SourceForge regarding HibernateTransactionManager's
> committing:
>  
> If the flushing of changes is delayed (i.e. HibernateTemplate is in
> FLUSH_AUTO mode, not in FLUSH_EAGER), the Hibernate commit call will first
> implictly cause a flush. The problem is that that flush might throw an
> exception which causes Hibernate's commit implementation to return with the
> exception, without a commit but unfortunately also without a rollback.
>  
> The current implementation of HibernateTransactionManager will not cause an
> explicit rollback in that case, as it assumes that the commit has failed and
> caused an implicit rollback. It takes that assumption from JTA: After a
> failed commit, you don't have to invoke rollback anymore; the failed commit
> has already completed the transaction with a rollback.
>  
> If I understand correctly, a JDBC driver should behave the same: If a commit
> has failed, rollback does not have to be invoked anymore. Would it actually
> hurt if one still invoked rollback? Would that get ignored with most drivers,
> or would it cause yet another exception, a la "transaction is already rolled
> back" or "no transaction anymore"?
>  
> I intend to fix the Hibernate transaction issue like this: First invoke flush
> manually and treat it like data access, i.e. just commit if it succeeded,
> else cause an explicit rollback. Aren't the semantics of Hibernate's commit
> method a bit odd? Shouldn't it cause a rollback itself if the flush failed?
>  
> I'm not sure about JDO in that respect. It will also flush changes at commit,
> and there is no chance to cause an explicit flush before. The spec and the
> javadocs are unclear about whether a flushing failure causes a rollback of
> the datastore transaction, but I assume so. At least, it seems that rollback
> can and should not be invoked after an exception on commit.
>  
> Has anyone practical experience with these issues?
>  
> Juergen
>  
>  
>  
>  
>