Re: [hibernate-dev] [ORM] Synchronization on AbstractLoadPlanBasedLoader
Hi,
We really need to test it thoroughly because the current pooled optimizer
are reasonably fast especially when used with a database sequence.
The table generators are slow because of the row-level locking, so I won't
include those in this discussion.
What strikes me is the synchronized block which might cause contention we
didn't have before.
I'd also vote for a new optimizer instead of modifying the pooled or the
pooled-lo ones.
The current optimizer are quite easy to grasp, and, if we are to add a
high-performance one, I think a new implementation is better suited for
this task.
Vlad
On Mon, Dec 14, 2015 at 6:25 PM, Sanne Grinovero <sanne(a)hibernate.org>
wrote:
Hi all,
while reviewing an improvement by Stale about reducing
synchronization, I'm having the impression that the synchronization
could be completely removed.
But there's a comment warning me against that, so for sake of safety
I'm merging the improvement without risking going too far:
// synchronized to avoid multi-thread access issues; defined as
method synch to avoid
// potential deadlock issues due to nature of code.
I tried to figure what "potential deadlock" it's referring to, but I'm
having the impression the comment might be outdated. So I've reduced
the contention to the only include the code block about which I'm not
confident.
By looking into git history, it seems the comment isn't related to any
specific fix but was included already when this class was first
created.
Would someone be able to point out what is the issue this is protecting
against?
That should allow us to provide an even better patch, although I'll
apply the safe one for now so to at least have the benefits already
when wrapping of result-sets is disabled.
thanks,
Sanne
_______________________________________________
hibernate-dev mailing list
hibernate-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/hibernate-dev