[infinispan-dev] Atomic operations and transactions

[Dan Berindei]

On Tue, Jul 5, 2011 at 1:45 PM, Sanne Grinovero <sanne at infinispan.org> wrote:
> That's an interesting case, but I wasn't thinking about that. So it
> might become useful later on.
> The refcount scenario I'd like to improve first is about garbage
> collection of old unused index segments,
> we're counting references from open searchers and allow to finish to
> run queries on still open segments
> while they are being deleted: at the same time, these delete
> operations are run in batch operations in background threads,
> I couldn't possibly run them in a transaction as it would likely not
> have enough memory to complete it, and anyway they're run async
> to the rest of the application. So timing is not very critical, but
> having a wrong increment/decrement on the counter can cause
> many issues and so this must rely on atomic operations. Current
> implementation acquires a pessimistic lock on the integer,
> having a distributed "AtomicInteger" as discusses on the train would
> be a simple improvement.
>

I think I got it now, your references are from application objects to
cache keys, so you don't have anything else that you need to modify in
the same transaction.
You only have reads, unless the refcount reaches 0, in which case you
also do a remove.

It's clear to me now that the distributed AtomicInteger approach would
be best in your case, so that means I haven't found a real-life
problem that requires in-transaction atomic operations yet ;-)

Dan


> Sanne
>
>>
>> Dan
>>
>> _______________________________________________
>> infinispan-dev mailing list
>> infinispan-dev at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>
>
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev