On 10 Jun 2013, at 17:30, Dan Berindei <dan.berin...@gmail.com> wrote:
> Yes, putAll is really heavy in non-tx (concurrent) mode, because the same > PutMapCommand is forwarded from each primary owner to all the backup owners > of the keys it primary-owns. However, I don't think > > However, in non-tx mode locks are owned by threads. A separate lock command > would acquire a lock and associate it with its execution thread, making it > impossible for a following write command to use the same lock. Changing > putAll to implement Radim's proposal would indeed make it very similar to a > transactional putAll: you'd need a pseudo-transaction object to associate the > locks with, and a reaper to clean up the pseudo-transaction objects when the > originator leaves the cluster. Indeed would require some significant changes as we'd need to support lock to command association for non-tx caches. We could use some generated UUID for managing the lock owner in this case, or even an GlobalTransaction object. Cheers, -- Mircea Markus Infinispan lead (www.infinispan.org) _______________________________________________ infinispan-dev mailing list infinispan-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev
