change of plans
I noticed that the write skew check is not supported by pessimistic
caches, so the test is totally invalid, because we have 10 threads
perform a put operation, but the key is locked only in prepare (and the
write skew check is performed).
I'm gonna change the test to use the IGNORE_RETURN_VALUES in the put
operation. This should not trigger the write skew check and all the
threads should run as expected.
On 07/10/2013 11:36 AM, Mircea Markus wrote:
On 10 Jul 2013, at 13:34, Pedro Ruivo <pedro(a)infinispan.org> wrote:
> The test is definitely wrong because the locks are only acquired during
> the prepare that gives an window of opportunity to the write skew happens.
>
> On other hand, I'm pretty sure that we have a bug in local mode write
> skew + optimistic caches (it is performing the write skew in the put and
> not in the prepare).
>
> So, are you suggesting to remove WriteSkewTest and extend/adapt
> ConditionalOperationsConcurrentWriteSkewTest to local mode Pessimist +
> optimistic write skew test?
>
> if yes, +1
+1 :-)
Cheers,