[infinispan-dev] Issue about propagation of the RollbackCommand in Infinispan 5.2.0

Pedro Ruivo pruivo at gsd.inesc-id.pt
Sat Aug 18 10:55:39 EDT 2012


On 08/18/2012 02:23 PM, Sebastiano Peluso wrote:
> On 8/18/12 3:14 PM, Pedro Ruivo wrote:
>> Hi,
>>
>> see inline.
>>
>> Cheers
>> Pedro
> (...)
>> My 2 cents if the following:
>> mark the transaction "as prepared" when the prepare reaches the
>> ReplicationInterceptor and, when the rollback command is created, send
>> this information with it.
>>
>> Each node, when received the rollback command, check this information:
>> 1) if it is marked "as prepared", waits for the prepare (or do other
>> stuff, e.g. mark the remote transaction as rollback only, ignore the
>> prepare when received and unlock possible locked keys)
>> 2) if it is not marked "as prepared", unlock possible locked keys
> I've already implemented a solution like this ;-) .
> But you can do better: you can simply avoid to send the rollback
> messages that trigger your point 2). In this way you avoid unnecessary
> flooding of the network. Do you agree with me?
it depends. if you use the Pessimist Locking, the locks are acquired in 
every node. So, you must send the rollback even if you didn't send the 
prepare to unlock them.
If you use the Optimistic Locking, I agree 100% with you =)
>
> Thanks for you reply Pedro.
>
> Cheers,
>
>      Sebastiano
>>> Thank you for the reply.
>>>
>>> Cheers,
>>>
>>>         Sebastiano
>>>
>>> [1] https://issues.jboss.org/browse/ISPN-2081
>>>
>>> _______________________________________________
>>> 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
>
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev



More information about the infinispan-dev mailing list