On 6 Jun 2013, at 10:09, Dan Berindei <dan.berin...@gmail.com> wrote:
> Say you have two transactions, tx1 and tx2. They both send a > LockControlCommand(k1) to the primary owner of k1 (let's call it B). > If the lock commands use SYNCHRONOUS_IGNORE_LEAVERS and B dies while > processing the commands, both tx1 and tx2 will think they have succeeded in > locking k1. > > So you're right, everything should be locked before prepare in pessimistic > mode, but LockControlCommands are also susceptible to SuspectExceptions. On > the other hand, you can use SYNCHRONOUS mode for LockControlCommands and you > can just retry the transaction in case of a SuspectException. > > Unfortunately, you can't retry the transaction if the PrepareCommand fails > (in pessimistic mode; or the CommitCommand in optimistic mode), because it is > executed in the commit phase. The transaction manager swallows all the > exceptions in the commit phase, making it impossible to see if it failed > because of a node leaving. I guess this means I should increase the priority > of https://issues.jboss.org/browse/ISPN-2402 ... +1 for increasing the priority. A bit too late for 5.3.0 stream though, as this is a rather large change. 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
