Re: [jbosscache-dev] GravitateDataCommand finds invalid "phantom" nodes
Manik Surtani wrote:
On 7 Aug 2009, at 10:18, Manik Surtani wrote:
> Thanks for the patch; that makes a lot of sense. Go ahead and check
> it in to trunk.
>
> Regarding the creation of nodes for deletion, well that's pessimistic
> locking for you. Unfortunately the lock is encapsulated in the node
> so you can't acquire the lock until the node is created. (Which is
> partially why the pessimistic lock design sucks).
>
> This is created in PessimisticNodeBasedLockManager (line 145). At
> this stage we already know if we are just creating a dummy node for
> deletion, so perhaps we could pass this as a flag into the constructor
> of a PessimisticUnversionedNode, and this could acquire the lock on
> creation. Hmm, need to think about this.
Right, so I have created a patch for this - it isn't the prettiest in
that it exposes some PessimisticUnversionedNode internals to the
PessimisticNodeBasedLockManager, but these two (deprecated) classes are
so tightly coupled anyway that if this fixes the race then I'm happy to
go with it.
Just as an FYI, this is checked in as rev. 8167.
Great; thanks! Rado is doing some reconfig of his test setup; once
that's done he'll run the tests with a build of your latest.
--
Brian Stansberry
Lead, AS Clustering
JBoss by Red Hat