Re: [infinispan-dev] JGroups 2.7 and 2.8 work with Infinispan now
I think we should drop the up-latch. I don't see how messages can get
received before a view (after the STOP-FLUSH message has been applied
(and ack'ed)). I also don't see how receiving a message before unblock
(but after VIEW) would break virtual synchrony...
Do you want to discuss this offline ?
Vladimir Blagojevic wrote:
On 6/2/09 3:23 PM, Bela Ban wrote:
>
> I do *not* think a message from a member can squeeze in *before* a
> VIEW because view installation is synchronous. However, you're right,
> a message can squeeze in after a VIEW and before an unblock() is
> received. But IMO that's fine I guess. Or do you see any issues here ?
>
That message gets dropped by FLUSH :( as things stand right now. That
is the question, what should we do? Should we wait for a latch to open
by unblock(with some timeout i guess) or should we let the messages
pass up the stack freely and forget this whole latch business?
Let me think about implications as well....
--
Bela Ban
Lead JGroups / Clustering Team
JBoss - a division of Red Hat