It is because of how my implementation is, instead of locking on a RPC call I call out for replicated data in a async manner. You have a very valid point, and it is a valid problem.
I will fix when time permits :) - the fix is to implement RPC instead of async request-reply between nodes. This has been tricky because of how flaky the NIO implementations are in the current available VMs. I might even switch back to java.io instead of java.nio, because so far, I have not seen a good NIO implementation out there.
And yes, if you want to look at source, take a look at T5, since the backport for T4 is not pretty and will get you pretty confused :)
OK, I just want to clarify then, both TC 4 and 5 suffer from the issue that they will start serving requests before the replication manager is fully configured.
-Dave
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
