[jboss-user] [JBossCache] - Re: newbie question - ReplicationException

[manik.surtani@jboss.com]

Both locking schemes will have one node fail in the described scenario.  The differences are, using the original poster's example, with PL the commit on JVM2 will fail while with OL the commit on JVM1 will fail.

Other differences are that with OL, the fail on JVM1 is local, i.e., it will fail before any remote calls are made and hence more efficient.  also with OL if the tx on JVM1 hangs, other txs on other JVMs (or even the same JVM) can proceed.

If you want proper atomicity, then you get into the realm of distributed locks (or fail-fast cooperative locks) but either way, we're talking about extremely non-scalable solutions, especially since in a cache, you're looking at a mostly-read use-case which should make such a failure rare and a retry acceptable.

If you are looking at using the cache to store a heavily updated value, like a counter, for example, then a cache is almost certainly the wrong tool for the job.  

That said, enough folk have asked for distributed locking and it is on the roadmap (albeit low prio - JBCACHE-1098) .  It will almost certainly not be enabled by default for scalability reasons though.




View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4117966#4117966

Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4117966