On 20 Mar 2013, at 18:18, Dan Berindei <[email protected]> wrote:

> The issue I'm seeing is that without deadlock detection the commands that 
> would normally time out are never executed on the thread pool, so after 
> running for a while you can end up with a lot "expired" commands in the 
> queue. The sender will give up after the replication timeout expires, but the 
> recipient will keep on checking if the commands can acquire their locks.

Hmm, this is no good.

> 
> Actually the sender should roll back the transaction when it gets a RPC 
> timeout, so the commands should be able to run at some point. It's just that 
> you'll be ignoring lockAcquisitionTimeout, which is usually smaller than 
> replTimeout. I'm not sure how big an issue this would be... Mircea, Manik?

But then the rollback command gets on the queue as well ... ?


--
Manik Surtani
[email protected]
twitter.com/maniksurtani

Platform Architect, JBoss Data Grid
http://red.ht/data-grid

_______________________________________________
infinispan-dev mailing list
[email protected]
https://lists.jboss.org/mailman/listinfo/infinispan-dev

Reply via email to