When we had outages, we had timeouts. Sometimes, the nodes would drop out of a cluster. Therefore, the applications would get timeouts. There is a newer version out that we're planning to upgrade to for some stability. It could also be caused by our design. The product must improve its reliability. It'd be nice if nodes could drop out and notify each other in a timely manner so that the applications wouldn't know if there was a problem.
