In a great moment of DUH, I realized I had the expireSessionsOnShutdown to true.
<Manager className="org.apache.catalina.ha.session.DeltaManager" expireSessionsOnShutdown="false" notifyListenersOnReplication="true"/> All working nicely now. On Tue, Aug 27, 2013 at 12:27 PM, Tomcat Random <tomcat.ran...@gmail.com>wrote: > Tomcat 7.0.42 / RHEL 6 / Two physical servers, with one tomcat instance on > each server. Physical loadbalancer with sticky sessions. No proxy servers. > > I've set up session-replication using the delta-manager. I can confirm it > works just lovely when the LB switches over from one box to the other. > Using a test get/set session value servlet, the manager app reports the > primary session value on the box where it was set, and the identical backup > value on the other box. So everything appears good there. > > The problem is if, as a crash test, I stop/restart or kill the tomcat > service on the box with the primary session: the backup session on the > other box gets removed. So, as you'd imagine, when the LB swaps to the > non-dead server the session value is gone. Not so good there. > > Shouldn't the backup session value remain? Isn't that sort of the whole > point. Any ideas? It would seem box 1 and box 2 can communicate enough to > create backup session values and detect the death of the other node. Why > would the backup session value be lost? > > Thanks, > Alec >