Scenario:

1. Set up 1 balancer (undertow or httpd) and at least 3 workers(EAP)
2. Deploy some app to the workers, for example clusterbench
3. Send a request and keep session
4. stop-context on worker that handled the request

   /subsystem=modcluster/:stop-context(context=/clusterbench, virtualhost=default-host)
   

5. Send few more requests with before established session
   At this point you`ll see that request jump between 2 workers that have left

Issue:
stop-context message for balancer just informs the balancer that he shouldn't send any request with specified context to that worker
and every new request that don't need to be send to that worker is send to another one. Requests with session set to that worker will get failovered to another worker. Issue is that worker stays session owner(ending of the JSESSIONID stays the same). Failover is then performed with every request and therefore can and will jump from worker to worker.