The following behaviour was observed:
1) Httpd is aware of tomcat1 and tomcat2 nodes.
2) Requests to the load-demo webapp are being equally served both by tomcat1 and tomcat2.
3) Load-demo webapp is stopped with .../manager/html/stop?path=/load%2Ddemo... on tomcat2.
4) Only tomcat1 is serving requests now.
5) Load-demo webapp is started with .../manager/html/start?path=/load%2Ddemo... on tomcat2.
6) Only tomcat1 keeps serving requests, mod_cluster_manager shows load-demo webapp as STOPPED (however it is started at the moment)
7) If you disable & enable all contexts in the mod_cluster_manager, it "helps" the balancer to recognize that the webapp is started again and behaviour 2) takes place.
8) If you repeat step 3) again at this point, the balancer DOES NOT behave as described in 4), however you get HTTP 200 from tomcat1 and HTTP 404 from tomcat2 (balancer is switching between them)