Real time for forgetting worker node is about half of the time set in WaitBeforeRemove property I am measuring time from stopping worker node (even remembering time it took) till the time it disappears from mod_cluster manager console. I stop server using normal clean shutdown and it takes for default WaitBeforeRemove value arround 5s + 2 seconds for stopping worker node. In case I set WaitBeforeRemove 20, it takes 10 + 2 seconds and in case I set WaitBeforeRemove to 30 it takes 15 + 2 seconds This behaviour I have tested only with mod_cluster 1.2.1.Final and tomcat 6 (both as part of EWS 2.0) on RHEL 6
[Wed Sep 19 11:54:20 2012] [debug] mod_manager.c(2598): manager_handler REMOVE-APP (/NODE_COMMAND) processing: "JVMRoute=498bb1f0-00d9-3436-a341-7f012bc2e7ec and [Wed Sep 19 11:54:40 2012] [debug] mod_proxy_cluster.c(2252): remove_removed_node: 498bb1f0-00d9-3436-a341-7f012bc2e7ec cluster-prod-01 with WaitBeforeRemove 20 that looks OK. I have tesed with 120 (2 minutes) it is working too.
I have found error in time measuring in my test, I tried it manually and fixed the error and now it behaves correctly => closing as this is not a bug, just mistake in test.