I'm quiet new to HA solutions that are out of JEE scope, so sorry if my questions are stupid.
Currently I just can't get a whole picture of really no single point of failure solution for JBoss AS7 clustering on Windows server 2008! Maybe someone has an idea on this:
If I set up cluster of 4 JBoss AS7 nodes, and 2 proxy/balancer nodes. I have situation where proxy-balancing nodes takes care of crashed JBoss worker nodes.
However If setup a 2 different IP adresses for balancers (XXX.XXX.XXX.XXX for first node and YYY.YYY.YYY.YYY for second node),
and lets say client knows that for service access they should access IP adress XXX.XXX.XXX.XXX and balancer node 1 goes down (not the server or operating system, but tomcat or httpd or jboss whitchever does this clustering). So clients should switch to YYYY.YYYY.YYYY.YYY address.
DNS round-robin would result 50% of errors. There should be intelligent solution. For Window NLB it works ok with virtual IP, but if understand correctly - it works on operating system level. So if OS is alive, but balancer service is dead - this server is seen as alive in NLB cluster, but is dead from the clients perspective and cluster services are not reachable. So NLB is not fully no-SPOF solution either (unless heartbeat of cluster nodes could be understood by Windows NLB).
Is there some basic stuff I am missing here? (like windows-compatible protocol or some third party software that could manage IP adresses of balancer nodes by monitoring some services or by the help SNMP)?
The question is how to setup whole system that from clients perspective it would be fully automatic-failover? (no need for a transparent failover where sessions are bounced to other nodes transparently, I just need that clients could re-try after failure and continue working and services are reachable)