Hi, There are basically two ways of getting an IP address within a zone if zone runs under a cluster. One is to put the zone IP address into zone's configuration so it is not managed by cluster. The drawback is that for example if all interfaces fail (in IPMP group) cluster won't detect it and won't failover a resource group for the zone.
The other option is to not configure the ip address into the zone configuration but rather create additional logicalhostname resource. Now the cluster will properly be able to detect network failure in above scenario. However it seems that the way the IP address is configured in in this scenario there is a race condtition when if zone boots too quickly (everything is cached in a memory) some services which depend on network will be started before the IP address has been configured which is causing them to fail, etc. For example it is often a case with ssh if configured to bind to the ip address and not to *. Sometimes with other apps which can't contact dns, etc. IMHO they way an ip address is being configured in in such a scenario should be changed so unless it is done notheing else starts after network-physical (or other proper smf service). ps. I know one could configure an ip address in a zone's configuration and another ip address as a logicalhostname which would workaround both the issues above. However as I have a resource group per zone and multiple configs like that on the same cluster I don't like the idea of having to allocate an extra IP address per each zone. -- Robert Milkowski http://milek.blogspot.com -- This message posted from opensolaris.org
