Hi Edward, I will answer your question in terms of Solaris Cluster 3.2 update 1/09, running on Solaris 10. The answer will differ somewhat on OpenSolaris (for example, OpenSolaris does not currently have "native zones"); however, Sun Cluster does not yet run on OpenSolaris.
On each Solaris host, there is one rgmd process for the global cluster, and one additional rgmd process for each zone cluster. All rgmd processes run in the global zone. The "global cluster" consists of the global zone and zero or more non-global zones of the 'native' brand. There may be any number of native-brand non-global zones on each Solaris host; but there is only one global-cluster rgmd per Solaris host, controlling all of these non-global zones plus the global zones. The name "global cluster" is confusing because it consists of global zones and non-global zones. We were going to change the name to "base cluster", but there was not enough time in the schedule to implement the change, so the term remains "global cluster". You might think of the global cluster as the "native cluster" since its membership consists of all global and non-global native-brand zones across all of the Solaris hosts in the cluster. Each "zone cluster" consists of at most one non-global zone per Solaris host; all of the zone-cluster zones have the same zonename, and are of the 'cluster' brand. Suppose I have a four-node cluster consisting of physical hosts phys1, phys2, phys3, and phys4. Suppose there is a zone cluster named 'zc1'. Then there will be a cluster-brand zone named 'zc1' on each of the four Solaris hosts. The membership of zone cluster zc1 consists of the four non-global cluster-brand zones (one running on each physical host) named 'zc1'. There is a zc1 rgmd running in the global zone of each Solaris host. Suppose then I halt zone zc1 on host phys4. The zc1 rgmd on phys4 continues to run; we call this a "proxy rgmd". The zone cluster does not have a member node on phys4, but you can still execute zone cluster commands (for example, "clrg list -Z zc1") in the global zone of phys4. The proxy rgmd is there to handle such commands. The proxy rgmd is prevented from being the zone cluster president. There are some bugs in this area which are being fixed in the next patch release coming soon after SC 3.2 1/09. rgmd processes themselves never run in non-global zones. The rgmd processes for the global cluster and the rgmd processes for the zone clusters do communicate with each other to implement inter-cluster resource dependencies and inter-cluster RG affinities. The communication is from rgmd president of one cluster to rgmd president of the other cluster, using orb calls. --Marty Edward wrote: > Hi Marty, > I still have some problems: > I don't know how the rgmd exist in the non-global-zone cluster and > global-zone cluster? > is there one rgmd for global-zone cluster and several rgmd for zone > cluster? > zone cluster's rgmd run in global-zone and called proxy rgmd? > zone cluster haven't rgmd runing in the zone ? > rgmd for zone cluster even called proxy rgmd ,yet they can be elected to > be president in the zone cluster? > rgmd for global zone and rgmd for non global zone need't communication? > global cluster can include rgmd from the zone in the same chassis? > > Thanks > BestRegards > Edward
