RMS operates on an object-oriented basis. The objects can be almost any system component, such as virtual disks, file system mount points, processes, and so on. These objects are defined as resources. A resource is categorized into a grouping called a object type. An object type has specific properties, or attributes, which limit and define what monitoring or action can occur in relation to that resource. Resources are monitored by programs called detectors. This very general object-oriented design permits a high degree of flexibility in the type and level of monitoring.
Resources which are dependent on each other can be grouped together to form logical applications. Failure of any resource in such a group generally triggers a reaction by the entire application.
Any failure of a resource triggers a user-defined reaction. In most cases, the most significant reaction to failures is switchover.
A switchover, sometimes known as failover, consists of first bringing an application into a well-defined Offline state and then restarting the application under the control of RMS on another node. RMS supports symmetrical switchover, which means that every RMS node is able to take on resources from any other RMS node. For example, if the node that is running an application fails, RMS automatically shuts down the node where the failure occurred and redistributes its application load to another operational RMS node.
The details of performing automatic switchover are defined in user-specified scripts and configuration files that RMS accesses when a failure is recognized. The RMS wizards are generally used to create these files.
Resources can be switched manually for such purposes as hardware maintenance. For example, in a two-node RMS configuration, all applications can be temporarily moved to one node while maintenance is performed on the other. Then all applications can be switched back to the operational node while maintenance is performed on the second node. The only impact to users might be a momentary interruption in service while the applications are being switched, and perhaps a slowdown in response time while all applications are operating on a single node.
RMS uses IP aliasing to allow switchover of the communication link. It is possible for several IP addresses (aliases) to be allocated to one physical network interface. With IP aliasing, the user is able to continue communicating with the same IP address, even though the application is now running on another node.
RMS ensures that all resources of an application are offline on the current node before initiating a switchover. These resources are then switched online on another operational node. This behavior ensures that multiple nodes do not access the same resource, thus protecting against data loss.
In the rare event of simultaneous multiple faults, RMS protects against data corruption by preventing a switchover. This design means that in certain circumstances, switchover may be prevented entirely.
Although high availability is the goal of RMS, data integrity takes priority over high availability.
