The timer facility is used by SBBs and SBB parts to manage single-shot and periodic timers. The default replication behaviour of Rhino when a timer is created depends on the replication mode of the activity that the timer is created against. Non-replicated timers are created for non-replicated activities, and replicated timers are created for activities that are replicated using the traditional savanna framework.

However, activities whose state is replicated using a key/value store only exist on one node at a time, and thus it does not make sense to replicate a timer created on such an activity across the entire cluster. Thus, Rhino’s default behaviour for timers created on a key/value store replicated activity is to not replicate those timers at all.

There are cases, however, where an application desires that such a timer survives a node failure. The firing of the timer may even be expected to trigger the retrieval of application state from the key/value store and adoption of the SBB entity tree on a surviving node after the node originally owning the SBB entity tree has failed.

In order to support replication of timers on key/value store replicated activities, Rhino extends the standard javax.slee.facilities.TimerOptions class with the com.opencloud.rhino.facilities.RhinoTimerOptions class. The RhinoTimerOptions class defines two additional timer options: replication factor and convergence name session ownership record.

Replication Factor

The replication factor specifies the number of remote timer server nodes that the timer will be replicated to.

The default value of this option is 0, meaning the timer will not be replicated.

The maximum supported replication factor in the current version of Rhino is 1. A larger replication factor may be specified by an application without error, but it will be treated as if the application had specified a replication factor of 1.

Also in the current version of Rhino, the remote timer server is implemented as an internal component of a Rhino node, thus creating a replicated timer by the mechanism described here arms the timer on some other Rhino node (as well as the local node). The remote timer server node chosen for replication is chosen at random - the application has no control over which node is chosen. If there is only one node in the cluster when a timer with a non-zero replication factor is created, the local timer will still be created but a remote timer will not.

A replication factor greater than 0 can only be specified for timers set on replicated activities, but the value is ignored for activities that are not replicated using a key/value store. A replication factor less than 0 is treated as if it was 0.

If a replication factor greater than 0 is specified, then the convergence name session ownership record option must also be set in the RhinoTimerOptions object.

Convergence Name Session Ownership Record

This option contains a reference to the convergence name session ownership record of the SBB entity tree arming the timer. This record is used to identify which Rhino node within the cluster, if any, currently owns the SBB entity tree when the replicated timer fires.

A convergence name session ownership record can be obtained from the SBB context of the SBB, or SBB part context of the SBB part.

This option must be specified if the replication factor option is set to a value greater that 0, otherwise it is ignored.

Previous page Next page