By Shreedhar Ganapathy on Jan 17, 2008
As usage of Shoal's clustering framework increases across products, new feature requests are coming in to enrich Shoal's offerings for employing applications.
The Sailfin project is building a Telecommunications Application Server with contributions from Ericsson and Sun. One of the parts of Sailfin is a Java based load balancer called ConvergedLoadBalancer (CLB). The CLB load balances both Http and SIP based requests. The CLB is unique in that it is a component of the Sailfin appserver instance. Thus, any instance in the cluster can be configured as a load balancer. Such an instance can be part of a Front End LB Cluster performing load balancing of requests on a separate application server cluster tier or it can be part of a Self LoadBalancing Cluster wherein one or more instances of the application server cluster also perform the role of a load balancer while also serving requests. The CLB component and other Sailfin components employ Shoal for cluster events, messaging and health monitoring.
The CLB (as with any load balancer) needs to know when instances of a cluster have not only joined a cluster but also when the instances are ready serve requests. Shoal was up until now providing a JoinNotificationSignal which would be triggered as soon as each instance in the cluster used Shoal's GMS to join the cluster. This was sufficient for many use cases but for the LB it needed to know not just that but also when the instance had completed startup operations. This requirement helped us design a new notification called JoinedAndReadyNotificationSignal that would be disseminated to the cluster members for each instance completing startup and reporting such a completion to the group.
Shoal's GroupManagementService provides a new API called reportJoinAndReadyState() which the employing parent application can now call when the parent application has completed its own initialization and startup indicating a point where operational activities on the instance can now start.
Sheetal and I recently committed the code supporting this new feature.
This feature can be particularly useful while building Compute Grid and Cloud Computing type services using Shoal wherein each Grid node can now report when they are ready to act as nodes in the grid. BTW Shoal's concept of a GroupLeader ties in well with the Grid's Compute Task Manager abstraction and is a good infrastructural fit.
Do please share your feedback with us at users alias about what you would like to see added in Shoal to serve your clustering and fault tolerance needs. Shoal's goal of clustering goes much beyond the realm of Data Grids (which we are looking into building ) with a wider spread in terms of building fault tolerance solutions.