CloudEXPO 2019

Dynamic Auto Scalable Software Applications

Dynamic auto scalable software applications having multiple custom schedulers running as part of a Multi-Cluster often have the need to execute code in exactly one Application Instance. This pattern can be used to implement processing which must only be performed once within the entire application. This is achieved via a leader election and even job distribution. Microservices utilize this software pattern to ensure all future schedulers that will wake up in the future will stream the events for marketing campaigns. It is so important that we should not reach a customer with several duplicate messages through multiple channels like email, Direct Mail etc. When a custom configured, persisted scheduler wakes only one Application Instance will start processing the customer information to determine the eligibility to reach a customer and stream those events through the specific channels. Ensuring exactly one Node/Scheduler election to run a custom scheduled job ensures events are not duplicated resulting in duplication of work and wasted computer resources from technology and duplicate campaigns from business perspective.

Also, a Scheduler is non-preemptive. Thus, if an Application Instance is running the Scheduler and another Application Instance is eligible to run the Scheduler, the eligible Application Instance must not preempt the Scheduler already running; running Scheduler is allowed to complete.
Apart from that having multiple custom configured scheduled jobs, this algorithm should be responsible to evenly and randomly choosing a node for load distribution of the job election leaders.
The leader election algorithm is aware of the Application Instance version. If a Multi- Cluster contains Application Instances with different versions the Scheduler must only run in the newer version Application Instances.