Leader Election Pattern

For instances that need a

Coordinator or an Aggregator

> Horizontal scaling with access to shared resource

> Parallel computation with aggregated results

When there is no natural leader

Several strategies

> Arbitrary selection ( lowest-ranked instance or process ID)

> Racing for shared Mutex

> Implementing election algorithms:

    Bully Algorithm

    Ring Algorithm

Issues and Considerations

> Election process should be resilient

> Leader failure must be monitored

> Horizontal auto-scaling might kill the leader when scaling down

> Mutex provider introduces single point of failure

> Single election process = simplicity , brings vulnerability

> Manual algo = greatest tuning and optimizing flexibility