Leader election

Consider a scenario where multiple application instances need to elect a leader among themselves and be aware of the current leader at any given time.

This scenario can be implemented using semaphores in YDB coordination nodes as follows:

1. A semaphore is created (for example, named my-service-leader) with Limit=1.
2. All application instances call AcquireSemaphore with Count=1, specifying their endpoint in the Data field.
3. Only one application instance's call will complete quickly, while others will be queued. The application instance whose call completes successfully becomes the current leader.
4. All application instances call DescribeSemaphore with WatchOwners=true and IncludeOwners=true. The result's Owners field will contain at most one element, from which the current leader's endpoint can be determined via its Data field.
5. When the leader changes, OnChanged is called. In this case, application instances make a similar DescribeSemaphore call to learn about the new leader.