Rewrite Zookeeper Elections

The current implementation of elections watches all election children (secondaries). If a non-primary process terminates, every other node will be notified and perform reads approximately at the same time and could overload zookeeper.

Why delay improving?

  • An implementation currently exists and works.
  • Should not optimise while essential features do not exist.
  • This will cause problems when scaling only, until then it is safe to wait.

Potential improvement

The official docs suggest a better implementation that I should aim for: