Use case: Having DB connection pool per node
Description: I want to deploy an API backed by a database on a Ray cluster on Kubernetes. For each node I want to have a db client to manage database connection pools. I am thinking of having one or a set of Actors per node for this purpose. However, I want to make sure each node has its own Actor(s) so that tasks within the node use a local db Actor vs a remote one to avoid data serialization/deserialization. How can ensure that there exists at least N (>= 1) Actors per node where there is no node without db actor and also tasks prefer their local db actors for database connections? And as the cluster scales the actors are scaled accordingly?