Designing a Real-time Presence System: Backend Architecture Choices?

Hey everyone, I'm starting to design a real-time presence system (think Slack's online/offline indicator for thousands of users) and I'm looking for some input on backend architecture. My main concern is efficiently handling persistent connections and quickly propagating presence updates with minimal latency. Are people using WebSockets exclusively, or is a combination with something like long-polling still relevant for certain clients? What are your thoughts on using a Pub/Sub system like Redis Pub/Sub or Kafka for distributing updates internally among services vs. direct service-to-service communication? Also, how do you manage user sessions and state in a distributed setup for something like this?