Service Fabric可靠服务：集群副本间是否支持Pub/Sub或广播事件？

Hey Ashish, great question! You’re correct that Azure Service Fabric Reliable Services don’t include a native built-in broadcast or Pub/Sub pattern out of the box, but there are several effective approaches to handle your scenario—where you need to notify all stateful service replicas of custom events when a primary replica’s state changes. Here are your options:

• Custom Pub/Sub using Reliable Collections
  Since you’re already familiar with Reliable State Manager events, this is the most integrated approach. You can build a lightweight Pub/Sub layer using Reliable Queues or Reliable Dictionaries:

  • Have every stateful replica register itself (e.g., store its communication endpoint in a shared Reliable Dictionary) on startup.
  • When a primary replica’s state changes, it can either:
    1. Write the custom event to a shared Reliable Queue, and all replicas listen for new messages via Reliable State Manager events.
    2. Iterate over the registered endpoints in the Reliable Dictionary and send direct notifications using Service Fabric’s remoting capabilities.
  • This leverages the built-in high availability and consistency of Reliable Collections, with no external dependencies needed.

• Leverage Reliable Actors' Event System
  If you’re open to integrating Reliable Actors alongside your stateful services, Actors have a native event publishing/subscription system via the IActorEventPublisher interface. Here’s how it could work:

  • Create an actor type that acts as a "subscriber proxy" for each stateful service replica.
  • Each replica instantiates its proxy actor and subscribes to a shared actor event.
  • When the primary replica’s state changes, it triggers the event via the actor system, which broadcasts it to all subscribed proxy actors—who then relay the event to their associated service replicas.
  • This is a good option if you want to reuse the actor framework’s built-in eventing logic.

• Integrate a Dedicated Messaging Service
  For more robust Pub/Sub needs (like message persistence, filtering, dead-letter handling, or cross-cluster communication), you can use a service like Azure Service Bus Topics and Subscriptions:

  • Set up a topic dedicated to your state change events.
  • Every stateful service replica creates a unique subscription to this topic on startup.
  • When a primary replica’s state changes, it sends the custom event to the topic, and the messaging service automatically delivers it to all subscribed replicas.
  • This adds an external dependency but gives you enterprise-grade messaging features.

• Direct Replica-to-Replica Communication
  Using Service Fabric’s built-in service remoting, you can have the primary replica discover all other replicas in the cluster and send direct notifications:

  • Use FabricClient to query the service’s partition information and retrieve the addresses of all active replicas.
  • Use ServiceRemotingClientFactory to create clients for each replica and send the custom event message directly.
  • This is a straightforward approach but requires you to handle replica discovery, retries, and fault tolerance manually.

Based on your specific scenario, starting with the Custom Pub/Sub using Reliable Collections is probably the best fit—it’s tightly integrated with Reliable Services, requires no extra resources, and aligns with your existing knowledge of the Reliable State Manager.

内容的提问来源于stack exchange，提问作者Ashish