Skip to main content

Multi-Cluster and Multi-Region Strategy

FLUX is currently single-broker. This chapter defines the path to true distributed topologies.

Current Topology​

  • one broker process
  • local disk persistence
  • local replication simulation

Target Cluster Topology​

Phase 4+ direction:

  • metadata quorum for controller state
  • multiple brokers with partition leadership
  • broker registration and fencing
  • deterministic failover behavior

Multi-Region (Future)​

Initial stance:

  • single-region strong consistency focus first
  • multi-region active-active is explicitly non-goal for early versions

Possible later patterns:

  • active-passive disaster recovery
  • asynchronous cross-region replication pipelines

Operational Questions to Solve​

  • leader election latency under failures
  • replication lag SLOs
  • partition reassignment workflows
  • client metadata refresh and routing behavior

Client Impact​

To fully mirror distributed cluster behavior across regions, SDKs need:

  • metadata discovery and refresh
  • leader-aware produce/fetch routing
  • retriable broker failover handling
  • backoff policies tuned for control-plane transitions