Consul Anti-Entropy Synchronization and Local State Divergence

Consul uses a distributed architecture where health checks are executed by local agents to minimize server load. While the gossip protocol (Serf) manages cluster membership, the synchronization of local agent state with the server cluster relies on anti-entropy mechanisms to ensure eventual consistency.

In environments with rapid service flapping—where a service oscillates between passing and critical states—there is a potential window where the local agent's view of the service health diverges from the server's global state. This creates a synchronization gap that can affect service discovery queries routed through the local agent.

Assuming a recent stable version of Consul, what are the specific triggers for anti-entropy synchronization during health state transitions? How does the system resolve conflicts when the local agent and server disagree on a service's health status during high-frequency flapping?