How it works
Replication stream
The primary sends an ordered change log that each replica applies, usually asynchronously to keep write latency low.
Route eligible reads
Analytics, feeds, and stale-tolerant lookups fit replicas; workflows that require the newest committed value may need the primary.
Read-your-writes
Session pinning, version tokens, or a short primary-read window can keep a user from seeing older state after an update.
Failover is separate
Promoting a replica requires checking its freshness, preventing the old primary from writing, and redirecting both writers and readers.