MongoDB uses Raft-inspired consensus for both leader election and data replication, guaranteeing durability and strong consistency by default with `w:majority` write concern. However, you can trade consistency for performance using `w:1`, which enables asynchronous replication but risks split-brain scenarios during network partitions. When a partition heals, MongoDB deterministically chooses the majority branch and rolls back operations from the losing side, preserving them as BSON files for manual conflict resolution. This rollback mechanism (Recover To a Timestamp) distinguishes MongoDB from vanilla Raft, allowing it to support lower latency, multi-region deployments, and migration workloads while maintaining auditability.
