Keeping copies of data in multiple regions so reads are fast worldwide and the system survives the loss of any one region.
Geo-replication is the practice of replicating data across multiple geographic regions, either active-active (every region accepts writes, conflicts are resolved) or active-passive (one primary region accepts writes, others are read-only standbys). The pattern serves two goals: low-latency reads for users in distant regions, and disaster recovery if the primary region goes offline. Cross-region replication adds 50-200ms of write latency and data-egress cost, so the design tradeoffs matter.
Single-region services impose fixed latency floors on users in distant regions, an Australian user hitting us-east-1 waits 200ms before the request even arrives. Geo-replication is the foundation for both global UX (reads served from the user's region) and DR (the system survives region loss). It's also the layer where eventual-consistency tradeoffs become real engineering work.
See the part of the platform that handles geo-replication in production.