Geo Replication
Copy data across regions while balancing latency, durability, locality, and consistency.
2–5 minute refresher
Leaderaccept writes
replicate ↓
Follower A
Follower B
Follower C
Place replicas by fault domain, then decide when remote durability enters the acknowledgement path
30second
refresher
refresher
Geo replication maintains copies across regions to survive large failures, serve local reads, and meet data-location requirements. Synchronous replication spends wide-area latency; asynchronous replication accepts a nonzero data-loss window.
What problem does it solve?
Zone-local replicas do not survive a regional disaster, while one remote primary creates poor latency and locality for global users.
How it works
- Choose regions based on latency, regulation, and correlated risk.
- Place replica groups across independent zones and regions.
- Decide whether remote acknowledgement is synchronous, asynchronous, or policy-based.
- Expose replication lag and safe recovery points.
- Practice promotion, client rerouting, backlog catch-up, and failback.
Decision guide
Key trade-offs
| Choice | What you gain | What it costs |
|---|---|---|
| Synchronous remote | Near-zero committed data loss | Wide-area write latency |
| Asynchronous remote | Fast local writes | Nonzero RPO |
| Local reads | Low read latency | Potential staleness |
What happens if?
The primary region disappears with seconds of unshipped writes
Promotion loses those acknowledged local writes unless the contract required remote durability. Preserve the recovery point, communicate the RPO, and reconcile clients that observed outcomes absent from the survivor.
Where it appears
- Cross-region database replicas
- Object storage replication
- Global log mirroring
- Disaster-recovery clusters
Senior interview modeHow do synchronous and asynchronous geo replication change RPO?Show answer
Synchronous remote acknowledgement can make committed RPO approach zero for covered failures; asynchronous replication leaves an RPO roughly bounded by unshipped or unapplied lag, subject to recovery correctness.