Replication

NodeDB replicates data via Raft log replication. Each vShard's Raft group independently replicates its WAL entries to followers.

Replication Factor

Configurable per collection or per vShard:

• RF=1 — no replication, single-node durability only
• RF=3 — default for production, tolerates one node failure
• RF=5 — high durability, tolerates two node failures

How Replication Works

1. Leader receives a write and appends to its WAL
2. WAL entry is sent to all followers in the Raft group
3. Each follower appends to its own WAL and acknowledges
4. Leader waits for a quorum (majority) of acknowledgements
5. Leader commits the entry and responds to the client

Follower Reads

Followers can serve reads with bounded staleness:

SET read_consistency = 'bounded_staleness_5s';
SELECT * FROM orders WHERE status = 'pending';

The follower checks that its replication lag is within the configured bound before serving the read. If the lag exceeds the bound, the read is forwarded to the leader.

Cross-Region Replication

Learner replicas (non-voting Raft members) can be placed in remote regions for read scaling. They replicate asynchronously and do not participate in quorum decisions, so they do not add write latency.