Copy data across nodes for durability and read-scaling.

Data Flow (master-slave)

   Master
     |
     v  (async log shipping)
  +--------+--------+--------+
  |slave_0|slave_1|slave_2|          (read replicas)
  +--------+--------+--------+

Two flavours

• Master-slave: writes hit master; slaves serve reads. Replication lag = delay before slaves see the write.
• Multi-master: writes accepted at any node; must resolve conflicts (CRDTs, last-writer-wins, vector clocks).

Pitfalls

• Stale read: a write is followed immediately by a read on a slave that hasn’t caught up.
• Split brain in multi-master during a network partition -> two nodes both think they’re authoritative.

Mitigation

• Read-your-writes consistency: route the session’s own writes to the master for some time window.
• Quorum reads: R + W > N (with N replicas, R reads, W writes).

Analogy

A library where the original book is at HQ and branches hold photocopies: branches lag behind HQ by a few hours.

Interview tip: Mention replication lag explicitly; it’s the cause of half the “data loss” production incidents.