Question 1

What problem does replication solve?

Accepted Answer

One copy of the data is one machine's worth of reads, one data centre's worth of latency for distant users, and one failure away from loss. Copies fix all three, and introduce the question of which copy is right when they differ.

Question 2

How does replication work?

Accepted Answer

The common arrangement is leader-follower: one primary accepts writes and streams its change log to replicas, which apply it in order. Synchronous replication waits for a replica to confirm before the write is acknowledged. A committed write survives the primary's loss, at the cost of a network round trip on every commit and stalls when the replica is slow. Asynchronous replication acknowledges first and ships the change afterwards. Writes are fast, but replicas lag (usually milliseconds, sometimes seconds or more under load), and a failover can lose the last few committed writes; see durability. Lag produces anomalies that users notice: Read-your-writes: a user creates something and the next read, served by a replica, does not show it. Fixes: read from the primary for a short time after a user's write, read from a replica only if it has caught up past the write's log position, or fall back to the primary on a miss. Monotonic reads: two reads hit different replicas and data appears to go backwards. Fix: pin a session to one replica. Multi-leader and leaderless replication accept writes in several places, which adds write availability and locality, and makes concurrent writes to the same data a conflict-resolution problem.

Question 3

What does replication assume?

Accepted Answer

Reads dominate, or are latency-sensitive in places far from the primary. The application can tolerate, or explicitly works around, replication lag.

Question 4

How does replication fail?

Accepted Answer

Stale read after write: The user's own change is missing on the next page. Data loss on failover: Asynchronously replicated writes committed just before the crash are missing on the new primary. Lag spiral: A replica falls behind under load, serves increasingly stale data, and may need to be removed from rotation. Caching a miss: A 404 served from a lagging replica is cached and outlives the lag by hours.

Question 5

What are the alternatives to replication?

Accepted Answer

Caching: Reads repeat the same keys; a cache gives read capacity without a full copy of the database. Partitioning: Writes or data size exceed one machine; replication alone does not scale writes.

Replication

Learn it

Quick reference

How it goes wrong

Instead, consider

In practice

It assumes

Explain it in your own words

Where you practise it

Further reading