CSE Roadmap

Core tier · about 3 hours · 8 parts

Part 2 of 8 · 1 min read

Why keep copies

Three goals of replication and the idea of a replica.

Think of an important document kept in the office safe. One copy in the safe is lost in a fire. Copies in two offices survive. A copy kept in a branch office also saves staff there a long journey. The same document, kept in several places, serves three goals at once.

The book "Designing Data-Intensive Applications" gives the reasons for replication: keep data close to users to cut latency, keep the system working when some parts fail, and spread reads over more machines to raise throughput.

Replica and replication

A replica is a node that stores a copy of the database. Replication is keeping those copies in step with each other as the data changes.

Example: a user database runs on 3 machines. Each holds the full set of rows, so each is a replica.

If the data never changed, replication would be easy: copy it once and finish. The difficulty lies in handling changes. The book names three families of method: single-leader, multi-leader and leaderless. This page covers the first and most common. Modules 3 onwards cover the other two.

The three goals pull on the design in different ways:

Goal What replication gives Cost
Availability Another copy takes over if one machine fails Failover is hard to get right (section 4)
Read capacity Reads are spread over many machines Copies can be behind (module 2)
Low latency A copy near each group of users Distance makes keeping copies in step slower

Replication does not by itself solve a data set too large for one machine, since every replica holds all the data. That job belongs to partitioning (module 4).

Check yourself:

  • Name the three goals of replication.
  • Does replication help when a database is too big for one disk? Why?