Skip to main content

2.8 Consistency

image.png

Strong Consistency (changes are reflected to all users immeidately) vs Weak Consistency (changes take some time to reach all users because there're different versions of the data in the system)

We need some terminologies in the middle: Consistency Models

Linearizability Consistency Model

When updates are completed, all clients will get the updated value when they read it.

Drawbacks:

  • It is slow
  • when there is a network partition, linearizability and availability are mutually exclusive

Typical usage: banking, e-commerce, booking systems, distributed locks

Eventual Consistency Model

If there are no additional updates made to the object, eventaully all reads will return the latest written value of that object. It does not guarantee when exactly replication will complete.

Inconsistency window is typically small (sub-second)

Can be much faster than linearizability

No need to sacrifice availability

DNS is the most popular example

Drawbacks:

  • May cause a lot of confusion

Monotonic Reads Consistency Model

Read-Your-Writes Consistency Model

Consistent Prefix Reads Consistency Model

 

image.png

Back to top