The CAP theorem explains the trade-off between the characteristics of distributed system which are Consistency, Availability and Partition tolerance.
Meaning of CAP
CAP stands for – Consistency, Availability and Partition tolerance.
Consistency– It means the data in all the nodes is consistent. And will return the fresh/latest data when requested.
Availability– It means when a request is made, the distributed system is guaranteed to give the response.
Partition Tolerance– It refers to the ability of distributed system to continue to work even in case of network or node failure.
Due to network limitations, achieving all three simultaneously is impossible. Systems must choose between CP (Consistency + Partition Tolerance) and AP (Availability + Partition Tolerance) in real-world scenarios.
Understanding CP, AP and CA
One has to choose two properties from the three.
CONSISTENCY + PARTITION TOLERANCE (CP):
Where the system will provide latest data when requested or give an error like- the Latest data can not be provided at the moment. [As AVAILABILITY is absent- client is not guaranteed to receive a response with data]
AVAILABILITY + PARTITION TOLERANCE (AP):
A query of client is always completed with data response. Even in case of network/node failure. [As CONSISTENCY is absent- it is not guaranteed that is it latest value, server might return stale data]
CONSISTENCY + AVAILABILITY (CA):
A client query will always receive the latest data. It’s only possible in systems without partitions, which is impractical in distributed systems. [This combination assumes that there will not be any network/node failure- which is considered as impractical scenario]
Real-world examples:
CONSISTENCY + PARTITION TOLERANCE:
Imagine a bank with multiple ATMs, each connected to a central database. If network partitioning occurs (meaning ATMs lose connection to the central system), an ATM must decide between:
- Availability (allowing transactions even if the data might be outdated)
- Consistency (ensuring all ATMs always have the latest account balance)
If the system prioritizes Consistency (CP), an ATM will reject withdrawal requests during a network failure. This prevents a scenario where:
- A user withdraws $100 from one ATM.
- Due to partitioning, another ATM doesn’t get this update immediately.
- The user withdraws another $100 from the second ATM, exceeding their balance.
By sacrificing availability, the system ensures users cannot withdraw more money than they have, maintaining data integrity.
This set can be used where latest data is of utmost priority.
AVAILABILITY + PARTITION TOLERANCE:
Let’s say a creator on YouTube, who has millions of viewers uploads a video. Assume that there is network partition, So now when two different users open the video they will see different view and likes count on the video. Due to partitioning, different servers might show different counts. The data will be received but it is not guaranteed if it is latest/updated data or stale data.
This set can be used where data availability is of utmost priority.
In the real-world, no partition scenario is hard to achieve. Node or network can go down at one point or another due to various reasons.
Hence one is left with two choices – CP or AP
Recommended articles:
https://en.wikipedia.org/wiki/CAP_theorem
Recommended videos:
https://www.youtube.com/watch?v=BHqjEjzAicA