What is consistency in the CAP Theorem?

Consistency in the CAP Theorem refers to the guarantee that all nodes in a distributed system return the same, most up-to-date version of data for any read request, regardless of which node handles the request. When a system prioritizes consistency, it ensures that after a write operation, all subsequent reads reflect that change immediately, even if some nodes are temporarily unreachable due to network partitions. This means the system will block or return errors for read/write operations during a partition until all nodes are synchronized, rather than serving potentially stale data. For example, if you update a user’s profile on one server, every other server must agree on that new value before confirming the write to the client.

A classic example of a consistency-focused system is a traditional relational database like PostgreSQL in a distributed setup. If a network partition occurs between two database nodes, the system might reject writes until the partition heals to prevent conflicting data. Another example is Google’s Spanner database, which uses synchronized clocks and strict consensus protocols to maintain global consistency across geographically distributed nodes. These systems prioritize data accuracy over immediate availability, ensuring users never see outdated or conflicting information, even if it means slower responses or temporary unavailability during network issues.

In practice, consistency in CAP often requires trade-offs. Most real-world systems operate in environments where network partitions are inevitable, forcing designers to choose between consistency and availability. For instance, banking systems typically prioritize consistency to prevent double-spending—rejecting transactions during a partition rather than risking inconsistent balances. However, systems like social media feeds might prioritize availability, accepting writes even during partitions and resolving conflicts later. Modern databases often provide tunable consistency levels (e.g., Cassandra’s configurable read/write consistency settings) to let developers balance these trade-offs based on specific use cases.