Skip to content

Consistency

Consistency is an overloaded term with multiple uses.

Flavors of Consistency

  1. CAP Consistency/Strong Consistency/Linearizability/Strict Consistency/Immediate Consistency

    • Means that simultaneous reads at different nodes in a distributed system always return the same data. This holds even if there is a simultaneous write at one of the nodes.
    • Example:
    • Updating passwords

  2. Weak Consistency

    • Means that simultaneous reads at different nodes could yield different values. Nodes can diverge in state, and there is no guarantee that they will converge - which is different from eventual consistency!

  3. Eventual Consistency

    • Type of weak consistency where if there are no new writes, data at all nodes will eventually converge. Updates and writes arepropogated throughout the nodes of a system, but those changes will not be immediately reflected. This is associated with BASE databases.
    • Examples:
      • The Domain Name System (DNS) is highly available with high throughput, but may not reflect the latest value at any given time.

  4. Sequential Consistency

    • Means that the ordering of write requests at a single node is preserved in a distributed system. Two sequential writes cannot be applied out of order at another node in the system.
    • Example:
    • Ordering of scial media posts of friends. We expect that for a given friend the comments or news posts are displayed in the order they were submitted.

  5. ACID Consistency

    • Means that data stored by a database is in a correct and valid state - e.g. if a schema specifies that a value must be unique, the database will ensure that the value is unique throughout all actions. If a transaction violates data integrity constraints, that transaction is rejected, and the state is reverted. This is associated with ACID databases.
    • Example:
      • if a foreign key is deleted, associated rows in other tables will also be deleted.

Tradeoffs

A common decision to make is the choice between strong and weak consistency - e.g. a database with BASE characteristics, transactions have eventual consistency, a form of weak consistency.

This is a choice because achieving consistency in a distributed system has additional overhead and costs e.g. synchronization between nodes, verification mechanism to verify data is most up to date,