The Forgotten Consistency Levels of Apache Cassandra

ANY

• Description: The ANY consistency level ensures that a write operation is successful as long as it is written to at least one node. If all designated replica nodes for the data are down, the write is recorded in a hinted handoff on any available node. Once the target node is back online, the hint is played back.
• Use Case: This level is typically used to ensure the highest possible write availability, even at the cost of consistency.
• Considerations: While it guarantees that a write doesn't fail when all replicas are down, it provides the weakest consistency guarantee. Data loss is risky if the node holding the hinted handoff fails before delivering the hint to the target node.

TWO

• Description: This level requires two replica nodes to acknowledge a read or write operation.
• Use Case: Offers a middle ground between consistency and performance. TWO is used when applications need better consistency than ONE but can tolerate slightly higher latency.
• Considerations: Provides more consistency than ONE but still can't guarantee the latest write in a multi-node environment.

THREE

• Description: Acknowledgment is required from three replica nodes for a read or write operation.
• Use Case: Chosen when stronger consistency is needed and the application can tolerate higher latency.
• Considerations: It provides higher consistency at the cost of increased latency and reduced availability (compared to ONE or TWO).

ALL

• Description: Requires all replica nodes to acknowledge a read or write operation.
• Use Case: Used when the application requires strong consistency and can tolerate the highest latency and lowest availability.
• Considerations: Guarantees the most recent data is read but is susceptible to high latency and low availability, as any unavailable replica can cause the operation to fail.

EACH_QUORUM

• Description: In multi-data center clusters, EACH_QUORUM requires a quorum of nodes in each data center to respond to a read or write operation.
• Use Case: Ideal for scenarios requiring strong consistency across multiple data centers.
• Considerations: It can lead to high latencies, especially in geographically dispersed data centers, and is less tolerant of node failures in smaller data centers.

SERIAL

• Description: SERIAL is used for lightweight transactions and provides linearizable consistency for conditional updates. It ensures that conditional writes are isolated and applied atomically.
• Use Case: Essential for scenarios where a high level of consistency is required for operations like incrementing counters or updating if a condition is met.
• Considerations: It is more latency-intensive than other consistency levels and should be used judiciously to avoid performance bottlenecks.

LOCAL_SERIAL

• Description: Similar to SERIAL, but the linearizable consistency is ensured only within the local data center.
• Use Case: Useful in multi-data center setups where strong consistency is needed for lightweight transactions within a local data center, but cross-data center consistency is optional.
• Considerations: Offers a balance between consistency and performance in multi-data center environments but does not guarantee global consistency across all data centers.