Designing Data-Intensive Applications – Leaderless Replication

Concurrent Writes

Handling concurrent writes in distributed databases is a complex challenge. Joe Zack and Michael Outlaw discuss strategies like using logical clocks and version numbers to detect conflicts when two clients write different values simultaneously. They emphasize the importance of conflict-free replicated data types (CRDTs) in managing these issues effectively 1. Joe explains that databases must choose a strategy to resolve these conflicts, such as the "last write wins" approach, which prioritizes the most recent write based on timestamps 2.

The goal here is to eventually become consistent, not correct.

--- Joe Zack

These strategies ensure that the system remains functional even when conflicts arise, although they may not always guarantee data correctness.

   

Version Vectors

Version vectors are crucial for managing data conflicts in distributed systems. Michael Outlaw explains that version vectors track the version numbers of records across multiple replicas, helping to identify and resolve conflicts 3. This method allows databases to determine whether a write operation is an overwrite or a concurrent update, facilitating more accurate conflict resolution. Joe Zack highlights the use of dotted version vectors in systems like Riak, which send version information back to clients during reads and writes 4.

This collection of those versions is called a version vector.

--- Joe Zack

These vectors play a vital role in maintaining data consistency and integrity across distributed databases.