MVCC

Multi-Version Concurrency Control (MVCC) is a database concurrency control method that tracks multiple versions of data items to improve concurrent access. By allowing readers to access data without blocking writers, MVCC provides non-conflicting reads and reduces lock contention. It is widely used in relational databases and some key-value stores to support higher throughput and consistent reads across transactions.

How it works: When a transaction modifies data, a new version of the affected row is created and stored alongside the previous version. Older versions remain visible to transactions that began earlier, while new versions are visible to transactions started after the change. Each version carries metadata (for example, creation and expiration identifiers) that the database uses to determine visibility for a given transaction snapshot. A transaction reads a single snapshot of the database; it does not see concurrent changes that occur after its start.

Cleanup and isolation: Versions that are no longer visible to any active transaction become candidates for removal by a vacuum or garbage-collection process. MVCC supports different isolation levels, including read committed and serializable; at higher isolation levels, the system may need to ensure that the results are serializable, potentially increasing contention on write operations.

Applications and examples: PostgreSQL uses MVCC with tuple versions and transaction IDs; MySQL InnoDB implements MVCC for row-level locking; Oracle uses a similar concept in its multiversion read consistency. MVCC can reduce read latency and improve concurrency but increases storage overhead and complexity in version cleanup and conflict handling.