Definition

MVCC is a concurrency control method that keeps multiple versions of data records to allow non-blocking reads[[22]](https://en.wikipedia.org/wiki/Multiversion_concurrency_control "Multiversion concurrency control - Wikipedia")[[23]](https://en.wikipedia.org/wiki/Multiversion_concurrency_control "Multiversion concurrency control - Wikipedia"). Instead of locking, transactions see the version of data that was current at their start.

Core concept

Each update creates a new version; the old version isn’t overwritten immediately. Each transaction has a timestamp or ID and sees only versions committed before its start. Thus readers don’t block writers and vice versa.

Use cases

MVCC is used by many databases (PostgreSQL, Oracle, SQL Server Snapshot, CouchDB). It provides good performance for mixed read/write workloads.

Trade-offs

Requires storing multiple versions (increased storage, need for vacuum/cleanup). Writers still conflict on the same row (only one can commit). Requires version cleanup/gc (or compaction in some systems).

Example

Transaction T1 reads row X (version 1). T2 updates X to version 2 but hasn’t committed. T1 still sees version 1 throughout. After commit, new reads see version 2.

References

“MVCC aims at solving concurrency by keeping multiple copies… Each user sees a snapshot; changes by a writer are not visible until commit”[[22]](https://en.wikipedia.org/wiki/Multiversion_concurrency_control "Multiversion concurrency control - Wikipedia")[[23]](https://en.wikipedia.org/wiki/Multiversion_concurrency_control "Multiversion concurrency control - Wikipedia").