Definition

Snapshot isolation is a transaction isolation level where each transaction operates on a snapshot of the database as of the start of the transaction[[21]](https://en.wikipedia.org/wiki/Snapshot_isolation "Snapshot isolation - Wikipedia"). Transactions see a consistent view of data that doesn’t change during the transaction. A transaction under snapshot isolation can commit only if its writes don’t conflict with concurrent writes.

Core concept

Readers don’t block writers and vice versa. Each transaction sees the last committed state at its start time. Writes use versioning: on commit, if two transactions have modified the same row, one will abort (to prevent write-write conflict). This avoids many deadlocks of stricter locks.

Use cases

Common in modern databases (PostgreSQL’s default), providing high concurrency. Good for read-heavy workloads where you want consistency within a transaction but still allow concurrent updates.

Trade-offs

Snapshot isolation is not fully serializable. It can still exhibit write skew anomalies. It provides a strong, but not strict, guarantee. It uses more storage (versions) to support multi-version concurrency (see MVCC).

Example

Two transactions T1 and T2 both read a row and try to update it. Under snapshot, whichever commits first will succeed; the second will abort because its write would conflict.

References

“Snapshot isolation… all reads in a transaction see a consistent snapshot of the DB (values at start), and a transaction commits only if no updates conflict with concurrent writes”[[21]](https://en.wikipedia.org/wiki/Snapshot_isolation "Snapshot isolation - Wikipedia").