Viewserializability

Viewserializability (often written as view serializability) is a correctness criterion for concurrent transaction schedules in database systems. A schedule is viewserializable if its effects are equivalent to those of some serial schedule, based on what each transaction reads (its view) and the final values produced for each data item. Informally, this means that the reads in the schedule must be explainable by reads from a single predecessor write (or from the initial value), and the final writes on each data item must match those of the chosen serial order.

Key concepts include the read-from relation and the final-write relation. For every read operation r_i(x) in a schedule, if it reads a value written by T_j, then in the equivalent serial schedule the read must originate from the same transaction T_j (or from the initial value if no prior write occurred). Additionally, the last write to each data item x in the schedule must be the same data item written by the same transaction in the serial schedule. These conditions ensure that the observable outcomes of the reads and the final states align with some serial execution.

View serializability is weaker than conflict serializability: every conflict-serializable schedule is view-serializable, but not every view-serializable schedule is conflict-serializable. To test view serializability, one can construct a view graph with transactions as nodes and edges T_i → T_j when T_i reads a value written by T_j (including an edge from an initial state node if a read uses the initial value). If this graph is acyclic, the schedule is view-serializable.

In practice, view serializability is a theoretical notion used to analyze what schedules can produce correct results; verifying it can be more complex than checking conflict serializability, and implementations often favor stronger guarantees for simplicity.