woundwait

Wound-wait is a deadlock-prevention strategy used in database transaction processing and distributed locking. It relies on transaction age, typically represented by a timestamp, to resolve conflicts when multiple transactions contend for the same data item. In wound-wait, when a transaction T requests a lock on an item currently held by another transaction S, the system compares their timestamps. If T is older (has an earlier timestamp) than S, the holder S is aborted (wounded) so that T can proceed. If T is younger than S, T is made to wait until the lock is released by S. The aborted transactions may be retried later with a new timestamp.

The approach enforces a global order among transactions and avoids circular wait, thereby preventing deadlocks. It favors older transactions, granting them progress by sacrificing younger ones, which are either aborted and restarted or forced to wait. Because younger transactions can be repeatedly aborted by interleaving older conflicting requests, wound-wait can lead to starvation of newcomers in heavily skewed workloads.

Applications of wound-wait include distributed databases and systems implementing timestamp-based locking or two-phase locking variants where serializability must be preserved without deadlocks. It is categorized as a pessimistic concurrency control technique, since it resolves conflicts by aborting transactions rather than allowing indefinite waiting.

Advantages include simplicity, clear deadlock avoidance, and strong progress guarantees for older transactions. Disadvantages include potential starvation of younger transactions and the overhead of abort-and-restart cycles. Related concepts include wait-die, timestamp ordering, deadlock, and two-phase locking.