concurrency-control

Prevent lost updates and race conditions by applying invariant-first concurrency design that selects the narrowest control strategy, such as optimistic locking, atomic SQL/CAS, or distributed leases with fencing, to ensure deterministic conflict behavior.

Handle duplicate submissions by implementing idempotency key storage and deduplication, which ensures that duplicate-submit risks from parallel workers or network retries do not result in duplicate effects on shared resources.

Avoid deadlocks by enforcing deterministic lock ordering and bounding lock or lease durations, which eliminates circular waits and prevents indefinite blocking when multiple actors acquire locks on shared resources.

You need the outbox pattern or post-commit guarantees when publishing events concurrently to prevent publish-before-commit anomalies, ensuring event publish safety alongside database mutations.

Queue partitioning works for at-least-once delivery by sharding execution, ensuring parallel workers process partitioned queues deterministically while idempotency keys handle any duplicate processing.

Choose between optimistic and pessimistic locking by naming the exact invariant that must survive overlap, then selecting the narrowest mechanism with explicit conflict behavior documentation, such as atomic SQL/CAS for simple updates or distributed leases for cross-node coordination.