Distributed transaction
A distributed transaction is a single logical operation that spans multiple databases or data sources – which are typically located on different servers within a network – guaranteeing atomicity in distributed systems.
A distributed transaction is a type of transaction that ensures that all operations across a distributed system, such as microservices, are completed successfully, or none at all, maintaining the data integrity of the overall system.
For example, in a distributed system, placing an order might involve charging the customer’s payment method, reserving stock, updating the inventory, scheduling delivery, and sending a confirmation email. If any of these operations fail, the entire transaction should be rolled back to prevent inconsistent state.
A distributed database runs distributed transactions internally whenever a write touches more than one of its nodes. In a sharded database, any write that touches more than one shard is a distributed transaction, which is why shard keys are usually chosen to keep common writes local to a single shard. Some databases support distributed transactions natively, but otherwise implementation requires additional tools. Phased commits, such as the two-phase commit (2PC) protocol or three-phase commit (3PC) protocol, are commonly used to manage distributed transactions. Sagas are another approach that can be used to manage long-running distributed transactions without relying on traditional locking or phased commit protocols.