Distributed Locks

In many situations the action against some context (or even single message) has to be performed in an exclusive manner. One example is an aggregator component where we have to check the message group state for the current message to determine whether we can release the group or just add that message for future consideration. For this purpose Java provides an API with java.util.concurrent.locks.Lock implementations. However, the problem becomes more complex when an application is distributed and/or run in the cluster. The locking in this case is challenging and requires some shared state and its specific approach to achieve the exclusivity requirement.

Spring Integration provides a LockRegistrty abstraction with an in-memory DefaultLockRegistry implementation based on the ReentrantLock API. The obtain(Object) method of the LockRegistrty requires a lock key for specific context. For example, an aggregator uses a correlationKey to lock operations around its group. This way different locks can be used concurrently. This obtain(Object) method returns a java.util.concurrent.locks.Lock instance (depending on the LockRegistry implementation), therefore the rest of the logic is the same as standard Java Concurrency algorithm.

Starting with version 6.2, the LockRegistry provides an executeLocked() API (default methods in this interface) to perform some task while locked. The behavior of this API is similar to well-known JdbcTemplate, JmsTemplate or RestTemplate. The following example demonstrates the usage of this API:

The method rethrows an exception from the task call, throws an InterruptedException if Lock is interrupted. In addition, a variant with Duration throws a java.util.concurrent.TimeoutException when lock.tryLock() returns false.

Spring Integration provides these LockRegistrty implementations for distributed locks:

Spring Integration AWS extension also implements a DynamoDbLockRegistry.