Infinite loop in unsynchronized HashMap in CrudMethodMetadataPostProcessor [DATAJPA-620]

[spring-projects-issues]

Henning Sudbrock opened DATAJPA-620 and commented

The method org.springframework.data.jpa.repository.support.CrudMethodMetadataPostProcessor.CrudMethodMetadataPopulatingMethodIntercceptor#invoke is not thread-safe, although it is called concurrently when Spring-Data-JPA-Queries are executed concurrently.

More concretely, since the method uses the non-thread-safe java.util.HashMap, concurrent calls to this method may result in an infinite loop when executing the Hashmap-Lookup 'metadataCache.get(method)' (line 101 in version 1.7.0.RELEASE), and I have actually encountered this behavior in an application. (Infinite loops in concurrently used hashmaps appear to be a known problem, see, e.g., https://jira.spring.io/browse/DATACMNS-518 or http://mailinator.blogspot.dk/2009/06/beautiful-race-condition.html.)

I therefore believe that this method should be made thread-safe, e.g., by using a java.util.concurrent.ConcurrentHashMap

---

Affects: 1.7 GA (Evans)

Referenced from: pull request #111, and commits 532adb1, 99a35c4, 7210a5c

Backported to: 1.7.1 (Evans SR1), 1.6.5 (Dijkstra SR5)

[spring-projects-issues]

Thomas Darimont commented

I replaced the HM with a CHM, please revise