JpaRepository memory leak with Pageable argument [DATAJPA-1647]

[spring-projects-issues]

Oleg Kolychev opened DATAJPA-1647 and commented

There seems to be a memory leak when JPA repository findBy... method accepts Pageable argument. If I remove Pageable from repository method signature, there is no leak.

Attached simple reproducible project that run repository query method in infinite cycle on ApplicationStartedEvent. If I run it with low heap limit (like 50-100mb) I get OutOfMemoryError in few seconds.

Tried with H2 and PostgreSQL databases, both cases have memory leaks.

Might also be related to Hibernate somehow, but I am not sure how to track it down to hibernate code.

Also screenshots from VisualVM attached, that shows Spring Data JPA related objects are in top.

 

 

---

Affects: 2.2.3 (Moore SR3)

Attachments:

• image-2019-12-15-22-57-37-213.png (186.66 kB)
• image-2019-12-15-22-59-15-486.png (317.08 kB)
• jpa-memleak.zip (5.68 kB)

Issue Links:

• DATAJPA-1575 Performance improvements in repository invocation handling

Referenced from: pull request #402, and commits 0b70952, 3476a21, 25ebe74, 88b942b

Backported to: 2.2.4 (Moore SR4)

[spring-projects-issues]

Jens Schauder commented

This is related to DATAJPA-1575 which introduced caching fo rparameter data. Unfortunatly it uses a class without proper equals and hasCode (ParameterMetadata)implementation as a map key

[spring-projects-issues]

Mark Paluch commented

We introduced the cache to avoid recomputation as createCriteriaBinder heavily create new collections. Shouldn't the fix rather be implementing proper equals and hashCode methods Jens Schauder?

[spring-projects-issues]

Jens Schauder commented

Maybe.

I didn't do anything in this directions for two reasons:

1. the memoization never worked as far as I could tell, so I wasn't not at all sure that it is actually beneficial/significant.
2. ParameterMetadata has indirectly a reference to the EntityManager
   I think it was via the ParameterExpression.
   Therefore it can't be easily memoized by just adding an equals and hashCode implementation.

I vote for having a separate issue for implementing proper memoization if it is worthwhile

[spring-projects-issues]

Mark Paluch commented

Good point. I didn't catch the reference to EntityManager in the first place. I'll move forward with merging the PR. Thanks a lot