Spring Data 4: Data Access Revisited

The Spring Data team is preparing a new generation of the Data framework for 2025 and beyond: Spring Data 4 will ship with Jakarta EE 11 and Kotlin 2 baseline upgrades, introducing major new themes of which AOT Repositories, JSpecify annotations, Vector store support, and a major Spring Data JPA revision are only a selection. This talk gives an outlook on what’s to come and focuses on key design decisions and roadmap considerations.

Transcript

1. Mark Paluch @mp911de • https://bsky.app/profile/mp911.de Spring Data 4 Data Access Revisited
2. Agenda • Spring Data 3.5 (May Release Train 2025.0) • Spring Data 4.0 (November Release Train 2025.1)
3. Spring Data 3.5
4. Spring Data 3.5 (May) • Vector Search: MongoDB, Cassandra • MongoDB: Queryable Encryption • Cassandra: Baseline upgrade, SAI Indexes • JPA: DTO Projections • JDBC & R2DBC: @Sequence support
5. Spring Data 3.5 (May or: 2025.0.0) • Vector Search: MongoDB, Cassandra • MongoDB: Queryable Encryption • Cassandra: Baseline upgrade, SAI Indexes • JPA: DTO Projections • JDBC & R2DBC: @Sequence support
6. Vector Search • Types: Vector • Encapsulates accidental complexity • Modules: MongoDB and Apache Cassandra
7. MongoVector.of(BinaryVector.int8Vector(…)) MongoVector.ofFloat(…) vs.
8. Spring Data MongoDB 4.5 • Vector Search: Aggregation Stage • Queryable Encryption: Schema support, Experimental converter
9. MongoDB Vector Search VectorSearchOperation $vectorSearch = VectorSearchOperation.search("cosine-index") .path("embedding") .vector(Vector.of(…)) .limit(Limit.of(10)) .numCandidates(20) .searchType(SearchType.ANN) .withSearchScore(); AggregationResults<Document> results = template.aggregate(newAggregation($vectorSearch), Comment.class, Document.class);
10. MongoDB Encryption • CSFLE: Client-Side Field Encryption (since 3.3) • Queryable Encryption: Range and Equality Queries (4.5) • Schema Support (Collection Creation) ⚠ Experimental Converters • Strict and Rigid
11. MongoDB Queryable Encryption class Person { String id; // @ValueConverter(MongoEncryptionConverter.class) @Encrypted(algorithm = "Indexed") @Queryable(queryType = "equality", contentionFactor = 0) String ssn; // @ValueConverter(MongoEncryptionConverter.class) @Encrypted(algorithm = "Range") @Queryable(queryType = "range", contentionFactor = 0L, queryAttributes = "{ 'sparsity': 0 }") Integer age; // @ValueConverter(MongoEncryptionConverter.class) // @Encrypted + @Queryable @RangeEncrypted(contentionFactor = 0L, rangeOptions = "{\"min\": 0, \"max\": 200, \"trimFactor\": 1, \"sparsity\": 1}") Integer income; }
12. Spring Data for Apache Cassandra 4.5 • Baseline Upgrade: Apache Cassandra 5 • Storage-Attached Indexes (SAI) • Vector Indexing via SAI • Vector Search: CQL support
13. Cassandra Vector Search Vector vector = Vector.of(…); Columns columns = Columns.from("comment") .select("vector", it -> it.similarity(vector).cosine().as("similarity")); Query query = Query.select(columns).limit(3).sort(VectorSort.ann("vector", vector)); List<CommentSearch> result = template.query(Comments.class) .as(CommentSearch.class) .matching(query) .all(); @Table class Comments { @Id UUID id; String comment; @VectorType(dimensions = 5) @SaiIndexed Vector vector; } class CommentSearch { }
14. Cassandra Vector Search Vector vector = Vector.of(…); Columns columns = Columns.from("comment") .select("vector", it -> it.similarity(vector).cosine().as("similarity")); Query query = Query.select(columns).limit(3).sort(VectorSort.ann("vector", vector)); List<CommentSearch> result = template.query(Comments.class) .as(CommentSearch.class) .matching(query) .all(); @Table class Comments { @Id UUID id; String comment; @VectorType(dimensions = 5) @SaiIndexed Vector vector; } class CommentSearch { String comment; float similarity; }
15. Cassandra Vector Search Vector vector = Vector.of(…); Columns columns = Columns.from("comment") .select("vector", it -> it.similarity(vector).cosine().as("similarity")); @Table class Comments { } class CommentSearch { } Query query = Query.select(columns).limit(3).sort(VectorSort.ann("vector", vector)); List<CommentSearch> result = template.query(Comments.class) .as(CommentSearch.class) .matching(query) .all();
16. Cassandra Vector Search Vector vector = Vector.of(…); Columns columns = Columns.from("comment") .select("vector", it -> it.similarity(vector).cosine().as("similarity")); @Table class Comments { } class CommentSearch { } Query query = Query.select(columns).limit(3).sort(VectorSort.ann("vector", vector)); List<CommentSearch> result = template.query(Comments.class) .as(CommentSearch.class) .matching(query) .all();
17. Cassandra Vector Search Vector vector = Vector.of(…); Columns columns = Columns.from("comment") .select("vector", it -> it.similarity(vector).cosine().as("similarity")); Query query = Query.select(columns).limit(3).sort(VectorSort.ann("vector", vector)); List<CommentSearch> result = template.query(Comments.class) .as(CommentSearch.class) .matching(query) .all(); @Table class Comments { @Id UUID id; String comment; @VectorType(dimensions = 5) @SaiIndexed Vector vector; } class CommentSearch { String comment; float similarity; }
18. Spring Data JPA 3.5 • JPQL Parsers: Refinements and bugfixes • Detail revisions: Fluent API Projections, Nulls Precedence in String queries • Projections: DTO Constructor Expressions
19. JPA Constructor Expression for DTO Projections public interface UserRepository extends Repository<User, Integer> { @Query("select new com.acme.myapp.UserExcerpt(u.firstname, u.lastname) from User u") List<UserExcerpt> findRecordProjection(); } record UserExcerpt(String firstname, String lastname) { }
20. Constructor Expression Derivation public interface UserRepository extends Repository<User, Integer> { @Query("select u.firstname, u.lastname from User u") List<UserExcerpt> findRecordProjection(); } record UserExcerpt(String firstname, String lastname) { }
21. Even Better Constructor Expression Derivation public interface UserRepository extends Repository<User, Integer> { @Query("select u from User u") List<UserExcerpt> findRecordProjection(); } record UserExcerpt(String firstname, String lastname) { }
22. Upcoming Spring Data 4.0
23. Spring Data 4.0 (November 2025.1) • Baseline Upgrade: Spring Framework 7, JSpecify 1.0, Kotlin 2.1, Jakarta EE 11 • Ahead-of-Time Repositories: JPA, MongoDB • JPA: JPA 3.2, JPQL Derived Queries, Revised Query Parsing & Enhancing • JDBC: Composite Id’s • Fluent API refinements: MongoDB, Cassandra
24. Repository Query Methods repository.countByLastname("White")
26. • Introspection: What Query is actually being executed? • Debugging: Where is the Query executed? Repository Query Methods
27. public Long countByLastname(String lastname) { String queryString = "SELECT COUNT(u) FROM User u WHERE u.lastname = :lastname"; Query query = this.entityManager.createQuery(queryString); query.setParameter("lastname", lastname); return (Long) query.getSingleResultOrNull(); }
28. @Generated public class UserRepositoryImpl__Aot extends AotRepositoryFragmentSupport { private final EntityManager entityManager; public UserRepositoryImpl__Aot(EntityManager em, FragmentCreationContext ctx) { … } public Long countByLastname(String lastname) { String queryString = "SELECT COUNT(u) FROM User u WHERE u.lastname = :lastname"; Query query = this.entityManager.createQuery(queryString); query.setParameter("lastname", lastname); return (Long) query.getSingleResultOrNull(); } // … }
29. • Opt-in: spring.aot.repositories.enabled=true • Generated Query Method Bodies • Only code required to run the query • Debuggable Query Methods 🤩 • Tooling Support: Repository.json AOT Repositories
30. JPA Repositories Startup Performance Milliseconds 0 1300 2600 3900 5200 Regular AOT AOT Repositories
31. JPA Repositories Startup Performance Milliseconds 0 1300 2600 3900 5200 Regular AOT AOT Repositories -11 % -9 %
32. JPA Repositories Startup Performance Memory 0 85 170 255 340 Regular AOT AOT Repositories Bootstrap First Request
33. JPA Repositories Startup Performance Memory 0 85 170 255 340 Regular AOT AOT Repositories Bootstrap First Request -11 % -4 %
34. { "name": "com.acme.UserRepository", "module": "JPA", "type": "IMPERATIVE", "methods": [ { "name": "countUsersByLastname", "signature": "public abstract java.lang.Long com.acme.UserRepository.countUsersByLastname(java.lang.String)", "query": { "query": "SELECT COUNT(u) FROM com.acme.User u WHERE u.lastname = :lastname" } }, { "name": "findPagedWithNamedCountByEmailAddress", "signature": "public abstract Page<com.acme.User> com.acme.UserRepository.findPagedWithNamedCountByEmailAddress(Pageable,java.lang.String)", "query": { "name": "User.findByEmailAddress", "query": "SELECT u FROM User u WHERE u.emailAddress = ?1", "count-name": "User.findByEmailAddress.count-provided", "count-query": "SELECT count(u) FROM User u WHERE u.emailAddress = ?1" } }, { "name": "saveAll", "signature": "public abstract <S extends T> java.lang.Iterable<S> org.springframework.data.repository.CrudRepository.saveAll(java.lang.Iterable<S>)", "fragment": { "interface": "org.springframework.data.jpa.repository.support.SimpleJpaRepository", "fragment": "org.springframework.data.jpa.repository.support.SimpleJpaRepository" } } ] }
37. • Generated Bean Definition wiring • Application must run in AOT mode (Native Image or -Dspring.aot.enabled=true) • Requires AOT processing • Can get out of sync • Requires AOT re-processing AOT Repositories (cont’d) mvn clean package gradle processAot
38. • Not every Method should be generated • Caching of rewrite inputs • Queries can become dynamic: Range<T> Queries, IS NULL • JPA (Hibernate only) and MongoDB available now, more modules to follow AOT Repository Method Limitations
39. Spring Data JPA 4.0
40. • Baseline Upgrade: JPA 3.2, Hibernate 7, Eclipselink 5 • Query Derivation: JPQL instead of Criteria API • Specifications: Revised Specification design • Query Enhancer Selectors: Configuration of Query Enhancers (e.g. JSqlParser) • JPQL Parsers: Limited JpaSort.unsafe(…) parsing for Specifications Spring Data JPA 4.0
41. Spring Data JPA 3.5 with Hibernate Queries per Second 0 100 200 300 400 500 600 700 JPQL Criteria Queries In thousands, H2 In-Memory Database via EntityManager
42. Spring Data JPA 4.0 with Hibernate Queries per Second 0 100 200 300 400 500 600 700 Derived Queries @Query EntityManager In thousands, H2 In-Memory Database, JPQL Queries
43. • Missing Update support, broken Delete support (nullability) • PredicateSpecification: Composition of Predicates • New variants: UpdateSpecification, DeleteSpecification Specification Revision
44. UpdateSpecification in Action PredicateSpecification<User> predicate = userHasFirstname("Oliver") .and(userHasLastname("Gierke")); UpdateSpecification<User> updateLastname = UpdateSpecification .<User> update((root, update, criteriaBuilder) -> update.set("lastname", "Drotbohm")) .where(predicate); long updated = repository.update(updateLastname);
45. UpdateSpecification in Action PredicateSpecification<User> predicate = userHasFirstname("Oliver") .and(userHasLastname("Gierke")); UpdateSpecification<User> updateLastname = UpdateSpecification .<User> update((root, update, criteriaBuilder) -> update.set("lastname", "Drotbohm")) .where(predicate); long updated = repository.update(updateLastname);
46. UpdateSpecification in Action PredicateSpecification<User> predicate = userHasFirstname("Oliver") .and(userHasLastname("Gierke")); UpdateSpecification<User> updateLastname = UpdateSpecification .<User> update((root, update, criteriaBuilder) -> update.set("lastname", "Drotbohm")) .where(predicate); long updated = repository.update(updateLastname);
47. Spring Data MongoDB 5.0 Spring Data Cassandra 5.0
48. • Type-based: Entity, Projection, or Converter<Document, Person> • EntityCallbacks: Application-wide or per Type Result Mapping Variants List<PersonProjection> result = template.query(Person.class) .as(PersonProjection.class) .matching(query) .all();
49. • Query Result Converter: Contextual conversion during query result reading • Actual result: Raw query result • Contextual Converter: Wrapping, custom mapping Fluent API Refinement List<PersonProjection> result = template.query(Person.class) .as(PersonProjection.class) .matching(query) .map((document, reader) -> ) .all();
50. But there is one more thing…
51. interface VectorSearchRepository extends JpaRepository<Comment, Integer> { SearchResults<Comment> searchTop5ByEmbeddingWithin(Vector embedding, Score distance); } Vector vector = Vector.of(…); SearchResults<Comment> results = repository.searchTop5ByEmbeddingWithin(vector,
52. interface VectorSearchRepository extends JpaRepository<Comment, Integer> { SearchResults<Comment> searchTop5ByEmbeddingWithin(Vector embedding, Score distance); } Vector vector = Vector.of(…); SearchResults<Comment> results = repository.searchTop5ByEmbeddingWithin(vector, Similarity.of(0.4, ScoringFunction.cosine()));
53. • Familiar Programming Model: Declarative Search Methods • Store Variance: Distance vs similarity, pre-filters, scoring functions in index • Similarity: Transparent normalization of Vector distance into similarity Repository Vector Search Methods
54. • Available Today: JPA via Hibernate 7 (Oracle 23, pgvector), MongoDB, Cassandra • Spring Data 2025.1.0-M3 • Exploring: Neo4j, Elasticsearch, JDBC, and R2DBC • Later today 15:30, Room 6: Modular RAG Architectures with Java and Spring AI by Thomas Vitale Feedback welcome! Repository Vector Search Methods
55. Mark Paluch @mp911de • https://bsky.app/profile/mp911.de Thanks for attending