Hibernate Pagination and Collection Fetching Memory Warning

Direct answer

Hibernate issues HHH000104 and falls back to in-memory pagination when you combine setFirstResult() and setMaxResults() with a fetch join on a collection-valued association, typically OneToMany or ManyToMany mapped as List, Set, or Bag. A fetch join on single-ended associations such as ManyToOne or OneToOne normally does not trigger this warning because those joins do not multiply root rows.

The cause is relational: joining a collection produces one database row per child element. A database LIMIT or OFFSET could cut through a parent's collection, returning a partial collection or losing the parent entirely. To avoid that, Hibernate loads the full matching result set and applies the page boundary in the JVM.

Confirmed behavior versus likely explanation

Item	Status
HHH000104 appears with collection fetch joins plus pagination	Well-established Hibernate ORM behavior
Database LIMIT/OFFSET is suppressed when the warning fires	Likely; verify with SQL logging for your dialect and version
Exact warning text and trigger vary by Hibernate ORM version	Version-dependent; check your dependency tree

Heap implications

When in-memory pagination applies, setMaxResults() does not bound the rows the database returns. Hibernate hydrates every matching parent row and its fetched collection into the persistence context, then discards everything outside the requested page. Heap pressure therefore scales with total matching rows multiplied by average collection size, not with the page size. Duplicate parent rows from the cartesian product add transient object overhead before de-duplication. If the transaction or session stays open, the first-level cache retains those entities until the session closes, extending retention and increasing GC pressure or OutOfMemoryError risk.

Steps for this case

1. Enable SQL logging (hibernate.show_sql and hibernate.format_sql) and reproduce the query. Confirm whether the SQL contains LIMIT/OFFSET and whether HHH000104 appears.
2. Remove the collection fetch join from the paginated query. Page the root entity using setFirstResult() and setMaxResults() so the database applies the limit.
3. Initialize collections afterward using @BatchSize, Hibernate.initialize(), or a second batch query keyed by the returned parent IDs.
4. Consider a DTO projection or scrollable streaming if you only need read-only collection data.

Verification

Check the Hibernate ORM version in your dependency tree, because in-memory pagination behavior is version-sensitive. Run the same page against thousands of parents with large collections, with and without the collection fetch join, and compare heap usage using a profiler or -XX:+HeapDumpOnOutOfMemoryError.

Missing diagnostic detail: which Hibernate ORM version you are using. Share it if the SQL log does not clearly show whether LIMIT/OFFSET is being pushed to the database.