EntityManagerBeanDefinitionRegistrarPostProcessor does not ensure unique SharedEntityManagerCreator beans

[imperatorx]

3.0.0-RC2 (and 1 too)

Given a trivial config:

<dependencies>
<dependency>
	<groupId>org.springframework.boot</groupId>
	<artifactId>spring-boot-starter</artifactId>
</dependency>
<dependency>
	<groupId>org.springframework.boot</groupId>
	<artifactId>spring-boot-starter-data-jpa</artifactId>
</dependency>
<dependency>
	<groupId>org.postgresql</groupId>
	<artifactId>postgresql</artifactId>
</dependency>
<dependency>
	<groupId>org.springframework.boot</groupId>
	<artifactId>spring-boot-starter-test</artifactId>
	<scope>test</scope>
</dependency>
</dependencies>

And datasource configured:

spring.datasource.driver-class-name=org.postgresql.Driver
spring.datasource.url=jdbc:postgresql://localhost:5432/test
spring.datasource.username=postgres
spring.datasource.password=postgres

And a @Component @Autowire-ing an EntityManager, the application fails to start when compiled with mvnw -Pnative native:compile, but starts normally when started in java mode from the IDE:

No qualifying bean of type 'jakarta.persistence.EntityManager' available: more than one 'primary' bean found among candidates: [org.springframework.orm.jpa.SharedEntityManagerCreator#0, org.springframework.orm.jpa.SharedEntityManagerCreator#1]

[snicoll]

@imperatorx thanks for the report but can you please share a small sample we can run ourselves? We do have smoke test for such trivial config and they are green.

[wilkinsona]

It's the injection of an EntityManager that's key and our smoke tests don't do that. The problem can be reproduced by making this change to the data:data-jpa smoke test:

diff --git a/data/data-jpa/src/main/java/com/example/data/jpa/CLR.java b/data/data-jpa/src/main/java/com/example/data/jpa/CLR.java
index 8b8ac80..612a374 100644
--- a/data/data-jpa/src/main/java/com/example/data/jpa/CLR.java
+++ b/data/data-jpa/src/main/java/com/example/data/jpa/CLR.java
@@ -5,6 +5,7 @@ import java.util.Collections;
 import java.util.List;
 import java.util.Set;
 
+import jakarta.persistence.EntityManager;
 import com.example.data.jpa.model.Author;
 import com.example.data.jpa.model.Book;
 
@@ -19,7 +20,7 @@ class CLR implements CommandLineRunner {
 
        private final BookRepository bookRepository;
 
-       CLR(AuthorRepository authorRepository, BookRepository bookRepository) {
+       CLR(AuthorRepository authorRepository, BookRepository bookRepository, EntityManager entityManager) {
                this.authorRepository = authorRepository;
                this.bookRepository = bookRepository;
        }

[imperatorx]

@imperatorx thanks for the report but can you please share a small sample we can run ourselves? We do have smoke test for such trivial config and they are green.

Sure here it is: https://github.com/imperatorx/reproducer-entitymanager

[snicoll]

It looks like SharedEntityManagerCreator is created at build-time and taken into account by AOT, and SD JPA is creating it again at runtime. It should rather check if it's there and skip it.

[mp911de]

The underlying problem is that multiple runs of EntityManagerBeanDefinitionRegistrarPostProcessor result in registering multiple auto-wire candidates of EntityManager via SharedEntityManagerCreator. We should consider unique bean names to ensure that we end up with a single EntityManager for each EntityManagerFactory bean even when running EntityManagerBeanDefinitionRegistrarPostProcessor multiple times.