Sample app showing differences in Spring startup based on run method.

Pre-conditions:

• using EntityManagerFactoryBuilder instead of automatic spring creation
  • org.springframework.boot.autoconfigure.orm.jpa.JpaBaseConfiguration.entityManagerFactory calls org.springframework.boot.autoconfigure.orm.jpa.HibernateProperties.applyScanner disables Hibernate Scanner by default
• having a lot of large files in /src/main/resource/static/
  • here auto-generated with for i in $(seq -w 1 200); do dd if=/dev/urandom of=filler_$i.bin bs=1K count=1000; done

Times observed (quite old hardware):

• ran each 3-4 times, times consistent to the first decimal
• executable jar java -jar spring-startup-test-0.0.1-SNAPSHOT.jar
  • Started SpringStartupTestApplication in 7.248 seconds (process running for 7.856)
• unpacked jar with jar -xf spring-startup-test-0.0.1-SNAPSHOT.jar
• JarLauncher java org.springframework.boot.loader.launch.JarLauncher
  • Started SpringStartupTestApplication in 4.567 seconds (process running for 4.941)
• natural main method java -cp "BOOT-INF/classes:BOOT-INF/lib/*" com.example.MyApplication
  • Started SpringStartupTestApplication in 4.243 seconds (process running for 4.561)

Notes:

• depends a lot on actual files both amount and size
• having multiple different persistence context (like for different databases) incurs the Hibernate Scanner cost for each of them
• hibernate scanner can be disabled by calling .properties(Map.of("hibernate.archive.scanner","org.hibernate.boot.archive.scan.internal.DisabledScanner")) on EntityManagerFactoryBuilder