Great performance improvement for CascadingAction.PERSIST_ON_FLUSH action !! ---------------------------------------------------------------------------- Key: HHH-3967 URL: http://opensource.atlassian.com/projects/hibernate/browse/HHH-3967 Project: Hibernate Core Issue Type: Improvement Components: core Affects Versions: 3.3.1 Environment: 3.3.1 GA , SQLServer Reporter: Guenther Demetz Attachments: Cascade.java Original Estimate: 30 minutes Remaining Estimate: 30 minutes Sometimes large stateful transactions are unavoidable and especially when having several large persistent collections in dirty state, flushing becomes gradually more cpu intensive and slowly. Analizing several stacktraces I detected that the thread most time is executing at the same point in reflection , see here a typical stacktrace: java.lang.Class.isAssignableFrom(Native Method) sun.reflect.UnsafeFieldAccessorImpl.ensureObj(UnsafeFieldAccessorImpl.java:36) ... java.lang.reflect.Field.get(Field.java:358) DirectPropertyAccessor$DirectGetter.get(DirectPropertyAccessor.java:55) PojoEntityTuplizer(AbstractEntityTuplizer).getPropertyValue(Object, int) line: 300 SingleTableEntityPersister(AbstractEntityPersister).getPropertyValue(Object, int, EntityMode) line: 3609 Cascade.cascade(EntityPersister, Object, Object) line: 172 ... Making further investigations, I saw that: - CascadingAction.PERSIST_ON_FLUSH requires NoCascadeChecking - for all kind of properties the noCascade implementation is called - regarding noCascade implementation checks only properties of type Entity ! public static final CascadingAction PERSIST_ON_FLUSH = new CascadingAction() { ... public void noCascade(EventSource session, Object child, Object parent, EntityPersister persister, int propertyIndex) { ... Type type = persister.getPropertyTypes()[propertyIndex]; if ( type.isEntityType() ) { ... check and throw eventually a TransientObjectException } } Thus most property values are retrieved by reflection in vain as they are not of type entity. With following code addition in org.hibernate.engine.Cascade.java I was able do avoid most reflection method calls without affecting the behaviour: org.hibernate.engine.Cascade.java ... public void cascade(final EntityPersister persister, final Object parent, final Object anything) throws HibernateException { ... else if ( action.requiresNoCascadeChecking() ) { // Line:162 // begin Improvement if (action == CascadingAction.PERSIST_ON_FLUSH) { Type type = persister.getPropertyTypes()[i]; if ( !type.isEntityType() ) { // Remark: goes only well as long PERSIST_ON_FLUSH noCascade implementation checks only entities continue; } } // end Improvement action.noCascade( eventSource, persister.getPropertyValue( parent, i, entityMode ), parent, persister, i ); ... best regards Guenther D.

Great performance improvement for CascadingAction.PERSIST_ON_FLUSH action !! ---------------------------------------------------------------------------- Key: HHH-3967 URL: http://opensource.atlassian.com/projects/hibernate/browse/HHH-3967 Project: Hibernate Core Issue Type: Improvement Components: core Affects Versions: 3.3.1 Environment: 3.3.1 GA , SQLServer Reporter: Guenther Demetz Assignee: Gail Badner Attachments: Cascade.java Original Estimate: 0.5h Remaining Estimate: 0.5h Sometimes large stateful transactions are unavoidable and especially when having several large persistent collections in dirty state, flushing becomes gradually more cpu intensive and slowly. Analizing several stacktraces I detected that the thread most time is executing at the same point in reflection , see here a typical stacktrace: java.lang.Class.isAssignableFrom(Native Method) sun.reflect.UnsafeFieldAccessorImpl.ensureObj(UnsafeFieldAccessorImpl.java:36) ... java.lang.reflect.Field.get(Field.java:358) DirectPropertyAccessor$DirectGetter.get(DirectPropertyAccessor.java:55) PojoEntityTuplizer(AbstractEntityTuplizer).getPropertyValue(Object, int) line: 300 SingleTableEntityPersister(AbstractEntityPersister).getPropertyValue(Object, int, EntityMode) line: 3609 Cascade.cascade(EntityPersister, Object, Object) line: 172 ... Making further investigations, I saw that: - CascadingAction.PERSIST_ON_FLUSH requires NoCascadeChecking - for all kind of properties the noCascade implementation is called - regarding noCascade implementation checks only properties of type Entity ! public static final CascadingAction PERSIST_ON_FLUSH = new CascadingAction() { ... public void noCascade(EventSource session, Object child, Object parent, EntityPersister persister, int propertyIndex) { ... Type type = persister.getPropertyTypes()[propertyIndex]; if ( type.isEntityType() ) { ... check and throw eventually a TransientObjectException } } Thus most property values are retrieved by reflection in vain as they are not of type entity. With following code addition in org.hibernate.engine.Cascade.java I was able do avoid most reflection method calls without affecting the behaviour: org.hibernate.engine.Cascade.java ... public void cascade(final EntityPersister persister, final Object parent, final Object anything) throws HibernateException { ... else if ( action.requiresNoCascadeChecking() ) { // Line:162 // begin Improvement if (action == CascadingAction.PERSIST_ON_FLUSH) { Type type = persister.getPropertyTypes()[i]; if ( !type.isEntityType() ) { // Remark: goes only well as long PERSIST_ON_FLUSH noCascade implementation checks only entities continue; } } // end Improvement action.noCascade( eventSource, persister.getPropertyValue( parent, i, entityMode ), parent, persister, i ); ... best regards Guenther D.

Great performance improvement for CascadingAction.PERSIST_ON_FLUSH action !! ---------------------------------------------------------------------------- Key: HHH-3967 URL: http://opensource.atlassian.com/projects/hibernate/browse/HHH-3967 Project: Hibernate ORM Issue Type: Improvement Components: core Affects Versions: 3.3.1 Environment: 3.3.1 GA , SQLServer Reporter: Guenther Demetz Assignee: Gail Badner Fix For: 4.1.0 Attachments: Cascade.java Original Estimate: 0.5h Remaining Estimate: 0.5h Sometimes large stateful transactions are unavoidable and especially when having several large persistent collections in dirty state, flushing becomes gradually more cpu intensive and slowly. Analizing several stacktraces I detected that the thread most time is executing at the same point in reflection , see here a typical stacktrace: java.lang.Class.isAssignableFrom(Native Method) sun.reflect.UnsafeFieldAccessorImpl.ensureObj(UnsafeFieldAccessorImpl.java:36) ... java.lang.reflect.Field.get(Field.java:358) DirectPropertyAccessor$DirectGetter.get(DirectPropertyAccessor.java:55) PojoEntityTuplizer(AbstractEntityTuplizer).getPropertyValue(Object, int) line: 300 SingleTableEntityPersister(AbstractEntityPersister).getPropertyValue(Object, int, EntityMode) line: 3609 Cascade.cascade(EntityPersister, Object, Object) line: 172 ... Making further investigations, I saw that: - CascadingAction.PERSIST_ON_FLUSH requires NoCascadeChecking - for all kind of properties the noCascade implementation is called - regarding noCascade implementation checks only properties of type Entity ! public static final CascadingAction PERSIST_ON_FLUSH = new CascadingAction() { ... public void noCascade(EventSource session, Object child, Object parent, EntityPersister persister, int propertyIndex) { ... Type type = persister.getPropertyTypes()[propertyIndex]; if ( type.isEntityType() ) { ... check and throw eventually a TransientObjectException } } Thus most property values are retrieved by reflection in vain as they are not of type entity. With following code addition in org.hibernate.engine.Cascade.java I was able do avoid most reflection method calls without affecting the behaviour: org.hibernate.engine.Cascade.java ... public void cascade(final EntityPersister persister, final Object parent, final Object anything) throws HibernateException { ... else if ( action.requiresNoCascadeChecking() ) { // Line:162 // begin Improvement if (action == CascadingAction.PERSIST_ON_FLUSH) { Type type = persister.getPropertyTypes()[i]; if ( !type.isEntityType() ) { // Remark: goes only well as long PERSIST_ON_FLUSH noCascade implementation checks only entities continue; } } // end Improvement action.noCascade( eventSource, persister.getPropertyValue( parent, i, entityMode ), parent, persister, i ); ... best regards Guenther D.

Great performance improvement for CascadingAction.PERSIST_ON_FLUSH action !! ---------------------------------------------------------------------------- Key: HHH-3967 URL: http://opensource.atlassian.com/projects/hibernate/browse/HHH-3967 Project: Hibernate ORM Issue Type: Improvement Components: core Affects Versions: 3.3.1 Environment: 3.3.1 GA , SQLServer Reporter: Guenther Demetz Assignee: Gail Badner Fix For: 5.0.0 Attachments: Cascade.java Original Estimate: 0.5h Remaining Estimate: 0.5h Sometimes large stateful transactions are unavoidable and especially when having several large persistent collections in dirty state, flushing becomes gradually more cpu intensive and slowly. Analizing several stacktraces I detected that the thread most time is executing at the same point in reflection , see here a typical stacktrace: java.lang.Class.isAssignableFrom(Native Method) sun.reflect.UnsafeFieldAccessorImpl.ensureObj(UnsafeFieldAccessorImpl.java:36) ... java.lang.reflect.Field.get(Field.java:358) DirectPropertyAccessor$DirectGetter.get(DirectPropertyAccessor.java:55) PojoEntityTuplizer(AbstractEntityTuplizer).getPropertyValue(Object, int) line: 300 SingleTableEntityPersister(AbstractEntityPersister).getPropertyValue(Object, int, EntityMode) line: 3609 Cascade.cascade(EntityPersister, Object, Object) line: 172 ... Making further investigations, I saw that: - CascadingAction.PERSIST_ON_FLUSH requires NoCascadeChecking - for all kind of properties the noCascade implementation is called - regarding noCascade implementation checks only properties of type Entity ! public static final CascadingAction PERSIST_ON_FLUSH = new CascadingAction() { ... public void noCascade(EventSource session, Object child, Object parent, EntityPersister persister, int propertyIndex) { ... Type type = persister.getPropertyTypes()[propertyIndex]; if ( type.isEntityType() ) { ... check and throw eventually a TransientObjectException } } Thus most property values are retrieved by reflection in vain as they are not of type entity. With following code addition in org.hibernate.engine.Cascade.java I was able do avoid most reflection method calls without affecting the behaviour: org.hibernate.engine.Cascade.java ... public void cascade(final EntityPersister persister, final Object parent, final Object anything) throws HibernateException { ... else if ( action.requiresNoCascadeChecking() ) { // Line:162 // begin Improvement if (action == CascadingAction.PERSIST_ON_FLUSH) { Type type = persister.getPropertyTypes()[i]; if ( !type.isEntityType() ) { // Remark: goes only well as long PERSIST_ON_FLUSH noCascade implementation checks only entities continue; } } // end Improvement action.noCascade( eventSource, persister.getPropertyValue( parent, i, entityMode ), parent, persister, i ); ... best regards Guenther D.