Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [eclipselink-users] Order of persist operations

I am saying that for a properly mapped and populated object graph, it shouldn't matter.

On 17/01/2013 8:55 AM, Deyan Tsvetanov wrote:
So you don't see a problem that eclipselink randomises the order of the db operations when it can't determine it from the relationships ?



On Jan 17, 2013, at 15:52 , Tom Ware <tom.ware@xxxxxxxxxx> wrote:

As soon as you add the possibility of cascading (a highly used JPA feature), there is no choice but for the persistence provider to pick an order.   With cascading, one persist or merge operation can result in a large set of insert, update and delete operations.  A JPA provider(any JPA provider) needs to be able to execute these statements in an order that allows it to satisfy all the database constraints it is aware of.

The key thing is that if you provide a JPA provider a properly mapped and populated object graph, it should be able to persist it without constraint issues regardless of the order your call your persist, merge and delete object in).

-Tom

On 17/01/2013 8:45 AM, Deyan Tsvetanov wrote:
I see your point guys but it won't work for me.

So EclipseLink tries to determine the best order of the database operations. That is OK.
But if it can't determine an order it executes them in a RANDOM order instead of the NATURAL ( set by developer ) order ?
That sounds spooky :)



On Jan 17, 2013, at 15:43 , Christopher Delahunt <christopher.delahunt@xxxxxxxxxx> wrote:

I believe Tom is stating that you are not setting the Role reference (parentRole is null when you insert).  If you set that, then EclipseLink will see the dependencies and order the inserts appropriately.

EclipseLink has a cache, so it should simply be a mater of calling newRole.setParentRole(em.find(Role.class, newRole.getParentId()));

Best Regards,
Chris

On 17/01/2013 8:34 AM, Deyan Tsvetanov wrote:
Oops :)

Wrong copy+paste.  Please consider RoleType as Role. Role has a reference to another parent Role.

I do have reasons indeed :)
But my mapping strategy is not the key point here.

You are free to modify the example and make it your way. It will still fail for the same reason:)

Regards,
Deyan


------------------------------------------

@Entity @Table(name="ROLE")
class Role {

    @Id @Column(name="ID", length=20, nullable=false)
    String id

    @Column(name="PARENT_ID", length=20, nullable=true)
    String parentId

    @Column(name="NAME", length=255, nullable=false)
    String name

    @ManyToOne
    @JoinColumn( name="PARENT_ID", referencedColumnName="ID", updatable=false, insertable=false)
    Role parentRole

}
On Jan 17, 2013, at 15:29 , Tom Ware<tom.ware@xxxxxxxxxx>  wrote:

How is RoleType mapped?

Is there a particular reason you have made the parentId mapping writable and the parentRole mapping readOnly, or are these mappings just for your import tool? If I were to map these entities, I would consider the opposite...  That would allow me to interact with the entities in a way that was much more Object-Relational, and I think in a way that would make using your mappings for other applications more practical. I'd write code more like this:

beginTransaction()
for (XMLRoleWrapper xmlRoleWrapper  xmlResults){
	Role role = new Role();
	em.persist(role);
	role.setId(xmlRoleWrapper.id);
	role.setName(xmlROleWrapper.name);
	role.setParentRole(em.find(RoleType.class, xmlRoleWrapper.parentId));
}
commitTransaction();

The problem with ordering operations exactly as the user writes them, is that it works well for very simple cases, but as soon as you add merging, cascading and the possibility of cycles it breaks down pretty quickly.

-Tom


On 17/01/2013 8:12 AM, Deyan Tsvetanov wrote:
Hi Tom,

here is a more specific test case:

@Entity @Table(name="ROLE")
class Role {

     @Id @Column(name="ID", length=20, nullable=false)
     String id

     @Column(name="PARENT_ID", length=20, nullable=true)
     String parentId

     @Column(name="NAME", length=255, nullable=false)
     String name

     @ManyToOne
     @JoinColumn( name="PARENT_ID", referencedColumnName="ID", updatable=false, insertable=false)
     RoleType parentRole

}

I have created a tool that imports data from XML file. In this case the XML would look like:

<data>
     <Role  id="MAIN" name="Main Role" />
     <Role  id="USER" name="User"  parent_id="MAIN"/>
     <Role id="ADMINISTRATOR" name="Administrator" parent_id="MAIN" />
</data>

The MAIN role has NULL in PARENT_ID. The rest of the records have MAIN as a parent role.
The order of the insert statements is vital for the successful import of the data.

I have implemented a XML parser that creates new entities and persists them in the order they are coming from the XML file.

HOWEVER

Hibernate inserts the data in the correct order - e.g. MAIN, USER, ADMINISTRATOR.

EclipseLink however would execute the insert statements in a random order every time, e.g.  USER, MAIN, ... which causes a FK error.

Flushing the session on each persist is not a solution.

EclipseLink needs to execute the database statements in the same order as the persist() and merge() calls.

Regards,
Deyan



On Jan 17, 2013, at 15:00 , Tom Ware<tom.ware@xxxxxxxxxx>  wrote:

Can you provide a specific test-case where the order actually causes a problem?

The JPA specification allows a persistence provider to order statements in the way that is most efficient for it.  This, in fact, is essential to solving the very problem you are worried about.  i.e. A JPA provider has to be able to compute an order of statements that allow foreign key constraints to be properly satisfied.  As long as your entities are configured so that JPA is made aware of all your constraints, the order of statements should not cause any problems of that type.

The power of this functionality is that when you are writing your JPA code, you do not have to think about the database constraints, you just manipulate your objects and the order will be taken care of.

-Tom

On 17/01/2013 6:27 AM, Deyan Tsvetanov wrote:
Well, the order that EclipseLink chooses is RANDOM :)
It is a fairly simple example, only 1 entity with only 1 column.
I'd really expect that the INSERT statements are executed in same order as the
persist() calls.

My real-life use case is importing data from an XML file into the database.
There are relations and FKs in my database and when exported and re-imported the
order of the XML entries, persist() calls and INSERT statements is critical.
EclipseLink basically inserts each XML entry randomly. Currently the only
workaround is to flush after each persist call. It could work for few hundred
calls,
but not for few thousand.

JPA says nothing about the order of the database operations. When writing the
spec they probably have assumed that it would be logical to execute the database
operations in the
same order as the persist() or merge() calls.
This is not the case of mixed remove(), persist() and merge() calls, in our case
we have only persist() calls and the case is very simple.

Best regards,
Deyan


On Jan 17, 2013, at 13:19 , Wim Bervoets<wbervoets@xxxxxxxxx
<mailto:wbervoets@xxxxxxxxx>>  wrote:

If you want to know the order in which the rows were inserted I use @OrderColumn
(eg. in combinantion with INDEX(..) function in a JPQL for example).

I think that EclipseLink can choose the order in which it commits the entities
to the database... (I haven't read the JPA spec so this is an assumption)

Wim


On Thu, Jan 17, 2013 at 12:09 PM, Deyan Tsvetanov<deyan@xxxxxxxxxxxx
<mailto:deyan@xxxxxxxxxxxx>>  wrote:

    Hibernate persists the entities in the correct order:

    2:57:58,867 TRACE TypeFactory:72 - Scoping types to session factory
    org.hibernate.internal.SessionFactoryImpl@395fa2b5
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,365 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST0
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,369 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST1
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,371 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST2
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,372 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST3
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,373 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST4
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,375 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST5
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,376 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST6
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,377 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST7
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,378 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST8
    Hibernate: insert into ROLE (ID) values (?)
    12:57:59,380 TRACE BasicBinder:83 - binding parameter [1] as [VARCHAR] - TEST9



    On Jan 17, 2013, at 12:19 , Deyan Tsvetanov<deyan@xxxxxxxxxxxx
    <mailto:deyan@xxxxxxxxxxxx>>  wrote:

    Hi guys,

    I am experiencing a weird imho behaviour of Eclipselink and I'd really like
    to hear some other opinions .

    I have a pretty simple entity with assigned IDs;

    @Entity@Table(name="ROLE")
    public class Role implements Serializable {

    privatestaticfinallongserialVersionUID= 1L;

    @Id @Column(name="ID", length=20, nullable=false)
    public String id;


    }


    I am executing the following operations:

    public static void main(String[] args) {
    EntityManagerFactory emf =
    Persistence.createEntityManagerFactory("EclipseLinkJPATest");

    EntityManager em = emf.createEntityManager();

    em.getTransaction().begin();
    for (int i = 0; i<  10; i++) {
    Role r = new Role();
    r.id = "TEST" + i;
    em.persist(r);
    }
    em.getTransaction().commit();

    em.close();
    emf.close();
    }

    And I'd expect that the INSERT queries will be executed in the same order as
    the persist() method is called:
    TEST0, TEST1, TEST2 , etc.

    But in the real life the insert queries are in a random order every time:

    EL Fine]: sql: 2013-01-17
    12:13:58.11--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST1]
    [EL Fine]: sql: 2013-01-17
    12:13:58.113--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST6]
    [EL Fine]: sql: 2013-01-17
    12:13:58.114--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST2]
    [EL Fine]: sql: 2013-01-17
    12:13:58.115--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST7]
    [EL Fine]: sql: 2013-01-17
    12:13:58.117--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST4]
    [EL Fine]: sql: 2013-01-17
    12:13:58.121--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST8]
    [EL Fine]: sql: 2013-01-17
    12:13:58.123--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST3]
    [EL Fine]: sql: 2013-01-17
    12:13:58.124--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST9]
    [EL Fine]: sql: 2013-01-17
    12:13:58.126--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST5]
    [EL Fine]: sql: 2013-01-17
    12:13:58.127--ClientSession(1694665796)--Connection(1795160456)--INSERT INTO
    ROLE (ID) VALUES (?)
    bind =>  [TEST0]

    As you can see the order if insert queries is:
    TEST1, TEST6, TEST2, TEST7, TEST4, etc.


    That is really weird and wrong ! :)
    I dug a lot and could not find a solution.

    Please help :)

    Thanks in advance,
    Deyan



    _______________________________________________
    eclipselink-users mailing list
    eclipselink-users@xxxxxxxxxxx<mailto:eclipselink-users@xxxxxxxxxxx>
    https://dev.eclipse.org/mailman/listinfo/eclipselink-users


_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx<mailto:eclipselink-users@xxxxxxxxxxx>
https://dev.eclipse.org/mailman/listinfo/eclipselink-users



_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users

_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users

_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users

_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users

_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users
_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users

_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users

_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users

_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipselink-users



Back to the top