NHibernate produces multiple queries on deletion of object in child collection - nhibernate

I have an object hierarchy as follows:
Object
Reports
Report Items
Based on other posts here, I set the Report.ReportItems mapping to Inverse, and CascadeAllDeleteOrphan. This makes it so when I remove a Report from an Object it goes to erase the ReportItems in that report.
The way it does it is very inefficient. It basically does a Delete ... Where ReportItemID = ?.
I've seen suggestions to increase the batch size to prevent this from being done in too many round trips, but that seems like a sloppy fix. Is there a way to make NHibernate generate a query like this instead:
Delete ... Where ReportID = ?
This way it would execute one query that would delete all the ReportItems instead of one statement per ReportItem.
Thanks in advance.
Edit
I have heard from a few people that NHibernate will simply not work this way with standard QueryOver etc.
I decided to start using HQL to fix the problem. I will later use some reflection to make sure there are no "magic strings" being used.
I had the idea to:
Use HQL to mass-delete the Report Items
Tell the Session to Refresh the object so it could detect that the Report Items are gone
Then tell the session to erase the Report, and let it clean up the remaining information
This does not work though. You can see the code below:
Session.CreateQuery("delete ReportItem r where r.Report= :report").SetEntity("report", SelectedReport).ExecuteUpdate()
Session.Refresh(SelectedReport)
Object.Reports.Remove(SelectedReport)
Session.Delete(SelectedReport)
Session.Update(Object)
I have also tried doing a Session.Evict after performing the HQL statements, but NHibernate keeps using the cache to try to delete the ReportItems. Any tips on how to do this?

If your foreign key has delete rule set to cascade then you can tell NH not to generate DELETE statements for child items when the parent is deleted. In this case NH will generate DELETE statement only for parent entity.
<bag name="ReportItems" cascade="all-delete-orphan" inverse="true">
<key column="report_id" on-delete="cascade" />
<one-to-many class="ReportItem" />
</bag>

After a lot of research I have found that this is simply not possible. You can use HQL, raw SQL statements, or rely on database rules to perform the deletes by enforcing a relationship. However, there is no way to tell NHibernate to delete a list of objects based on their owner's ID.

Related

NHibernate prevent cascade delete

Suppose I have a class Foo. I also have a view on Foo called Foo_Foo that lists a many-to-many association between Foos. I mapped this association as a simple immutable set on each Foo, with cascade="none":
<set name="association" table="Foo_Foo" cascade="none" mutable="false">
<key column="ParentFoo" />
<many-to-many class="Foo, MyAssembly" column="BaseFoo" />
</set>
However, when I try to delete a Foo, NHibernate tries and rightly fails to delete the Foo.association.
How can I prevent NHibernate from trying to delete the association to a view?
The collection belongs to Foo. You can't share the collection, so there is no need to keep it in the database. Cascade is used to tell NH if the referenced Foos should be also deleted or not.
Why do you want the Foo_Foo records to keep in the database? If this should be a bidirectional many-to-many self reference, it doesn't work like this.
Edit, after understanding the question.
Cascade doesn't work in your case, because it affects only the referenced Foos.
To avoid inserts / updates and deletes of the collection table, you may try one of the following:
First obvious attempt is mutable="false", which you already tried. I don't really understand why it isn't working. You may ask in the Nhibernate user group.
Less obvious, but promising is inverse="true". Inverse tells NH that the collection is mapped somewhere else and doesn't need to be stored from here. So it just omits inserts, but I don't know about deletes.
If this doesn't work, you need to explore more complex solutions. You could map it as a one-to-many of an intermediate entity which references the Foos. The intermediate entity is a mapping to the view. It is immutable (which still may lead to delete statements). In this case, cascade="false" will work (because it is the referenced entity). It will also work configure insert, update and delete sql statements (which are empty), but this is most probably not even necessary.

nhibernate intercept select query

I'm looking at the nhibernate interceptor. It seems to be able to intercept save, update and delete queries but is there anyway I can intercept a select query.
The problem I have is that I automatically want to append some additional sql filters to the executing sql statement in certain cases.
Any thoughts
Thanks
Mat
The interceptor framework only allows you to intercept entity operations like save, update, and delete. Queries themselves aren't intercepted, only entity operations (and selecting/viewing doesn't count).
Consider using some sort of inheritance for a set of query classes that append to your ICriteria to suit your purposes.
I think you can accomplish that with a custom NHibernate persister; however, I'm having trouble finding examples on how you'd write one...
You can use the where attribute on the class tag:
<class name="MyClass" where="deleted=0">
...
</class>

Override delete behaviour in NHibernate

In my application users cannot truly delete records. Rather, the record's Deleted field gets set to 1, which hides it from selects.
I need to maintain this behaviour and I'm looking into whether NHibernate is appropriate for my app. Can I override NHibnernate's delete behaviour so that instead of issuing DELETE statements, it issues UPDATES, as described above?
I would obviously also need to override its SELECT behaviour to include the 'AND Deleted = 0' clause. Or read from a view instead. I'm not sure.
To implement a soft delete just bypass the Hibernate delete mechanism. Instead, map your table's Deleted field to a .Net boolean property by the same name. To delete an item, set item.Deleted = true. Then add a where attribute to your class mapping to filter out the deleted items. If you like, create another mapping for deleted items. Otherwise they will become invisible to your application, but maybe that's what you want.
Edit: Here is perhaps a better approach: use the <sql-delete> tag to write a custom delete operation for your mapping. See http://docs.jboss.org/hibernate/core/3.3/reference/en/html/querysql.html#querysql-cud. I think this in combo with the where attribute would be just the ticket. For example:
<class name="MyClass" table="my_table" where="deleted=0">
...
<sql-delete>UPDATE my_table SET deleted=1 WHERE id=?</sql-delete>
</class>
I implemented this using INSTEAD OF DELETE triggers on SQL Server to set a delete flag bit when a SQL DELETE is issued. This has two benefits: 1) I can just issue deletes from my app. without worry and 2) It enforces soft deletes for all database access (i.e. the trigger must be temporarily disabled to hard delete). I then set up views that select the active records only and mapped NHibernate to those. This solution has worked very well for me.
I think the best way to get such behaviour would be by implementing the IInterceptor interface which would allow you to perform your own code as shown within the NHibernate Documentation.
Otherwise, you could simply create a trigger on delete that would perform an update. This solution is simpler, but is this suitable for your needs?
As for the SELECT, you only need to write method that will use Criterion with a Where clause to specify the Deleted=0 thing.

Single class maps to two table join

looking at an existing NHibernate implementation that maps a single class to two database tables that are joined. The current functionality is read-only. The table join is actually hidden away via a readonly view and it's the view that's referred to in the NHibernate mapping. Works fine for readonly behaviour. Except I need to add Insert, Update, Delete behaviour and only one of the tables needs to be inserted/updated/deleted. How best to do this? I can take readonly off the view of course, I could replicate the join in the NHibernate mapping, but how do I tell NHibernate that insert/update/delete affects one table only?
Thanks DC
You can set the update and insert attributes on the property mappings to false and they will be excluded from updates and inserts:
<property name="MyProperty" update="false" insert="false" />
Keep in mind that your view will need to have a single primary key defined to enable updates or inserts.
You can map to a view instead of a table, then define custom sql for your inserts, updates and deletes.
I would recommend calling a stored procedure.
See the docs: http://nhibernate.info/doc/nh/en/index.html#querysql-cud
I've used this technique quite a lot and it works fine.

Nhibernate many-to-one problem

I've got a problem with mapping many-to-one in the following code:
...
<property name ="CustomerID"/>
<!-- Many-to-one mapping: Customer -->
<many-to-one name="Customer"
class="Customer"
column="CustomerID"
insert="false" update="false"/>
<property name="Date" />
...
You may notice that i have mapped two CustomerID to Customer table. The reason i do this because when i want to create an Order, i just only assign value to CustomerID and other require fields then do save. many-to-one mapping, I just want to get detail of each customerID.
But, the problem is that: after i update customerID of an Order and Executre SaveOrUpdate with Session.Flush() also (I'm using HibernateTemplate), i still got the old figure when accessing to Order.Customer.
i.e:
Order = getOderByID(1);
Order.CustomerID=3 // Suppose value of CustomerId is 1. Now I changed to 3
SaveOrUpdate(Order);
Print(Order.Customer.CustomerID)// it returns 1 which is wrong. It should be 3
Pls help...
Thanks,
I would suggest you to look at this problem from an NHibernate point of view. And not from a relational database view. Let me start with what i feel you should be doing.
var customer = session.Load<Customer>(3);
var order = session.Load<Order>(1);
order.Customer = customer;
//assuming this is a one directional mapping. otherwise you might
//have to do some more steps to disassociate the order from the old
//customers collection and add it to the new customers collection
session.SaveOrUpdate(order);
Now, order.Customer.CustomerID will return 3.
As Serkan suggested, its better and more feasible to work with objects instead of primary keys.
Also, there really shouldnt be any performance impact here. Nhibernate is able to proxy a lot of the associations as long as the classes have virtual public methods. Because of this, as long as you only query for the Id of the customer, it will not generate a separate sql query. The Id is already there with the proxy object.
With regards to the original question, I have a hunch. NHibernate dynamically generates the sql query for the update and the inserts. This case here is of an update. You have explicitly set the CustomerID property to 3. But the Customer property of the order object still points to the customer object with Id 1. So, when NHibernate generates the sql query, it trys to set the value first to 1, as you asked it to. Then it also sees that the Customer is still pointing to the old object, so reset the CustomerId property to 1. I think NHibernate is getting confused with the dual mappings.
There are two things that you can do. First enable the "show_sql" property in the NHibernate configuration.
<nhibernate>
...
<add key="hibernate.show_sql" value="true" />
</nhibernate>
Check what is the sql being generated when you save the order. That will explain things better.
Second, after saving the order, do session.Refresh(order); You can read about the Refresh() method here. Its towards the end of the section 9.2. It will reload the order object with fresh values from the database. Calling order.CustomerID should show what value you have stored in the database.
I think you'd feel much happier in the long run if you try to forget about Id's of entities once you are done with the OR mapping. You are in a different level which you should think in objects only. If I were you I would remove CustomerId property all together.
If you have performance issues try to solve them in NHibernate way, caching etc.
Two things to try
Flushing the session via Session.Dispose if it's NH 2.x otherwise use Flush()
Make your IDs client assigned.
NHibernate will create the IDs for you and try to manage them unless you specifically tell it not to.