I'm moving a project off of a custom DAL and onto Nhibernate but I've run into a tricky mapping scenario that I don't know how to resolve. Essentially, there are sub classes of a subclass without a discriminator value.
The primary subclass has a discriminator value so that was trivial. The issue arises when I get to the third level. The only difference between the sub-sub types is which table references them (in a HasOne style map).
Any help would be greatly appreciated, thanks :)
If you want the third-level objects to be records in the same table, I'd add a discriminator to the third-level type. It can be a calculated property based on the parent type. Using FluentNH, I think it does have to be visible to the mapper (you can't use the tricks that vanilla NHibernate has to reflectively examine private members).
That's if you want to have them all in the same table. If you don't care too much about table count in your schema, you simply cannot include a discriminator, and/or the schema definitions of the third-level types are different enough that mapping to a single table would provide "odd" results (like a lot of null columns or very general/nondescriptive column names), then simply create a mapping for each concrete subtype to a different table.
You can also de-normalize the schema. Since it's a HasOne() relationship, the subclass can contain the columns of the third-level class using a Component mapping. If all of the third-level types map to similar fields, and you think it'll be unlikely that this relationship will change from 1:1 to 1:n, this is probably the way to go.
Related
I know that the underlying ORM used in Orchard is NHibernate and it does support the so-called ClassMapping which may help customize the mappings the way we want.
However I'm not sure about how Orchard utilizes the mapping methods supported by NHibernate. In this case it seems to always use the strategy similar to Table Per Type in EF as well as some other ORMs. With that strategy, the base class will be mapped to some common table whereas the derived class will be mapped to another table which contains all properties of its own (not declared in the base class). Those 2 tables will have a one-one relationship.
Now I really want to make it use the strategy similar to Table Per Concrete Type in which the base and derived classes will be mapped to 2 different tables with all properties (including inherited properties) being mapped to columns. Those 2 tables will not have any relationship, so querying for columns in just one table will not unexpectedly generate an inner JOIN (for one-one relationship).
Actually that requirement makes sense in case we just need to partition our data (from 1 big table to 2 or more small tables that have the same schema). We don't want to re-declare or use some kind of duplicate model classes (with different names), instead we just need to create a new model class and let it inherit from one base model class containing all necessary properties.
With the current code like this:
public class ARecord {
//properties ...
}
public class BRecord : ARecord {
//empty here
}
Currently we cannot use BRecord because it is understood as another part of the ARecord, the auto-generated query (always with INNER JOIN) will be failed because of some does-not-exist table or column names.
How can I solve this?
You're not going to like it ;) In a nutshell, the answer is don't do inheritance at all. Orchard was very deliberately designed around the idea of composition, steering well clear of inheritance in its content models. Maybe the central idea in Orchard is to make the concept of content part the "atom of content", and to design those basic units as very simple and composable pieces of functionality that do one thing well.
After all these years, the concept has held remarkably well, and I've yet to see an example of a content model where inheritance would have been more elegant and appropriate. This is reflected in the way nHibernate is customized and used in Orchard, as you've discovered.
So the solution to your problem can likely be one of two things:
You're modeling contents, and you should re-think your approach towards composition of parts. If you give a little more details about your specific scenario (maybe in a new question), I'm happy to help specifically in this direction.
You're modeling non-content data, in which case you might want to consider opting out of Orchard's specific nHibernate content-specialized idiosyncrasies and do things closer to the metal. Again, if you give more specifics about your scenario, I'm happy to take a look and give some pointers.
I have a frustrating situation owing to this little quirk of EF. Here's a simple demo of the behavior. First the DB schema:
As you see, RestrictedProduct is a special case of product, which I'm intending to make a subclass of Product with some special code.
Now I import to an EF data model:
Oops! EF saw that RestrictedProduct had only 2 fields, both FKs, so it mapped it as a one-to-many relationship between Product and Restriction. So I go back to the database and add a Dummy field to RestrictedProduct, and now my EF model looks much better:
But that Dummy field is silly and pointless. Maybe I could delete it? I blow away the field from the DB table and the entity model, then refresh the model from the DB...
Oh, no! The Product-Restriction association is back, under a new name (RestrictedProduct1)! Plus, it won't compile:
Error 3034: Problem in mapping fragments starting at lines (x, y) :Two entities with possibly different keys are mapped to the same row. Ensure these two mapping fragments map both ends of the AssociationSet to the corresponding columns.
Is there any way to prevent this behavior, short of keeping the Dummy field on the RestrictedProduct table?
I just came across the same issue, and as an alternative to putting the dummy field in your RestrictedProduct table to force the creation of an entity you can also make your RestrictedProduct.RestrictionId field nullable and EF will then generate an entity for it. You can then modify it to use inheritance and any subsequent "Update model from database" will not cause undesired nav properties. Not really a nice solution but a work around.
Let's walk slowly into your problem.
1st thing you need to decide is if the restricted product is
really a special case of product or is it a possible extension
to each product.
From your original DB Scheme it seems that any product may have
a relation to a single restriction however a single restriction
can be shared among many products.. so this is a simple 1 to many
situation which means that restricted product is NOT a special case
of product! Restriction is an independent entity which has nothing
to do with product in a specific way.
Therefore EF is correct in the 1st importation of your scheme:
1. a product can have 0 or 1 restrictions.
2. a restriction is another entity which can be related to many products.
I do not see your problem.
I am trying to come up with a database design that would work with Entity Framework 4 Code First. Actually, I have no experience yet of EF4 Code First but as I understand it, if I write the code, it will create the database and tables.
The issue is this. There are various types of auctions, they all have some common fields and some specific ones. In the code I envisage having a base abstract class called Auction and subclasses like LowestUniqueBidAuction and EnglishForwardAuction etc.
Nothing surprising there. The problem is that I imagine the database structure to mimic this. I imagine an Auction table and a LowestUniqueBidAuction table and a EnglishForwardAuction table. In the Auction table I imagine a foreign key into one of these two tables for each row depending on the type of auction that that row is. I also imagine another column in the Auction table with the name of the derived auction table (such as EnglishForwardAuction).
The problem is that whenever I've ever created a foreign key I've had to specify the name of the foreign table into which the key points (which makes sense). In this case, however, there is one of many tables that the key could point. So there are many issues here. Firstly, I could simply not use a foreign key and just use an ordinary field, but then the database will not be able to maintain data consistency for me. The second issue is how will EF Code First handle this? In other words, how will it know that if I ask for all EnglishForwardAuction rows from the Auction table that it should look at the column with the table name and then join on the EnglishForwardAuction table to get the extra fields?
Has anyone ever faced similar issues?
Thanks,
Sachin
This problem is solvable in Entity Framework in a number of ways - read up on how EF handles inheritance and what strategies are available.
There are basically three strategies how to handle this:
(1) Table per Hierarchy
You have only one single table, that represents all possible sub classes. Of course, this means, several rows (that only exist in a given subclass) must be nullable, since they don't show up / don't exist in super classes or other subclasses.
(2) Table per Type
Each subclass gets its own table, and by default, the sub-types table shares the PK with the base classes' table - e.g. PK = 1 in Auction will also be PK = 1 in EnglishForwardAuction. So your subclass tables reference the base table - not the other way around.
(3) Table per Concrete Type
Each concrete subclass (your separate auction types) gets its own table, but that table contains everything - all the columns, from that specific type, but also its base type.
Read more here:
Inheritance in the Entity Framework
Inheritance and Associations with Entity Framework Part 1
Entity Framework Modeling: Table Per Hierarchy Inheritance
Entity Framework Modeling: Table Per Type Inheritance
Searching for Entity Framework Inheritance and/or one of these strategies will reveal a lot more hits, too - that topic is very well covered and discussed on the interwebs! :-)
I have a table that, some of its columns are unknown at compile time. Such columns could either be of an integer value, or some Enum value. There is a table that holds all the names of such dynamic columns and also holds the column's type. This "metatable" has the following columns:
DynamicColumnId (Pk)
Name
TypeId (Integer / Enum, as Fk from a separate table)
Integer columns have the Name from this table, whereas Enum columns are Fk columns from a table that has that Name, with some modification (e.g. a "DynamicTable" prefix).
The only solution I could think of for this situation is using Reflection.Emit to dynamically create an Entity class and a corresponding Mapping class. Admittedly, I'm new to NHybernate / Fluent NHybernate and it seems like a relatively simple hierarchy between the tables, and so I wanted to verify my solution isn't as ugly as it initially appears...
I would also welcome solutions that completely disregard my table hierarchy, in order to effectively acheive the same results (that is, to enumerate the rows on the dynamic table, going over all the columns, with knowledge of whether they are Enums and, if they are, their possible values as well).
(Edit: Additional information re problem domain)
I initially included minimal details, as to avoid Too-Much-Info related confusion.
This description is much more complex, but it unravels the motives behind this design.
The application involved is designed to automate log/dump analysis. Analysis-scenarios are frequently provided by the log/dump experts and so, in order to streamline the typical process of requirements=>implementation=>verification cycle, such analysis-scenarios are implemented by the experts directly as an Iron Python code snippet, with some domain-specific constructs injected into the snippets' scope. Each snippet has a "context" for which it is relevant. An example of "context" could be "product," "version," etc... So, the snippet itself is only invoked in certain contexts - this helps simplifying the Python code by eliminating branching (you could view it as Aspect Oriented Programming, to some extent). A non-expert could use the application, with a given code-context database, to analyze a log/dump, after choosing values for the various contexts.
When an expert decides that a new context is required for cataloging a certain code snippet, he could add a context, indicating the possible values it could have. Once a new context is added to the database, a non-expert that runs an analysis will be given the option to choose a value for the newly-added context.
The "dynamic table" is the table that associates a code snippet with values of the various contexts (columns) that existed when the snippet was issued, plus default values for the columns that did not exist at that time.
I won't claim to fully understand your scenario, but it seems to me that you'd be better off using a key-value store such as Redis or a schema-less database like CouchDB instead of SQL. This doesn't seem to be a problem for a relational database, but if you really need to use a RDBMS I'd map NHibernate as closely as possible to the real schema (DynamicColumnId, Name, TypeId) then build whatever data structure you need on top of that.
I'm thinking of building a ecommerce application with an extensible data model using NHibernate and Fluent NHibernate. By having an extensible data model, I have the ability to define a Product entity, and allow a user in the application to extend it with new fields/properties with different data types including custom data types.
Example:
Product can have an addition fields like:
Size - int
Color - string
Price - decimal
Collection of ColoredImage - name, image (e.g. "Red", red.jpg (binary file))
An additional requirement is to be able to filter the products by these additional/extended fields. How should I implement this?
Thanks in advance.
I think this link describes kind of what you want...
http://ayende.com/Blog/archive/2009/04/11/nhibernate-mapping-ltdynamic-componentgt.aspx
More info on dynamic-component:
http://www.mattfreeman.co.uk/2009/01/nhibernate-mapping-with-dynamic-component/
http://bartreyserhove.blogspot.com/2008/02/dynamic-domain-mode-using-nhibernate.html
The idea behind dynamic-component is that you can build your data model by not having a one to one mapping of databse columns with properties. Instead you have only a dictionary property that can contain data from as many properties as you like. This way when you fetch the entity, the dictionary gets the data of all columns configured to belong in there. You can extend the database table's schema to include more columns and that will be reflected to the databse model if you update the mapping file accordingly (manually or though code at application start).
To be honest I do not know you can query such entity using the "attributes" property but if I had to guess I would do an IN statement to it.
One of the options is EAV model (Entity-Attribute-Value).
This model is good to apply if you have a single class in your domain, which table representation would result in a wide table (large number of columns, many null values)
It's originally designed for medical domain, where objects may have thousands of columns (sympthoms).
Basically you have
Entity (Id) (for example your Product table)
Attribute(Id, ColumnName)
Value(EntityId, AttributeId, value)
You can have some additional metadata tables.
Value should better be multiple tables, one for a type.
For example:
ShortStringValue(EntityId, AttributeId, Value nvarchar(50));
LongStringValue(EntityId, AttributeId, Value nvarchar(2048));
MemoValue(EntityId, AttributeId, Value nvarchar(max));
IntValue(EntityId, AttributeId, Value int);
or even a comple type:
ColorComponentsValue(EntityId, AttributeId, R int, G int, B int );
One of the things from my experience is that you should not have EAV for everything. Just have EAV for a single class, Product for example.
If you have to use extensibility for different base classes, let it be a separate set of EAV tables.
Onother thing is that you have to invent a smart materialization strategy for your objects.
Do not pivot these values to a wide row set, pivot just a small number of collumns for your query criteria needs, then return a narrow collection of Value rows for each of the selected objects. Otherwise pivoting would involve massive join.
There are some points to consider:
. Each value takes storage space for foreign keys
. For example row-level locking will behave different for such queries, which may result in performance degradation.
. May result in larger index sizes.
Actually in a shallow hellow world test my EAV solution outperformed it's static counterpart on a 20 column table in a query with 4 columns involved in criteria.
Possible option would be to store all extra fields in an XML structure and use XPath/XQuery to retrieve them from the database.
Each extensible entity in your application will have an XML field, like ExtendedData, which will contain all extra properties.
Another option is to use Non-relationnal Databases which are typically suited for this kind of things.
NOSQL databases(couchDB, mongoDB, cassandre...) let you define dynamically your propretyfields, you could add fields to your product class whenever you want.
I'm searching for similar thing and just found N2 CMS (http://n2cms.com) which implements domain extensibility in quite usable way. It also supports querying over extension fields which is important. The only downside I find out is that it's implemented using HQL so it would take some time to reimplement it to be able to query using QueryOver/Linq, but the main idea and mappings are there. Take a look on ContentItem, DetailCollection, ContentDetail classes, their mappings and QueryBuilder/DetailCriteria.