According to the Hibernate docs, MappedSuperclass allows mapping inheritance, where the super class is not treated as an entity, and where polymorphic queries that fetch objects by the base class are not supported.
On the other hand, the table per class strategy is similar in terms of the generated schema in the DB, except that it maps the super class to a table if it is not an abstract. However it supports polymorphic queries.
My question is: why would anyone use the #MappedSuperclass strategy? Is there a performance issue implied if the parent class is treated as an entity? Any ideas around this is appreciated.
#MappedSuperclass
By default, properties of the superclass are ignored and not persistent!
You have to annotate the superclass with #MappedSuperclass to enable embedding of its properties in the concrete subclass tables.
You can override column mappings from the superclass in a subclass with the #AttributeOverride annotation or several with #AttributeOverrides.
You could declare the identifier property in the superclass, with a shared column name and generator strategy for all subclasses, so you wouldn’t have to repeat it. But it’s optional!
The main problem with implicit inheritance mapping is that it doesn’t support polymorphic associations very well. In the database, you usually represent associations as foreign key relationships. If the subclasses are all mapped to different tables, a polymorphic association to their superclass can’t be represented as a simple foreign key relationship.
Polymorphic queries that return instances of all classes that match the interface of the queried class are also problematic. Hibernate must execute a query against the superclass as several SQL SELECTs, one for each concrete subclass.
A further conceptual problem with this mapping strategy is that several different columns, of different tables, share exactly the same semantics. This makes schema evolution more complex. For example, renaming or changing the type of a superclass property results in changes to multiple columns in multiple tables. Many of the standard refactoring operations offered by your IDE would require manual adjustments, because the automatic procedures usually don’t account for things like #AttributeOverrides. It also makes it much more difficult to implement database integrity constraints that apply to all subclasses.
So, this approach is a good chose for the top level of your class hierarchy,
where polymorphism isn’t usually required, and when modification of the superclass
in the future is unlikely.
#Inheritance(strategy = InheritanceType.TABLE_PER_CLASS)
The database identifier and its mapping have to be present in the superclass, to share it in all subclasses and their tables. This is no longer optional, as it was for the #MappedSuperclass mapping strategy.
Note that the JPA standard specifies that TABLE_PER_CLASS is optional, so not all JPA implementations may support it. The implementation is also vendor dependent — in Hibernate, it’s equivalent to a <union-subclass> mapping in the old native Hibernate XML metadata.
The advantages of this mapping strategy are clearer if we examine polymorphic queries. The tables are combined with a UNION operator, and a literal is inserted into the intermediate result; Hibernate reads this to instantiate the correct class given the data from a particular row. A union requires that the queries that are combined project over the same columns; hence, you have to pad and fill nonexistent columns with NULL. You may ask whether this query will really perform better than two separate statements. Here you can let the database optimizer find the best execution plan to combine rows from several tables, instead of merging two result sets in memory as Hibernate’s polymorphic loader engine would do.
Another much more important advantage is the ability to handle polymorphic associations. Hibernate can use a UNION query to simulate a single table as the target of the association mapping.
So, this approach is a good chose when you need to use polymorphic queries and associations.
P.S. This is almost verbatim quotation from the excellent book: Java Persistence with Hibernate (Bauer, King, Gregory)
meaning a class whose mapping information is applied to the entities that inherit it.
Related
Short version for the hasty:
There's various tables/entities in my domain model which have the same field (a UUID). There is a table where I need to link rows/instances of such entities to other JPA-managed entities. In other words, the instance of the field in that link table won't be known up-front. The two approaches I can think of are:
Use an abstract entity and a TABLE_PER_CLASS strategy, or
use an #MappedSuperClass store the class name of the instance in the link table as well, or something similar that lets me define logic for getting the actual instance from the right table.
Both have advantages and disadvantages in terms of complexity and performance. Which do you believe to be best, is there maybe a third option, or have you tried something like this in the past and would advice/strongly warn against?
Long version in case you want more background:
I have a database/object model wherein many types have a common field: a universally unique identifier (UUID). The reason for this is that instances of these types can be subject to changes. The changes follow the command model and their data can be encapsulated and itself persisted. Let's call such a change a "mutation". It must be possible to find out which mutations exist in the database for any given entity, and vice-versa, on which entity a stored mutation operates.
Take the following entities with UUIDs as an (extremely simplified) example:
To store the "mutations", we use a table/entity called MutationHolder. To link a mutation to its target entity, there's a MutationEntityLink. The only reason this data isn't directly on the MutationHolder is because there can be direct or indirect links, but that's of little importance here so I left it out:
The question comes down to how I can model the entity field in MutationEntityLink. There are two approaches I can think of.
The first is to make an abstract #Entity annotated class with the UUID field. Customer, Contract and Address would extend it. So it is a TABLE_PER_CLASS strategy. I assume that I could use this as a type for the entity field, although I'm not certain. However, I fear this might have a serious performance penalty since JPA would need to query many tables to find the actual instance.
The second is to simply use #MappedSuperClass and just store the UUID for an entity in the entity field of MutationEntityLink. In order to get the actual entity with that UUID, I'd have to solve it programmatically. Adding an additional column with the class name of the entity, or something else that allows me to identify it or paste it in a JPQL query would do. This requires more work but seems more efficient. I'm not averse to coding some utility classes or doing some reflection/custom annotation work if needed.
My question is which of these approaches seems best? Alternatively, you might have a better suggestion, or notice I'm missing something; for example, maybe there's a way to add a type column even with TABLE_PER_CLASS inheritance to point JPA to the right table? Perhaps you've tried something like this and want to warn me about numerous issues that would arise.
Some additional info:
We create the database schema, so we can add whatever we want.
A single table inheritance strategy isn't an option. The tables must remain distinct. For the same reason, joined inheritance doesn't seem a good fit either.
The JPA provider is Hibernate and using things that are not part of the JPA standard isn't an issue.
If the entities don't have anything in common besides having a uuid I'd use the second approach you describe: use MappedSuperclass. Making the common superclass an entity would prevent you to use a different inheritance strategy if needed, would require a table for that super entity even if no instances exist and from a business point of view it's just wrong.
The link itself could be implemented in multiple ways, e.g. you could subclass MutationEntityLink for each entity to map (e.g. CustomerMutationEntityLink etc.) or do as you described it, i.e. only store the uuid as well as some discriminator/type information and resolve programatically (we're using that approach for something similar btw.).
You need to use #MappedSuperclass while inheriting associations/methods/properties whereas TABLE_PER_CLASS is generally used when you have entity and sub-entities. If there are entities having an association with the base class in the model, then use TABLE_PER_CLASS since the base class behaves like an entity. Otherwise, since the base class would include properties/attributes and methods which are general to such entities not related to each other, using #MappedSuperclass would be a better idea
Example1: You need to set alarms for some different activities like "take medicine", "call mom", "go to doctor" etc. The content of the alarm message does not matter, you will need a reminder. So use TABLE_PER_CLASS since alarm message, which is your base class is like an entity here.
Example2: Assume the base class AbstractDomainObject enables you to create login ID, loginName, creation/modification date for each object where no entity has an association with the base class, you will need to specify the association for the sake of clearing later, like "Company","University" etc. In this situation, using #MappedSuperclass would be better.
Under a specific requirement such as not using an abstract base class (or super class), I need to implement a simple hierarchy of two entities one of which is supposed to extend the other but have a different #Id of its own.
My googling about this seems to conclude this is impossible or only on condition that I use a mapped super class (which is forbidden in my case due to a certain policy).
I don't want to duplicate the code of the entity with several dozen attributes and then mutate the duplicate by adding / overriding attributes in order to avoid future maintenance problems, but then I'm stuck in JPA restrictions.
Any help / suggestion will be appreciated.
Having different id types for non-abstract derived entities is not compatible with the JPA inheritance strategies.
What I mean is:
Imagine you have succeeded and different classes in the hierarchy use different incompatible types for the id.
how would you define the DB constraints for a single table inheritance in such a case?
and for joined inheritance?
EDIT: JPA does not distinguish between inheritance strategies when it comes to id definition. And you cannot even be sure that you can use TABLE_PER_CLASS with pure JPA. Virtually all providers implement it, but it is specified as optional and thus the least portable inheritance strategy.
The question remains however. How can the DB constraints look in order to make the table usable unambiguously by the persistence provider? E.g. Which columns should comprise the primary key on DB level?
If you cannot make the parent entity neither abstract nor embeddable nor use the same id, you will have to work around that. How you do that is highly dependant on what you want to achieve and what organizational constraints you have.
There are several possibilities - the least invasive would be composition, having A as a field in B, effectively creating a 1-1 relation.
More ugly approaches could be native and constructor queries but I doubt you want to descend that far.
tl;dr No, it is not possible.
Why is #ForceDiscriminator or its equivalent #DiscriminatorOptions(force=true) necessary in some cases of inheritance and polymorphic associations? It seems to be the only way to get the job done. Are there any reasons not to use it?
As I'm running over this again and again, I think it might help to clarify:
First, it is true that Hibernate does not require discrimination when using JOINED_TABLE mapping. However, it does require it when using SINGLE_TABLE. Even more importantly, other JPA providers mostly do require it.
What Hibernate actually does when performing a polymorphic JOINED_TABLE query is to create a discriminator named clazz on the fly, using a case-switch that checks for the presence of fields unique for concrete subclasses after outer-joining all tables involved in the inheritance-tree. You can clearly see this when including the "hibernate.show_sql" property in your persistence.xml. In my view this is probably the perfect solution for JOINED_TABLE queries, so the Hibernate folks are right to brag about it.
The matter is somewhat different when performing updates and deletes; here hibernate first queries your root-table for any keys that match the statement's where clause, and creates a virtual pkTable from the result. Then it performs a "DELETE FROM / UPDATE table WHERE pk IN pkTable" for any concrete class withing your inheritance tree; the IN operator causes an O(log(N)) subquery per table entry scanned, but it is likely in-memory, so it's not too bad from a performance perspective.
To answer your specific question, Hibernate simply doesn't see a problem here, and from a certain perspective they are correct. It would be incredibly easy for them to simply honour the #DiscriminatorValue annotations by injecting the discriminator values during entityManager.persist(), even if they do not actually use them. However, not honoring the discriminator column in JOINED_TABLE has the advantage (for Hibernate) to create a mild case of vendor lockin, and it is even defensible by pointing to superior technology.
#ForceDiscriminator or #DiscriminatorOptions(force=true) sure help to mitigate the pain a little, but you have to use them before the first entities are created, or be forced to manually add the missing discriminator values using SQL statements. If you dare to move away from Hibernate it at least costs you some code change to remove these Hibernate specific annotations, creating resistance against the migration. And that is obviously all that Hibernate cares about in this case.
In my experience, vendor lockin is the paradise every market leader's wildest dreams are about, because it is the machiavellian magic wand that protects market share without effort; it is therefore done whenever customers do not fight back and force a price upon the vendor that is higher than the benefits reaped. Who said that an Open Source world would be any different?
p.s, just to avoid any confusion: I am in no way affiliated to any JPA implementor.
p.p.s: What I usually do is ignore the problem until migration time; you can then formulate an SQL UPDATE ... FROM statement using the same case-switch-with-outer-joins trick Hibernate uses to fill in the missing discriminator values. It's actually quite easy once you have understood the basic principle.
Guys let me try to explain about #DiscriminatorOptions(Force=true).
Well , it is used in single table inheritence, i have recently used this in one of the scenario.
i have two entities which was mapped to single table. when i was trying to fetch the record for one entity i was getting list of result containg records from both the entities and this was my problem. To solve this problem i have used #DiscriminatorOptions(Force=true) which will create the predicate using Discriminator column with the specified value mapped to the corresponding entity.
so the query will be look like this after i used #DiscriminatorOptions(Force=true)
select *
from TABLE
where YOUR PREDICATE AND DiscriminatorColumn = DiscriminatorValue
I think this is more of my opinion but I think some will agree with me. I prefer the fact that Hibernate enables you to not use a discriminator. In several cases the discriminator isn't necessary.
For example, I have a Person entity which contains stuff like a name, a date of birth, etc. This entity can be used by several other entities like Employee or Customer. When I don't reference Person from other entities, but reference Employee or Customer instead, the discriminator isn't used as Hibernate is instructed to fetch either one.
#yannisf ForceDiscriminator is not the only solution to solve this issue.
You can do instanceof tests for each child class. Though this will be like hardcoding your classes in your code but is a cleaner way to solve the problem if the discriminator column is not populated.
This also helps your code avoid mixing jpa and hibernate annotations.
As pointed out by yannisf, instanceOf is kind of an antipattern in the OO world.
Another solution could be changing your entity mapping. Suppose an entity A has a refernce to a superclass B and B has child classes of type C1 and C2, the instead of A pointing to B, you can have C1 and C2 have a foreign key pointing to A. It all comes down to changing the entity design so as not to mix annotations.
Thanks
Vaibhav
I use JPA 2 with Hibernate. There are two entity classes, CardElement and IdentityDocumentKind. The last one is an inherited entity of the first one. SINGLE_TABLE inheritance strategy is used. When I try to select an instance of the parent class by query from CardElement where id = '123456' the following error occures:
Object with id: 123456 was not of the specified subclass:
org.cp.cardsystem.CardElement (Discriminator: SDocClass)
I don't have a subclass for "SDocClass" discriminator value. Actually at the moment of developing IdentityDocumentKind class querying of CardElement was used widely across the application. So I can't create a subclass of CardElement for each discriminator value and replace CardElement with it in all existent queries. It would be cost too much efforts for me. Is there a way to instantiate parent entity class when SINGLE_TABLE inheritance strategy is used?
I am not sure if I understand your problem correctly. You are using Single Table strategy to store the whole inheritance hierarchy. However, you have only mapped some of the discriminators, and this time, it is the unmapped discriminator causing the problem (because Hibernate dunno what that subclass means). Am I understanding your problem correctly?
Consider work against a special DB view instead of the real table. That view expose only records with discriminator you can handle.
Problem is solved. I've annotated root entity class of inheritance hierarchy (CardElement) in this way: #DiscriminatorValue(value = "not null"). Now I can select objects of this class without creating subclass for each discriminator value. not null and null seem to be Hibernate's special discriminator values which match in discriminator column anything except null and null respectively. I've not found any information about these values in Hibernate's official documentation. So it could be some kind of undocumented feature.
I am taking a table per subclass approach to map some data using hibernate. Typically at the database layer I would introduce a type column in the abstract table and it's subtables, which would enforce that an abstract record can only relate to a subrecord of a matching type. However from looking at various resources it seems to state that the hibernate discriminator is used for legacy databases. Does this mean I am taking the wrong approach?
Table Per (Sub)Class means you have one table per class and no abstract table. Thus a discriminator is not needed. If you have Single Table or Joined inheritance, then you'd need a discriminator.
Edit: actually, Hibernate's Table Per Subclass seems to be the Joined inheritance strategy. However, this is stated by the documentation:
Hibernate's implementation of table per subclass does not require a discriminator column.
As stated above, Hibernate table per subclass does not require a discriminator.
According to the same source:
"... If you want to use a discriminator column with the table per subclass strategy, you can combine the use of subclass and join ..."
See section 9.1.3 in http://docs.jboss.org/hibernate/core/3.3/reference/en/html/inheritance.html#inheritance-tablepersubclass