An Overview of Object-Relational Mapping (ORM)

It is no secret that relational databases are the most common type of databases in a majority of organizations today when compared to other formats (for example, object-oriented, hierarchical, network). Product names such as Oracle, Microsoft SQL Server, MySQL, IBM DB2, and Sybase are common terms used by developers in our line of work.

On the computer languages side of things, object-oriented (OO) programming has become the norm. Languages such as Java, C#, C++, and even OO scripting languages are common discussion topics among developers.

A majority of the software applications that use relational database and OO languages end up writing code to map the relational model to the OO model. This can involve anywhere from cumbersome mapping code (because of the use of embedded SQL or stored procedure calls) to heavy-handed technology, such as EJB's entity beans.

Because most of us seem to like both relational databases and OO, Object-Relational Mapping (ORM) has become a natural choice for working with POJOs (plain old Java objects), especially if you don't need the distributed and secure execution of EJB's entity beans (which also map object attributes to relational database fields).

Although you still need to map the relational model to the OO model, the mapping is typically done outside of the programming language, such as in XML files. Also, once this mapping is done for a given class, you can use instances of this class throughout your applications as POJOs. For example, you can use a save method for a given object and the underlying ORM framework will persist the data for you instead of you having to write tedious INSERT or UPDATE statements using JDBC, for example.

Hibernate is one such ORM framework and given its popularity in the world of Java today, we will use it for Time Expression. A few others, such as JDO, iBATIS, Java, and Apache ObJectRelationalBridge, are listed at the end of this chapter under "Recommended Resources."

Hibernate also supports the EJB 3.0 standard, so should you need to move to EJB 3.0, it'll be an easy transition (in fact, EJB 3.0 is based on many of the concepts and techniques found in Hibernate). EJB 3.0, as you might already know, aims to simplify working with EJB technology prior to this release; for example, EJB 3.0 provides a lighter-weight persistent API similar to the one provided by Hibernate. However, if you do not need the many services provided by EJB technology, you can use the Hibernate core technology by itself (without needing a big EJB container product such as an application server).

Before delving into Hibernate, let's review some basic concepts common across ORM technologies. Later we will look at lots of Hibernate Java code and XML file examples. After you have the hang of coding using an ORM framework, you will almost certainly not turn back to the old ways of working with relational databases.

Relationships and Cardinality

Database relationships are typically defined in terms of direction and cardinality (multiplicity in OO terminology). From an OO perspective, relationships are defined as association, inheritance, or aggregation. Many software development projects use ORM either with existing databases or are required to conform to standards established by a database group within the organization; hence, I will approach our relations discussion from a database perspective.

Note

Relationships can be viewed as unidirectional or bidirectional for objects. On the other hand, relations in a relational database are bidirectional by definition because related tables know of each other. However, if we were designing objects that map to the database, we would factor in both types of relations because object relationships have to be made bidirectional explicitly. So for the sake of our discussion on relationships and cardinality, we will pretend that the database can have bothunidirectional and bidirectionalrelations.

Unidirectional is when one table knows of another, but not vice versa. For example, you might have a record that uses a unique primary key; this same primary key can be used as a foreign key by records in a child table, thereby establishing a unidirectional relationship. In a bidirectional relationship, records in both tables would know about each other. For example, assume we have two tables named Employee and Project to store information about which employees worked on which project. In the Project record, we might have an EmployeeId foreign key. On the flip side, we might have a ProjectId key in the Employee table.

Cardinality can be defined as one-to-one, one-to-many (or many-to-one depending on which direction you look at the relationship), and many-to-many. We look at each briefly:

• A one-to-one relationship is when a record in table 1 can have exactly one associated record in table 2. For example, a record in a Person table might have exactly one related record in a JobTitle table.

• A one-to-many relationship is typically seen in parent-child relationships where a parent record can have several related records in a child table (for example, related via the parent's primary key).

• A many-to-many relationship is where a record in table 1 can have several related records in table 2 and vice versa. For example, an Employee table might have more than one record in a Project table (because an employee can be involved in multiple projects). On the flip side, a record in the Project table might have several related records in the Employee table because a project can have multiple employees assigned to it. Also, this type of relationship is typically achieved by using an (extra) association table (for example, a ProjectEmployee table that contains foreign keys pointing to the two main tables).

Note

We will be looking at examples of relationships in this chapter from various perspectives, namely diagrams, code, and mappings. For example, Figures 5.1 and 5.2 show examples of one-to-many relationships.

Figure 5.1. Domain model for Time Expression.

Figure 5.2. Physical database model for Time Expression.

Object Identity

An object identity (or simply, object id) is something that uniquely defines a persisted object (that is, a record in the database). It is commonly mapped to the primary key of a database table.

Cascade

Cascading can be defined as an action on a given entity flowing down to related entities. For example, if we wanted to maintain referential integrity between related parent-child tables in a database, we would delete records from a child table whenever its related parent is deleted, so that no orphan records are left lingering in the database. Similarly, when you read a parent record in, you may also want to read in all its children records. Cascading can be defined for each of the four CRUD operationsthat is, create, read, update, and delete. Also, cascading is often handled via the use of database triggers.

Mapping

Before we can begin working with objects that store and retrieve data from a relational database, we must create mappings (usually in an XML file) between the database tables and Java classes. The mapping file typically contains properties, which essentially map an attribute (variable) in a class to a column in database. If you are new to some of these concepts, don't worry; after you see some examples later in this chapter, it'll start to become a bit clearer.

There are various mapping strategies we can employ, such as horizontal mapping, vertical mapping, and union mapping. In vertical mapping, each class in a hierarchy (abstract or concrete) is mapped to a different table. For example, if we have concrete classes named Dog and Cat, both inheriting from an abstract class named Animal, we would end up having three tables in the databaseone for each class. In horizontal mapping, each concrete class is mapped to a table. In union mapping, many classes (presumably part of the same hierarchy) map to a single table.

Although vertical mapping is more flexible, it is also more complex because it requires multiple tables to extract all the data. Hence, we will use horizontal mapping because it is a simpler design and can provide faster performance, especially for simple to reasonably complex applications. To be more specific, our approach will involve one table per class mapping strategy.

In-Memory Versus Persisted Objects

When we are working with ORM technologies, there is a distinction between database objects we have in memory versus persisted ones. If the object does not exist in the database, or its attribute values do match the corresponding column values in the database, it is considered an in-memory object. For example, Hibernate distinguishes object states as persistent, detached, or transient (each is explained later in this chapter).

Another way to look at this distinction is that if we remove an object from memory (for example, by removing it from a Java collection), it does not necessarily mean the record has been physically deleted from the database (unless, of course, we mapped the collection in Hibernate to have automatic cascading during parent deletes).