The technique of using ORM has been becoming the defacto standard, when developing new application, but after passed the level of school books example, the usage of using ORM is not so simple. In this blog I will talk about minimizing code for simple CRUD actions, but also arguing about getting back to basic for complex side of ORM.
Before started, one must recognised that the number one implementation of ORM is hibernate and that is the implementation framework used here for the specification JPA 2.0 contained in EE 6.
ORM Best Practise
All domain object contain some common characteristic, e.g. primary key, toString method, maybe created and last modified date time, etc. Put all those things in a abstract domain class
Note about abstract domain class.
- Every domain class will be serializable.
- We add a toString method, which is realized with reflection. Reflection is not the fastest technique, which is OK for a toString method, since it should not be called often.
- We add javax.persistence.Cacheable(true), for preparing for entity second level cache.
Next is to create a concrete domain class.
Note about concrete domain class.
- We make use of javax.persistence.AttributeOverride to specialize a primary key for concrete domain class.
- WE DO NOT MAP RELATIONSHIP WITH ORM. Se below.
KISS. Back to basic for relationship.
Not using ORM for our relationship, might sound crazy at first, but the fact is that managing relationship with ORM is hard and comes with several not obvious side effects.
So what went wrong? Lets start from the beginning. First we mapped all our relationship with ORM. Next we was challenged with how do we load all relationship. You might then started with loading all relationship eagerly. But after testing your implementation with more production like data volume, you realize that you are loading big chunk of your database in memory. This does not hold. OK, you comes to your senses and add lazy loading to all your relationship.
So what happened next? You started to optimize all your queries so you did not need to load each sub children separately. You probably now came across exception like org.hibernate.loader.MultipleBagFetchException: cannot simultaneously fetch multiple bags. And problem like loading loading N number of extra data. Or why do you need to load all children for inserting a new child to a parent?
When reaching this point you start to doubt the entire usage and justification of ORM. And that in fact is where I somewhere also landed. So lets get rid of all complexity it means of handling relationship with ORM and get back to basic - KISS.
We extract common CRUD task to a DAO base class.
Web Application Test
The testing of Second Level Cache with Infispan is not easily done outside the container. So lets write a simple REST service which we can simply test with e.g. REST client - Google Code
And the web.xml
To make this work on JBoss 7 and later we need to add src/main/webapp/WEB-INF/jboss-deployment-structure.xml.
And to make CDI work we add src/main/webapp/WEB-INF/beans.xml
Finally deploy it on JBoss and play around and watch log with below extra logging in JBoss.