Hibernate vs JPA vs JDO - pros and cons of each?
Some notes:
- JDO and JPA are both specifications, not implementations.
- The idea is you can swap JPA implementations, if you restrict your code to use standard JPA only. (Ditto for JDO.)
- Hibernate can be used as one such implementation of JPA.
- However, Hibernate provides a native API, with features above and beyond that of JPA.
IMO, I would recommend Hibernate.
There have been some comments / questions about what you should do if you need to use Hibernate-specific features. There are many ways to look at this, but my advice would be:
If you are not worried by the prospect of vendor tie-in, then make your choice between Hibernate, and other JPA and JDO implementations including the various vendor specific extensions in your decision making.
If you are worried by the prospect of vendor tie-in, and you can't use JPA without resorting to vendor specific extensions, then don't use JPA. (Ditto for JDO).
In reality, you will probably need to trade-off how much you are worried by vendor tie-in versus how much you need those vendor specific extensions.
And there are other factors too, like how well you / your staff know the respective technologies, how much the products will cost in licensing, and whose story you believe about what is going to happen in the future for JDO and JPA.
Which would you suggest for a new project?
I would suggest neither! Use Spring DAO's JdbcTemplate together with StoredProcedure, RowMapper and RowCallbackHandler instead.
My own personal experience with Hibernate is that the time saved up-front is more than offset by the endless days you will spend down the line trying to understand and debug issues like unexpected cascading update behaviour.
If you are using a relational DB then the closer your code is to it, the more control you have. Spring's DAO layer allows fine control of the mapping layer, whilst removing the need for boilerplate code. Also, it integrates into Spring's transaction layer which means you can very easily add (via AOP) complicated transactional behaviour without this intruding into your code (of course, you get this with Hibernate too).
I have recently evaluated and picked a persistence framework for a java project and my findings are as follows:
What I am seeing is that the support in favour of JDO is primarily:
- you can use non-sql datasources, db4o, hbase, ldap, bigtable, couchdb (plugins for cassandra) etc.
- you can easily switch from an sql to non-sql datasource and vice-versa.
- no proxy objects and therefore less pain with regards to hashcode() and equals() implementations
- more POJO and hence less workarounds required
- supports more relationship and field types
and the support in favour of JPA is primarily:
- more popular
- jdo is dead
- doesnt use bytecode enhancement
I am seeing a lot of pro-JPA posts from JPA developers who have clearly not used JDO/Datanucleus offering weak arguments for not using JDO.
I am also seeing a lot of posts from JDO users who have migrated to JDO and are much happier as a result.
In respect of JPA being more popular, it seems that this is due in part due to RDBMS vendor support rather than it being technically superior. (Sounds like VHS/Betamax to me).
JDO and it's reference implementation Datanucleus is clearly not dead, as shown by Google's adoption of it for GAE and active development on the source-code (http://sourceforge.net/projects/datanucleus/).
I have seen a number of complaints about JDO due to bytecode enhancement, but no explanation yet for why it is bad.
In fact, in a world that is becoming more and more obsessed by NoSQL solutions, JDO (and the datanucleus implementation) seems a much safer bet.
I have just started using JDO/Datanucleus and have it set up so that I can switch easily between using db4o and mysql. It's helpful for rapid development to use db4o and not have to worry too much about the DB schema and then, once the schema is stabilised to deploy to a database. I also feel confident that later on, I could deploy all/part of my application to GAE or take advantage of distributed storage/map-reduce a la hbase /hadoop / cassandra without too much refactoring.
I found the initial hurdle of getting started with Datanucleus a little tricky - The documentation on the datanucleus website is a little hard to get into - the tutorials are not as easily to follow as I would have liked. Having said that, the more detailed documentation on the API and mapping is very good once you get past the initial learning curve.
The answer is, it depends what you want. I would rather have cleaner code, no-vendor-lock-in, more pojo-orientated, nosql options verses more-popular.
If you want the warm fussy feeling that you are doing the same as the majority of other developers/sheep, choose JPA/hibernate. If you want to lead in your field, test drive JDO/Datanucleus and make your own mind up.
Make sure you evaluate the DataNucleus implementation of JDO. We started out with Hibernate because it appeared to be so popular but pretty soon realized that it's not a 100% transparent persistence solution. There are too many caveats and the documentation is full of 'if you have this situation then you must write your code like this' that took away the fun of freely modeling and coding however we want. JDO has never caused me to adjust my code or my model to get it to 'work properly'. I can just design and code simple POJOs as if I was going to use them 'in memory' only, yet I can persist them transparently.
The other advantage of JDO/DataNucleus over hibernate is that it doesn't have all the run time reflection overhead and is more memory efficient because it uses build time byte code enhancement (maybe add 1 sec to your build time for a large project) rather than hibernate's run time reflection powered proxy pattern.
Another thing you might find annoying with Hibernate is that a reference you have to what you think is the object... it's often a 'proxy' for the object. Without the benefit of byte code enhancement the proxy pattern is required to allow on demand loading (i.e. avoid pulling in your entire object graph when you pull in a top level object). Be prepared to override equals and hashcode because the object you think you're referencing is often just a proxy for that object.
Here's an example of frustrations you'll get with Hibernate that you won't get with JDO:
http://blog.andrewbeacock.com/2008/08/how-to-implement-hibernate-safe-equals.html
http://burtbeckwith.com/blog/?p=53
If you like coding to 'workarounds' then, sure, Hibernate is for you. If you appreciate clean, pure, object oriented, model driven development where you spend all your time on modeling, design and coding and none of it on ugly workarounds then spend a few hours evaluating JDO/DataNucleus. The hours invested will be repaid a thousand fold.
Update Feb 2017
For quite some time now DataNucleus' implements the JPA persistence standard in addition to the JDO persistence standard so porting existing JPA projects from Hibernate to DataNucleus should be very straight forward and you can get all of the above mentioned benefits of DataNucleus with very little code change, if any. So in terms of the question, the choice of a particular standard, JPA (RDBMS only) vs JDO (RDBMS + No SQL + ODBMSes + others), DataNucleus supports both, Hibernate is restricted to JPA only.
Performance of Hibernate DB updates
Another issue to consider when choosing an ORM is the efficiency of its dirty checking mechanism - that becomes very important when it needs to construct the SQL to update the objects that have changed in the current transaction - especially when there are a lot of objects. There is a detailed technical description of Hibernate's dirty checking mechanism in this SO answer: JPA with HIBERNATE insert very slow