Followup - http://blogs.tedneward.com/2006/06/27/Thoughts+On+Vietnam+Commentary.aspx
Some counter punches
- I am not convinced as to why should I learn HQL or similar QLs provided by other OR Mapping solutions? Why should I not write and benefit from SQLs which have been proven to be so powerful over the ages and often outlive the lifetime of an application. So many times I have seen SQLs and Stored procedures being reused as applications moved from say .NET to Java and vice-versa.
- There is significant learning curve to an ORM add to those fancy concepts like lazy loading(if you need lazy loading you must first think that there is some problem, why are you afraid of object creation with modern day JVM and hardware or is it that you fear you will create too many objects?).
- Why do I generate, tweak so many XMLs worry about so much XML based syntaxes, how do I efficiently manage these XMLs and alter later if required? There are too many XML elements and attributes some hidden undocumented? Oh this is real pain. Annotations solve some of the meta data hell, but I am still in a dilemma if mixing meta data with code is a good practice?
- Why do think of my joins in terms of objects when they are best done with relational tables? Sorry I can only think of joining tables and not my objects.Live and let them live in their own world.
- There is no doubt ORMs are powerful and tries to make life easy. But the learning curve and deadline pressures often forces developers with somehow acquired knowledge by reading few articles and getting started guides from here and there to code and then finally when issues crop up especially too many objects, JVM crashes and sever performance worries the application falls apart and you focus more on solving these than business logic.
- Even dangerous trend that I see these days is the advent of ORM specific profilers. What does this imply? Yes the ORMs are serious cause for performance bottlenecks in your applications. I know someone came to me few months back, asking why his JVM crashes when he is running his search queries and paging? His table just had about 150k records and JVM heap was 2GB. Not a bad setting at all? Is it because the ORM was fetching a significant subset of the records based on the search criteria,first to application server before starting pagination? or my friend had a bad QL statement?
- I know ORMs can help with optimistic lock? But do they help with phantom records problem?
- Why should I cache on application server and allow or create a slightest possibility of my JVM to burst? Does caching on JVM really increase the application performance? I always knew and still continue to believe the best place to cache database records is in the database memory. Can we not gain on performance by keeping these data to where it belongs, which have been long tuned and proven to optimize performance, reducing network trips, setting tips to bulk fetch data when required, fetch data in optimal blocks? You may be tempted to use caching systems which hides behind the ORM. But think in case of an issue you have the pain of learning another framework and then fixing your errors, if Googling does not help.
- How does your ORM populate the object graph? Does it fire query after query going down the graph? If yes think about the domain model? Or think about your ORM implementation? Is there any way out? My answer is yes, smart SQL + smart mapping. I will try to show this with an example.
- However, ORMs are good, if you have indepth knowledge of the framework under consideration. Then with an ORM you can significantly cut down code in your DAO layer. This is also a myth. With Spring JDBC you can reduce your code even more or equal to what you can with an ORM. You can read an article I wrote long back on OTN(actually wrote in erstwhile BEA DEV2DEV) here to start thinking about this - http://www.oracle.com/technology/pub/articles/dev2arch/2006/10/spring-jdbc-dao.html
- One area I thought I might use ORM is when I design a product which needs to support multiple databases. But then I realised this is not true, if we manage to externalize our SQL and have an efficient strategy to map SQL result set to our domain object we can achieve database independence. 99% of the applications are not meant to be db independent. If your customer has investment so much money to procure license, build competency buy hardware for that big blackbox for Java developers called RDBMS why would he want that box of money, time and effort go down the drain by replacing it with another. So if you are trying to sell me an ORM for db independence I will not simply not buy in.
- ORMs dependent on several 3rd party libraries. Do I need to increase 3rd party dependencies? Yes you will need 3rd party jars but we must make effort to keep them to minimal. There may be version collisions between 3rd party libraries used by your ORM and main framework.