Remove unnecessary layers from the stack to simplify a Java solution

When writing new software, it is natural to simply follow the beaten paths. People pick up what they are already used to, or what they are told to use, and then get going. There may or may not be proper directions for choosing the technology, but there is invariably a safe bet option that will be chosen. This isn’t wrong per se. What is though, is when no alternatives were ever considered when there was a chance to do so. AppServers are not necessary for deploying scalable Java solutions. JDBC and ORMs are not necessary for building high performance and reliable Java programs. The fact that a lot of people keep doing this, is no good justification for adopting the practice.

An item in my news feed this morning triggered the writing of this post. It read shadow-pgsql: PostgreSQL without JDBC. It’s a Clojure project on GitHub. Indeed, “without JDBC”. That’s it. This is particularly pertinent in the Java software world, where a flurry of popular initiatives are driving down the adoption of once revered vendor solutions. Some of the arguments I make here apply as well to other software and programming environments, not just Java.

As you start writing a Java application, destined to be deployed as part of an Internet enabled solution, it is important to carefully consider the requirements before selecting technology to build your solution with. Everyday, a new tool pops up promising to make it easier and faster to start writing software. People lose sight of the fact that getting started isn’t the hard part, keeping running and growing, adapting to change, all at a reasonable cost, are by far the hardest part with software. And to boot, once something is deployed into production it instantly becomes a legacy that needs to be looked after. So, don’t be lured by get started in 10 minutes marketing lines.

I would argue that, not knowing the expected characteristics of your target solution environment, is a strong indication to choose as little technology as possible. There are lots of options nowadays, so you would expect that people always made proper trade-off analysis before settling on any particular stack. A case in point is database access. If you are not forced to target an AppServer then don’t use one. If you have access to a component with native support for your chosen database platform, it might be better to just skip all the ORMs and JDBC tools and use native database access libraries. This would simplify the implementation, reduce the deployment requirements and complexity.

It used to be that, if you set out to write a server side Java program, certain things were a given. It was going to be deployed on an AppServer. The AppServers promoted the concept of container where your application is shielded from its environment. The container took charge of manageability issues such as security, authentication, database access, etc. To use such facilities, you wrote your code against standard interfaces and libraries such as Servlets, JDBC, or Java beans. If the company could afford to, it would opt for an expensive product from a large vendor, for example IBM WebSphere or BEA WebLogic (nowadays Oracle). And if the organisation were more cost-conscious, a bit of risk taker or an early adopter for example, it would choose a free open source solution, Apache Tomcat, jBoss, etc. There were always valid reasons for deploying AppServers, they were proven, tested and hardened, properly managed and supported in the enterprise. Such enterprise attributes were perhaps not the most attractive for pushing out new business solutions or tactical marketing services. That was then, the Java AppServer world.

As the demand for Internet scale solutions started to rise, the AppServer model showed cracks that many innovative companies didn’t hesitate to tackle. These companies set aside AppServers, went back to the drawing board and rebuilt many elements and layers. The technology stack had to be reinvented.

Over last ten years I would say, people started writing leaner Java applications for the Internet services. Mostly doing away with AppServers. This allowed the deployment of more nimble solutions, reduced infrastructure costs, increase the agility of organisations. Phasing out the established AppServers came at a cost however, one now needed to provision for all the things that the AppServers were good at: deployment, instrumentation, security, etc. As it happens, we ended up reinventing the world (not the wheel as the saying would have it, but well a certain well known world). What used to be an expensive and heavy vendor solution stack, morphed into a myriad of mostly free and open source elements, hacked together parts and parcels, to be composed into lean and agile application stacks. This new reinvented technology stack typically has a trendy name, Netflix OSS is one that sells rather well. That’s how things go in the technology world, in leaps and bounds, lots of marketing and hype, but within the noise lie hidden some true value.

Arguably, there are a lot of new java based solutions being implemented and deployed with stacks of unnecessary layers. That is a shame, because there is an unprecedented amount and variety of options for implementing and deploying good solutions in Java, or targeting the JVM. Understanding the characteristics of the envisioned solutions is one aspect that keeps driving Java and JVM based languages and technologies. AppServers are not necessary for deploying scalable Java solutions. JDBC and ORMs are not necessary for building high performance and reliable Java programs. The fact that a lot of people keep doing this, is no good justification for adopting the practice. The move towards smaller and leaner specialised applications is a good one because it breaks down and compartmentalise complexity. The complexity that is removed from the application stack, is diluted and migrated over to the network of interlinked applications. Not all problems would automagically vanish, but some classes of problems would disappear or be significantly reduced. That is the same principle that should motivate into challenging and removing as many layers as is reasonable from individual application stacks. I resist the temptation of mentioning any specific buzzword here, there are plenty of good examples.

When writing new software, it is natural to simply follow the beaten paths. People pick up what they are already used to, or what they are told to use, and then get going. There may or may not be proper directions for choosing the technology, but there is invariably a safe bet option that will be chosen. This isn’t wrong per se. What is though, is when no alternatives were ever considered when there was a chance to do so. Keeping your options open would aways prove more profitable than just going with the flow. One characteristic of lagging incumbents is that they often just went with the flow, chose solutions deemed trusted and trustworthy without regard for their own reality, usually not challenging enough the technology stack.

 

Java is the new Cobol, of course. How else can it be? Enter Eclipse Xtend.

Eclipse Xtend may be a response to Jetbrain’s Kotlin, definitely making it plain that Java isn’t cutting it when it comes to programmer productivity.

I just saw an announcement of a new programming language from none other than the Eclipse group, you know of the Java IDE fame. A very quick scan of this new language, called Xtend, it looks like they’ve borrowed a bit from Groovy, something from Scala and maybe something from CoffeeScript. The fact that this language is directly supported by Eclipse IDE is a huge advantage, and that may be to the detriment of Scala which still hasn’t got a great IDE support yet. I think this move by Eclipse may also be a preemptive strike against JetBrain’s Kotlin. If the community picks up on this then I expect initiatives like Redhat’s upcoming Ceylon programming to struggle in gaining any foothold.

I’ve been wanting to write a post on how I thought Java just wasn’t good for programmer productivity, seeing this announcement encouraged me to finally post it. I did something with JBoss Seam a few months back, took a step back to look at the code and thought to myself what a waste!. I can’t believe anyone in 2011 would want to create new code using JSF for example, it simply feels wrong.

From the moment that I renewed my focus on functional languages, I find pure Java code to be an eyesore, the web tooling around Java are simply atrocious compared to what you can do with functional languages like Ruby, Clojure or Scala. Clojure way well be the most elegant of the lot, I like it so much that I’m increasingly considering doing more hacking than I’ve done in years.

Competition is a good thing. But I’m wondering if Xtend is a more of a Groovy clone and if that could be such a good thing for the wider programmer community. I haven’t written a single line of code with Xtend yet, this is just a spontaneous reaction, in fact a post I had drafted about Java as the new Cobol, but refreshed in the light of this announcement by Eclipse.