Use Code Generation to Bootstrap Your Next Devops
Resistant With Good Reason
As a seasoned developer, I am lazy by nature. Actually, let me rephrase that: I am no longer interested in writing certain parts of an application's code base. If you're like me, you look forward to the challenging aspects of what needs developing, and not the parts you have likely written time and time again.
Unfortunately, for most applications of a certain size and complexity, there's a category of code that must exist: the bland code. Bland code itself is extremely important, but rather redundant. Sure, each domain puts its own flavor to bland code, but the core pieces of most apps look alike. Although necessary, bland code is predictable and lacks the pizzazz (read: challenge) most developers seek. However, the code we want to work on is useless without the bland code.
So, we need the bland code and will for some time. Wouldn't it be great if it just appeared when we need it? Whether somebody else writes it, or if it appears with the push of button, it doesn't matter -- we want to view bland code as any other reusable, robust, and trusted API.
Regardless where your opinion lies regarding Sprint Zero, let's agree there's a period of time between figuring out what/how to begin DevOps and actually doing DevOps.
With all the mature tools to help with the before and during DevOps phases it is no wonder Sprint Zero is often overlooked or seen as a preparation formality.
Since the early 2000s, there have been many attempts to standardize the concept of software models and architectures. One such standard is Model Driven Architecture (MDA), a concept developed by the Object Management Group (OMG) that defines an independence between a model and the technical environment in which it is realized.
During this same time as MDA's development, application generators were popping up everywhere. Application generation has always held out promises that have often failed to be kept. In theory, it's a good thing to generate code. In practice, most provide marginal value, often lacking the flexibility to keep pace with ever-changing model descriptions, technology stacks, application frameworks, and deployment platforms.
The process of Generation-as-a-Service needs to appear as the other DevOps services we've become used to. It must be effective, but not intrusive. It should have a low-to-no learning curve. It should be available as both an application and programmatic API. It should seamlessly fit into any existing traditional DevOps process, be it Agile in nature or otherwise. Most importantly, the effort to provide input to the service and leverage its output should be negligible compared to its benefits.
Another Trusted Service
More than ever, software development is reliant upon external services. We trust GitHub and Bitbucket to our source control needs, JIRA to handle our issue tracking, and Docker Hub to secure our container images.
We have integrated Jenkins, Chef, Puppet, Maven and many other technologies into DevOps. They have become so commonplace that running a substantial project without them will demonstrate just how reliant we are. The reason is simple: They work and fill a need in the DevOps process. Also, when chained together, they form a continuous process of app integration and delivery.
What a great thing.
So, is it a stretch to think of application generation in the same way? Each of the services and technologies mentioned have input needs and provide an output as a service. If enough structured input could be gathered without significant effort, and if a description of a target technology stack could be specified, it may be possible to see application generation as a real DevOps service.
We need an engineer who is responsible for turning business models and technical models into our bland code. A single trusted technical resource who can take our "order" and has the platform and skillset to give us what we need when we need it, so we can prioritize and mobilize. In order to enforce reuse, this engineer would know which previous orders were the same or similar to our order. This engineer would rely on a platform to merge the business and technical models to instantly generate tens to hundreds of thousands of lines of bland code. The platform would allow the engineer to choose from a set of corporate standardized templatized technology stacks to support a host of languages, design patterns, architectures and target environment.