Everyone knows it takes forever to get something through the process and into production. Who are these CD folks trying to fool?
Let's start by asking ourselves one very important question: What is Continuous Delivery?
Thousands of years from now, when mankind has destroyed the Earth and reaped extinction as the reward, and birds, whose ancestors were the dinosaurs, have at last gained mastery over our world, they will determine that humans derived most of their information from two sources: Wikipedia and Entertainment Tonight.
Unfortunately, ET seems to have very little to say about Continuous Delivery, so we'll fall back on Wikipedia: "Continuous Delivery (CD) is a software engineering approach in which teams keep producing valuable software in short cycles and ensure that the software can be reliably released at any time."
Jez Humble, a Continuous Delivery authority, describes it as "a set of principles and practices designed to reduce the cost, time, and risk of delivering incremental changes to the users."
Currently, Continuous Delivery, in all its incarnations, is one of the hottest topics in the programming world and one of the areas most being pursued by IT departments around the world. Web-based IT departments, that is. IBM i shops not so much.
CD is not a theory. Amazon uses it to deliver a code change every 11.2 seconds (2013 figures). Etsy and Netflix are two other companies that have gone into CD big time and dramatically increased their deployments per unit of time. In the process, they have found that it not only reduces the cost and time for software development, but also dramatically cuts the number of bugs that make it to production and reduces the time to restore to a working version of the code when a bug does appear.
How Things Used to Work
A typical lifecycle for a software project or modification used to go like this:
Step 1: Someone decides that something is wrong and what, in general, should be done about it.
Step 2: Someone puts a copy of the program that needs to be modified in some sort of development environment where no one can get hurt, makes the change, and tests it within that environment. This test might be a full test or just partial. In the end, it's enough to make the programmer feel like this just might work.
Step 3: The code change is promoted to a test environment, and either another programmer or a user representative tests it to see if it works OK and, especially, to make sure that it doesn't break anything else. In many shops, one test environment isn't enough, and sometimes two or even three are used serially. You can never be too careful, you know. What's important is that the testing and evaluation is all manual.
Step 4: Somewhere along the line here, the software change is reviewed and approved by someone at the management level. This might be part of the test process, or it may just be a cursory check that testing, documentation, and other such stuff has been done. What's worse, it may not just be one management type. Many companies have developed what can only be called "painful" change control systems, in which several layers or silos of management need to approve even the smallest change. Or there may be a separate QA department that's responsible for this check. In police lingo, they are the Internal Affairs department and the most likely to have a bomb wired to the ignition of their cars. Bottom line: somebody (and generally several somebodies) have to check off that the change was done and that it works correctly. This information will be required at the trial.
Step 5: Finally, the change is moved to production. If you're in an RPG environment, this is usually just a library transfer, and it may or may not be automated. In a web environment, it generally occurs as part of a "build," where the build may contain not just one but a number of modifications. Either method probably inserts some wait time into the process (unless it's an emergency), and often shops will limit non-emergency fixes to a certain day of the week or month.
In general, it's a step-by-step, very labor-intensive process that's designed to make sure that what goes into the system is solid. As practiced in most shops today, the main emphasis is disaster prevention. No, I'm wrong about that. Theoretically, it's about disaster prevention, about making sure that something isn't put in that brings the system down. But, practically, it's about assigning blame. Everybody knows that not everything is going to go right all the time, and when it doesn't, we want to know who is responsible and make sure that never happens again.
If the above cycle sounds familiar, it should. This is the basic cycle that most midrange shops follow, and it was the basis for where web development took its start. Currently, there's a lot of difference between the two development cycles, primarily because of automated testing on the web side.
All I really want to point out here is that the above is what we have done historically and that Continuous Delivery is designed to help streamline that process, reducing the labor-intensiveness and shortening the time frame.
A Thumbnail Sketch of Continuous Delivery
Now we're going to talk more about the web side things because that's the bailiwick in which CD and its cohorts exist.
Historically, promotion of web code is centered around a "build." The build is just another release of a program or script and is identified by a build number. From this standpoint, it's no different from an RPG program promotion. It's just a different word.
A given build could consist of a single program or many programs and represent either a single development event or the accumulated output for several developers over a period of time.
For example, a company that I know does builds on an "occasional" basis. They're more or less shooting for several times a month, but it's not unheard of for things to build up for a couple of weeks as they try to work out last-minute bugs. This is one negative thing about accumulated builds. Inevitably, five items will be held up because something else in the build is having problems.
Builds are often assisted by build automation tools, and this is a whole area of web development. For compiled languages like Java, the tools will automate the compiling and linking of modules. For scripted languages like PHP or Ruby, it handles the packaging of the modules. Build tools also take care of moving objects to test libraries, support source code dependency management, keep track of the various builds (think iterations) for error tracking and history purposes, incorporate automatic testing, and other stuff. No one in the web world does builds without using some sort of build automation tool, preferably one that's appropriate to the language they're using.
In some ways, these are similar to some of the change management systems currently in use on the i, but they tend to be a bit more sophisticated. One thing to keep in mind, however, is that the environment where most web development is done is much less user-friendly than we're used to with RPG and the i. It's hard for me to compare even the nicer text editors to PDM and definitely hard to compare them to RDi. But even with the i, the move to ILE has made the build process more complex. For example, instead of just doing a '14' compile, now we're often doing a CRTBNDRPG and then wrapping several modules into a CRTPGM, a two-step process. And that two-step process is a requirement with every service program. Granted, you can roll your own CL to automate some of that, but it's still command-oriented. Build types of tools do have a certain utility.
Anyway, getting back to continuous delivery, the intent there is to stop doing large builds that contain multiple elements. What's the point if you have good build tools at your disposal? Rather than have a couple of enhancements that are ready to go and then have to wait for one that isn't, put each one in independently and eliminate the wait time.
This becomes even more important if you have multiple teams working on a single application. Rather than have people work for weeks or months and then try to integrate things, CD integrates small modifications, always making sure that you have code that works.
Obviously, the key thing here is streamlining your build process so that it happens automatically and flawlessly.
What Makes Continuous Delivery Possible?
OK, I have to admit that, even though I wrote this and should know better, so far, this sounds great.
Imagine being able to have so much done for me automatically (I am thinking here from a web, not RPG, point of view, which is another argument for another day, but I won't forget it). But the question right now is, what's continuous delivery based on? I mean, it doesn't just happen. I was at ZendCon last year (a great show, by the way) and was amazed by how many companies were trying to do CD and how achieving that was definitely a process that took time and experimentation. From that, I've determined that certain things are key to doing Continuous Delivery.
A New Philosophy
First and foremost, CD requires a change in philosophy. The old lifecycle is focused on keeping the stakeholders apprised of what's going to happen and on being able to affix blame in the event that something goes wrong. In many ways, it's a philosophy of failure: "This probably isn't going to work, and we need to know who to blame and who to notify when it fails."
But CD is bottom-up driven. It's designed so that deployments aren't a big deal; they become routine, boring. And they incorporate feeds on what's being done so that any stakeholders who want to are able to check what's going on by themselves.
It puts responsibility for the whole process in the hands of the developer and then gets out of the way to let them do their job. In the end, its focus is providing users with as much new functionality as possible, rather than on protecting the organization from any negative consequences of something that doesn't work as planned.
You need buy-in to this philosophy from everyone—from stakeholders and sponsors down to programmers and users. In the end, it will be the philosophy, not the tools, that allows you to do CD. But there are some tools that you'll need as well.
Agile
Second, you have to be agile. Oops, sorry, Agile. I mean capital "A," really Agile, not just wishing you were. That rules out any i shop I have ever been involved with. If you are i and Agile, let me know. I want to work with you!
Agile means that you do everything in short iterations. Yes, there may be grand projects, things that make the VP of Future Directions drool with delight, but we don't actually try to estimate our time against them. Agile projects require short time frames (not two months but a week), and they require that work be done in that time frame (rather than saying, "We're making progress, but it's not quite done yet.").
Without Agile, don't even bother.
Build Tools
If you look at a traditional software cycle, it's all human involvement. A hundred spots where someone like me could screw things up and cause a delay.
Build tools automate much of the tedium that causes errors in even the best programs. The question isn't whether you're automated. The question is how automated you are. Automation reduces the personal labor required and also reduces the possibility of manually introduced errors requiring us to redo steps.
Automation is essential to CD.
Automated Testing
Of the four, this is probably the most important. After the other three, that is.
Ask developers what takes the most time in a software project and they'll unanimously declare "testing." It's the hardest, least-liked, most easily put off part of any development project.
Yet, even though we know our coding skills are exceptional, this is the one element that must be done to know for sure if things were done correctly or not.
The problem is that manual testing is uncertain. The simple fact is that some people are better testers than others. But automated testing should take most of that out of the equation.
The trick is, how do you do automated testing? And I guess that would be another topic for another day, but it starts with Agile and the ability to write things as simple, uncoupled modules where the outcomes are pretty much defined. You then have to focus your testing efforts on whether the change you made broke the module. And you use automated tests to determine if the module still works. In other words, the focus becomes not just a line of code but how the module as a whole works.
So What Is Continuous Delivery?
Continuous Delivery is not a trick. Nor is it something that's obvious or easy to implement. Nor is it something that the i can't participate in. But that's yet another topic for another day.
In the end, it's a set of practices designed to simplify and automate the software pipeline. And that's where we need to go. There are too many programs. Too many enhancements. We can't do it the old way. Whether web or midrange. A new way has to be found, and maybe CD is it.
Some References
http://www.thoughtworks.com/insights/blog/case-continuous-delivery
http://www.infoq.com/news/2014/03/etsy-deploy-50-times-a-day
http://www.slideshare.net/mikebrittain/principles-and-practices-in-continuous-deployment-at-etsy
http://www.slideshare.net/beamrider9/continuous-deployment-at-etsy-a-tale-of-two-approaches
LATEST COMMENTS
MC Press Online