Saturday, 25 July 2015

Time To Hop On The DevOps Ship



When I started my 1st job in 2007 as part of a graduate programme I was signed up as a Java developer on the Java skills track. After joining, due to having a degree in engineering which involved software and hardware, I was instead put on a different skill track called “Development Architecture” or “DevArch” for short as part of a group called “Development Control Services” which  specialised in automating configuration and environment management workflows for projects.  Many of those in my starter group who were in a similar position, complained that this was not what they signed up for and wanted to be developers.  At this point I would like to tell you I stuck with it due to my vision and belief that this was indeed the future but this wasn’t the case, like many graduates i wasn't really sure what I wanted to do long term. However, in time I became passionate about the job and I loved the fact that you never did the same activity for more than a few days in a row. The variety and creative freedom the job brought, along with the ability to help others do their jobs and make a real difference to projects was certainly appealing.

My first project at my new job was working to build and deploy the norwegian pensions website and I was tasked with creating continuous integration processes using a relatively new tool called “CruiseControl”. In that role i looked after deploying configurable items such as code and related config through environments that had multiple test quality gates “component test”, “integration test”, “pre-production” and production environments using our in house build and deployment tool.

Sounds familiar right? I was part of the team, that looked after the continuous integration and deployment of applications and all configuration was held in source control management system at the time “ClearCase”.  Daily our team interfaced with developers and the large environment management team that built physical tin from the ground up. That environment management team were the go to team when anyone wanted new servers for new test environments or simply wanted to scale out production or on board a new application. The thing to note was that we didn’t have ticketing systems to interface with each other, we did the unthinkable and spoke to each other and worked without team silos.

This team sounds a lot like a DevOps initiative today that is being branded a “fad” or “new philosophy” when in reality it has always been a core component of the software delivery. Every project I have worked in since that very first one, I strive to encourage the same principles of knowledge sharing and learning. When you couple this with individuals that are respectful of each other, keep calm under pressure and operate without ego or want to be heroes then great things can happen. All this contributes to a team dynamic that allows teams to be more productive, deliver products to market faster and allows people to enjoy their jobs. The want for repeat-ability and predictability in processes are not new aims and are still as valid in 2015 as they were then in 2007. This is not magical or mythical or unicorns it is completely achievable.

In 2010 while still at the same company, having climbed the corporate ladder slightly, and running “configuration management” training for new graduates, they would question being put in the “Development Architecture” skill track and ponder how they may struggle to find a job outside the company with these sets of skills. It seems crazy today that this was questioned given how sought after and in vogue “DevOps” skills are today. At the time I tried to tell those graduates the key thing that would change all this was “cloud computing” and how it will make the difference and push forward prominence of our skill set and other companies would catch up and automation wouldn't just be needed for huge projects at massive scale to save costs. Those skills and values would be applicable to any company and give them a competitive edge as more activities could be automated.

At that time in 2010 cloud computing was still taking off. Any experienced engineer worth their salt working in the configuration management space could see that cloud infrastructure would give an adrenaline shot to “Development Architecture” but I don’t think anyone imagined it would see the boom it has today. The principle was that the expensive 12 man environment management team that was required per project, to manually build infrastructure, could be replaced by a smaller team and virtualised infrastructure. Instead a set of API’s and scripts could be created by a few highly skilled resources that understood infrastructure and used to provide infrastructure as a service once cloud platforms had stabilised and actually worked as they should. Like many other early adopters of cloud technologies we had lots of fun using many of the first attempts at cloud from vendors which were very unstable and buggy.


The first stable infrastructure platform that we worked with was VMWare based and it worked like a charm and I will testify that vsphere really is a fantastic platform. I wish VMWare would stop trying to add layers on top of it and instead build all the networking capability into it along with new functionality into vsphere. Instead they toil by adding product suites above vsphere then eventually end up throwing the product away which confuses customers over what vcloud, vcac or vrealize product and set of apis they should be using.  But vsphere was a completely stable platform that could finally be relied upon and it is no coincidence VMWare excelled in the market due to this. Now instead of just controlling and delivering configurable items to manage only code, the remit for configurable items extended to controlling the underlying servers that the code was deployed on and quality assurance test packs were run on. Consequently, the configuration management remit increased from just controlling code configurable items to infrastructure too. How do we do configuration management for infrastructure, how do we treat infrastructure as code? It is no coincidence at this time that early versions of tools such as Puppet and Chef emerged to cater for and meet those needs.

Now  “infrastructure as code” is pretty much a staple of every forward thinking company and taken for granted now, with many options to manage cloud based infrastructure such as Puppet, Chef, Ansible or Salt. The ease of use of truly open platforms like AWS for public cloud and Openstack for private cloud are like a nirvana for DevOps practitioners, providing feature rich set of consumable APIs and letting users use them as they wish. The beauty of AWS and Openstack is that they don’t try and enforce rigid standards on users about how to use them that other vendors always seem to try and enforce to make an additional service charge from.  Infrastructure in under two minutes is now the demand of the day with the DevOps buzzword placed on pretty much every job specification, seeking golden gooses who will bring automated self service to their enterprises like some magical wizard. Essentially this is not a new movement or some crazy idea but an illustration that industry has finally realised the importance of configuration management. For that fact alone we should be applaud, but please don’t label this some new revelation or new invention or quote stories of magic unicorns. The branding, packaging and name given to it may be different but the underlying concept is the same and whisper it pretty simple.


Nuage networks logoNow this brings us to 2015 and the next blocker after infrastructure is networking, teams need an easy way to consume networking like they do infrastructure as waiting for a networking team to service tickets will just not do and a process is only as quick as it’s slowest component. This has seen a wealth of “software defined networking” technologies burst onto the scene with Nuage, NSX, ACI, Contrail, Plumgrid and OpenDayLight to name but a few. This is extending the configuration management remit once again this time to include network configurable items and manage the network like we first did with code and then the underlying infrastructure. It is a natural progression and makes for exciting times.The potential to simplify data center deployments using software defined networking is massive. A lot of complexity of networking is brought in by security requirements so putting the network in software opens up many possibilities to keep the networking as simple as possible, while at the same time adhering to all security requirements.

Not everyone is well versed in configuration management and the principles though. I have seen developers and operations staff discuss things like “DevOps teams” and be very disparaging. I put this down to a fundamental lack of understanding on their part, wrought out of fear, that this new buzz word called “DevOps” will end all that they hold dear and somehow take away their jobs or livelihood, which of course is nonsense. In 2007 when I was doing “DevOps” before it was called that and still called “Configuration Management” or “Release Management” we still had a development team and operations team those roles still stand and are not going anywhere they are just evolving like the technology landscape. This means embracing change though, maybe learning some scripting or at least a DSL, changing the way you work and questioning what you have done before and being open to new ideas and learning these new skills.

In 2007 the successful project team had roles that spanned different skill sets we set out to have “T shaped” people. A slightly rubbish and consultancy type analogy maybe, but a “T shaped person” had deep dive knowledge in 1 expert area which was the main reason they were part the team so that is the depth of the T. That person though also had a breadth of knowledge in other areas which made up the breadth of the T. So taking myself as an example I was an expert in automation but had a breadth of knowledge in infrastructure, networking and storage.  Each person in that team cross skilled with other T shaped experts and shared information to the point the breadth of skills across the team and in each individual's T became better and better and wider and wider. This is what DevOps is all about, creating true multi-discipline teams and sharing information to build new skills.

Recently have even seen Amazon offering absurd certifications on DevOps and it really irritates me. DevOps doesn't mean a team, it is a mindset about sharing information and collaborating with others so we can deliver software to market quicker and become better engineers. By definition a certification in DevOps should really be granted to any child that works with other children to build a spaceship with lego as they are successfully working together to build something. But if organisations form a team and call it the “DevOps Team” don't get caught up too much in debating if this is the wrong approach as it may create another silo. Instead be positive and thankful that the company have at least considered doing the correct thing even if they may not fully understand what they want to achieve. But hopefully the experts that they recruit will show them the correct way.

Adopting a DevOps mind-set should mean the polar opposite of sitting in an ivory tower and pointing someone towards a ticketing system, when you can’t be bothered to talk to them and help them solve their problem, it isn’t shut-up and forget for a day. It isnt about using existing process as an excuse for something failing, not helping someone do their job, if that happens frequently it needs to be challenged and changed. If you do have a problem with DevOps then you have an issue with sharing information, self-improvement and talking with people which is absurd and you are in the wrong job. DevOps is about working together, challenging the status quo if something doesn’t make sense and is blocking productivity. It promotes the simple option to speak to other teams and change things free of the bureaucratic processes that inhibit us in IT.


On the other hand DevOps isn’t about re-inventing the wheel and is sometimes about doing the most simple thing, using the simplest solution, over complex processes equally hinder the ability to deliver software quicker so processes need to be lean and quick. It is also about seeing if there are gaps where we can build tools that can make our jobs easier daily and sharing them with others which is summed up by the “open source” initiative. Finally DevOps means people working together to create better streamlined process, it has nothing to do with tools they are simply facilitators of process. Tools come and go but great processes and principles maintain the test of time. DevOps really is a no brainer so please get on board and be part of the revolution, it has an open door policy so it isn't too late so get on board, the ship hasn’t sailed yet.

3 comments: