Managing the multitude of software releases the enterprise needs to support can feel a lot like drowning in release pipelines
One of my favorite scenes from Disney's film Fantasia is that of the 'Sorcerer's Apprentice' (originally, a poem by Goethe). Mickey Mouse, as the apprentice, is being trampled by an army of enchanted, cloned brooms. The brooms multiply automatically in an endless march to get the job done, fetching water in buckets to fill up the room and threatening to drown Mickey in the process.
For release managers responsible for software delivery in large organisations, managing the multitude of software releases can feel a lot like drowning or being trampled by an army of brooms.
Just like Mickey, sometimes you may feel you've found the perfect magic, the one, simple, automated pipeline to command the release of your one application and daisy-chain your tools of the trade along the pipeline. And that feels great. As you configure this one 'broom' to automate your chores and release this software to production.
And so, different groups in the organisation may repeat this process across hundreds (if not thousands) of individual teams and applications, each focusing on their one broom, ignoring the implications of the bigger picture.
One application or a simple pipeline is pretty easy. But some of our enterprise customers need to support 20,000 concurrent applications, all in various stages of being built, tested, deployed and released. What do you do then?
High volume, high complexity, high stakes
If all you had to worry about was one single track, things would be simple. But the realities of software delivery for enterprises are rarely that simple, and the scale of software production and releases that large organisations need to support can be daunting, for two primary reasons.
The first is sheer volume. As enterprises digitise more of their business and applications become more critical, software production is going into hyperdrive, putting strains on the organisation to release an ever-growing number of applications at a faster and faster pace. And for some enterprises, we're talking thousands of applications and interdependent components.
This volume of applications and velocity of releases are also mirrored in the volume of geographically distributed teams and infrastructure that develop and run these apps that need to be supported as well.
The other big issues are diversification and complexity. DevOps implementation usually starts with one small team and a pretty simple pipeline. But as you want to scale DevOps and optimise your release processes across other teams, another tool (or 50) gets introduced into the mix.
This requires another complex process-branching. Team B needs a different pipeline than Team A. Your security officer needs to approve code promotions for teams A-through-N and review the output of tests. Team C can't have access to certain environments. Team D has a unique configuration mandating the use of 10 other similarly one-off tools and processes. You need a process to manage priorities when writing tests to a specific environment configuration that's too costly to replicate across all locations, and lock-in artifacts so that teams competing for the same resource pool do not override each other. And on and on it goes.
Keep in mind, too, that while we all know the basic four stages path of the pipeline -- CI build, testing, deployment, and release -- for some organisations, each of these stages can be comprised of hundreds of different processes, encompassing thousands of tasks, and sometimes millions of jobs being executed. Sure, we want to keep it simple, but the real-world complexities of legacy code, regulatory requirements, and others often make enterprise processes difficult. Furthermore, software delivery pipelines become more complex as organisations find they need to support both the applications of yesterday and those of tomorrow.
As pipelines multiply, release managers are left struggling to stay afloat and bring some order, visibility, and predictability to the multitudes of tracks that they need to coordinate and make sense of.
Reining it in
While IT struggles to keep up, the ROI of DevOps, along with the continued advances in DevOps adoption in the enterprise, are leading organisations to look for ways to scale DevOps practices throughout the organisation. As a next phase to this evolution, enterprises are looking to address this Sorcerer's Apprentice challenge of software releases.
How can you gain shared visibility, centralised management, and governance over all your brooms across your entire software delivery processes? How can you ensure that you don't end up battling to take back control over an ever-growing number of separate automation tracks running amok that are not aware of each other, and are not coordinated as part of a larger effort?
As in Fantasia, so in DevOps: As it matures, in comes the 'Sorcerer' to rein-in your sprawl of automation gone wild. Enterprises realise the need for a seasoned 'Conductor' to command and orchestrate all their disparate DevOps tools, processes, pipelines, and multitudes of islands of automation to bring order, predictability, and scale.
It's not magic
But it does take a lot of work and planning.
To improve developer productivity, product quality, and resource utilisation -- as well as to enable enterprise-scale and cross-project visibility and management -- organisations need to automate and orchestrate their entire end-to-end software delivery pipeline. 'Automate All the Things' is a key tenet to any DevOps or continuous delivery initiative, and it's a requirement to achieve quality, compliance, speed and efficiency at scale.
This end-to-end orchestration enables standardisation and consolidation of all tools and processes under a centralised, shared, platform. This allows for re-use across teams, shortening of cycle times, cost reduction, and more. Mainly, it reduces the risk of software releases by having predictable processes, security checks, consistent monitoring and one pane of glass for the entire organisation.
End-to-end orchestration and standardisation are required for scaling DevOps effectively for today's large enterprises. To get it right, you need to map all of your pipelines across all teams and applications and design your DevOps solution from the get-go in a way that will allow you to scale while avoiding the Sorcerer's Apprentice trap.
Make sure your end-to-end solution enables you to:
Pipeline models allow you to define your end-to-end software delivery process, encompassing all the teams, tools, stages, tasks, approval gates, artefacts, and environments involved in this process. Modeling your application, environments and pipeline enables reuse and consistency.
Automate and orchestrate your entire toolchain and workflows to eliminate manual handoffs and silos of automation to accelerate your pipeline and improve quality.
Be able to support off-the-shelf plugins, robust DSL and APIs allow for extensibility and flexibility, making it easy to tie-in any tool chain, technology stack or cloud resources to gain shared control and visibility.
Role-based access control, approval gateways and automatic logging ensure security, visibility and compliance for governance and auditability.
Ensure HA and scale jobs and workloads predictably and efficiently across pools of on-premises or cloud resources.
DevOps provides enterprises with an effective approach to eliminate risk from software releases, and support the ever-growing demand for more frequent application updates. By automating your entire end-to-end pipeline and managing all release processes in one centralised platform, organisations gain visibility into the progress of all releases and DevOps processes, and ensure they can control, govern and optimise their delivery pipelines as they grow exponentially -- and not get run over by an army of automated brooms.
Anders Wallgren at Electric Cloud
Image Credit: Disney