Deployments vs Releases
When a team is looking to move towards Continuous Delivery, it comes a time when it’s essential to differentiate deployment from release.
Traditionally, new versions of the software are built, tested, and made available to users in cycles that vary from once every few days to every few months. In that context, each new version will most likely contain some visible impact to the user, so the words release and deployment tend to be used interchangeably to express when a new version has reached users.
In recent years, however, many development teams have been focusing on reducing the lead time to deploy any given change down to hours or even minutes. That has been made possible by increasing the focus on automating most of the steps required to verify, assemble and install new versions, as well as increasing the focus on instrumenting and monitoring the software so problems can be spotted and fixed quickly.
The ability to deploy new versions of the software more frequently means that often a new version of the software doesn’t necessarily contain a visible impact to the user. Those new versions are still worth deploying though, as they represent small, valuable increments, and having the latest version of the software being exercised by users earlier will:
- Help the team learn how their changes behave in production, where it actually matters.
- Make potential problems (and bugs, in general) easier spot to remediate.
- Exercise the tools and procedures used to build and deploy each version.
- Promote a culture of continuous improvement.
Some examples of small, “invisible” increments that benefit from being deployed as soon as they are ready are security patches, performance improvements, or internal changes to make things simpler to evolve or maintain.
Being able to deploy sooner also means the team can take advantage of feature flags to hide incomplete features from users until they are ready. During that period, the team may have the ability to enable them for specific users (internal or external) to evaluate how it behaves before it can reach a wider audience. The feature can then be released with greater confidence and at the most appropriate time.
The splitting of deployment and release may represent a significant shift for those used to longer release cycles. The longer the release cycle is, the more significant are the changes, and higher is the chance of things going wrong. The natural response, in this case, is to be more reluctant to deploy new changes, especially when the new version doesn’t contain something visible to users.
In reality, deploying more frequently de-risks releasing changes to the users. That, in addition to the ability to better control when and how changes are made available, should make this approach very attractive to modern software development teams.