Abstract
In this webinar, we discuss a layered approach to progressive delivery, presented by Dave Karow, a Continuous Delivery Evangelist at Split Software. The webinar begins with Karow asking three questions related to team releases, release experiences, and measuring release success.
Karow then introduces the concept of progressive delivery and its roots, highlighting the importance of learning early in the process and limiting the blast radius. He defines progressive delivery as a set of skills and technologies focused on modern software development, testing, and deployment, with the goal of achieving faster, safer, and smarter releases.
The webinar mentions two role models that have been practicing progressive delivery: Walmart Expo and LinkedIn experimentation. Walmart built its own platform for gradual rollouts and learning from them, using a “test to learn” and “test to launch” approach. LinkedIn uses experimentation to determine the impact of releases on business metrics, and they employ guardrails to monitor key indicators such as errors and response time.
The next section discusses the foundation of progressive delivery, which involves decoupling deployment from release. Three methods are mentioned: blue-green deployment, canary releases, and feature flags. Blue-green deployment allows for a seamless switch between different versions of infrastructure, minimizing downtime and enabling quick rollbacks. Canary releases involve routing a small percentage of network traffic through new release containers to limit the blast radius and allow for quick reversions. Feature flags, when integrated with data, provide granular control over code releases and enable learning during the process by collecting telemetry and analyzing user experiences.
The webinar concludes by summarizing the benefits and limitations of each decoupling method, highlighting the importance of combining feature flags with data for automated and data-informed practices, which form the upper layers of the progressive delivery pyramid.