May 2018 Summaries

In episode 45 of To Be Continuous, Sam Guckenheimer, Product Owner for Microsoft's Visual Studio Team Services (VSTS), discusses how continuous integration/continuous delivery (CI/CD) has shaped the way teams approach DevOps today and how things might look in the not-so-distant future. He highlights that CI/CD allows developers to move faster, experiment more, and connect users without all the waste and barriers in between. Guckenheimer also shares Microsoft's experience with adopting continuous delivery, emphasizing its benefits such as high confidence green signal on changes going out and the ability to remediate quickly across multiple data centers worldwide. He talks about how Microsoft has evolved from a monolithic approach to microservices and how this transformation can be applicable for other organizations regardless of their scale.

In the final session of April Test in Production Meetup, a panel discussion was led by Albert Wang, Senior Software Engineer at Uber. The featured speakers were Michael McKay from IBM and Andrew Seigner from Bouyant. They discussed deploying using Kubernetes and feature flags at scale, as well as debugging Kubernetes clusters in production. Michael McKay shared how his team uses feature flags for sending their deployments and how they keep deployment feature flags separate from business logic feature flags within services. Andrew Seigner explained the process of debugging Kubernetes clusters using a proxy layer. The discussion also touched upon the use of distributed tracing, which provides more detailed information about service behavior but requires application code changes. The panelists emphasized that their approach to observability and visibility is based on pushing all data to one place for easy access. They also mentioned how they modeled their visibility part off GitHub's practices, including checking Twitter feeds for potential issues after deployment.

In his talk at the April Test in Production Meetup, Andrew Seigner, Software Engineer at Buoyant, discussed visibility and debuggability in service mesh software. He explained that service mesh is a piece of software that runs on top of platforms like Kubernetes to provide security, reliability, debuggability, and visibility into clusters uniformly. The talk focused on Conduit, an open-source service mesh developed by Buoyant, which splits its control plane and data plane for managing routing policy, telemetry, observability, and administrative interaction with the cluster. Seigner demonstrated how to deploy an app, layer a service mesh onto it, and use tools like stat and tap to identify and fix failures in the system.

Matt Knox has had an eclectic career, which led him to work on feature flagging at Twitter in 2009. He built Decider, a tool for dynamic decision-making that combined Releaser and Darkmode systems. Feature flagging is most useful when companies need to turn features "on" or "off" without deploying code changes. It can be used for A/B testing, migrating from monolithic to microservice architectures, and handling thundering herd problems. The number one mistake made around feature flagging is not cleaning up after oneself, which can lead to chaos if storage fails. Feature flags play a significant role in DevOps and Continuous Delivery by allowing frequent changes without deploying code. As companies grow, it's crucial to introduce dynamic decision-making systems early on to avoid having multiple idiosyncratic systems.

Software is becoming an essential component of products sold by traditional enterprises, leading to a shift from project-centricity to product-centricity in IT shops. This change is particularly important for application development groups and is part of the DevOps cultural shift. Product-centric testing differs from traditional software testing as it involves leveraging feature management across the entire customer journey. Feature management, such as that provided by LaunchDarkly, can be an integral part of a product-centric approach to building digital journeys, giving more power and control to diverse digital product teams responsible for delighting customers in the digital era.

In April, IBM hosted a Test in Production Meetup where Michael McKay discussed managing deployments at scale using Kubernetes and LaunchDarkly. He explained the concept of Razee, which is used to manage Kubernetes with Kubernetes. The presentation covered the scaling of Armada, challenges faced while managing thousands of clusters, and how IBM has improved its deployment process by focusing on visibility, transparency, decentralization, standardization, and simplicity. McKay also demonstrated RazeeDash, a tool that provides visibility into what's running on particular clusters, and explained the use of LaunchDarkly for managing thousands of deployments across multiple clusters.

Ian Dugas, Sales Manager at IOpipe, shares how LaunchDarkly has improved efficiency in his company's product development process. As a seed stage startup, IOpipe focuses on iterating its product rapidly, and LaunchDarkly helps minimize disruptions to developers by enabling sales teams to unlock features for specific users without interrupting the product team's sprints. This results in faster delivery of functionality and more active participation in feedback processes with developers and product managers.