August 2019 Summaries

The 2019 Accelerate State of DevOps report by Google Cloud's DORA team reveals that elite performers deploy code 208 times more often than low performers and can get code changes from commit to deployment 106 times faster. Elite performers also recover from an outage or incident 2,604 times faster than low performers. The report highlights the importance of speed in software development and emphasizes that smaller, faster software changes are less likely to break a system. Feature flags can help avoid problems in production by allowing testing in production without impacting users. Additionally, the report identifies five pillars of productivity: psychological safety, useful tools, internal search, external search, and reducing technical debt. The top three performing groups use mainly open-source and COTS tools with little customization and have a clear change process, loosely coupled architecture, code maintainability, disaster recovery testing, effective use of cloud services, and conduct organization-level work. High performers favor strategies that create community structures at both low and high levels in the organization.

Using Beamer with LaunchDarkly allows companies to send targeted in-app announcements by leveraging feature flags to tailor messages to specific user groups. This approach enables different user experiences by using the same flags for both feature rollouts and announcements. Beamer's Advanced Segments feature supports this initiative by allowing the creation of Custom Segments, which are based on the target users of a feature flag, thereby ensuring that only relevant users receive notifications about new features or changes. This integration facilitates incremental and faster releases by ensuring that announcements are made only to the intended audience, enhancing the overall efficiency and relevance of communication. For further details, interested parties can contact sales at LaunchDarkly or start a free trial.

In a session of Test in Production, Christian Posta, Global Field CTO of Solo.io, discussed chaos engineering and service meshes. He explained that with distributed systems, there are cases that cannot be found deterministically or during lower environment testing. Chaos injection is used to prove the resiliency of these systems by injecting failure scenarios and observing their impact on the system's steady state. Solo.io has built a chaos experimentation engine called Gloo shot, which works with service meshes like Istio to enable controlled fault injection and telemetry analysis.

In July, the Test in Production meetup featured a discussion on progressive delivery and its relationship to continuous delivery. Progressive delivery emphasizes release progression and control points in code, allowing developers to incrementally deliver changes to users. This approach is designed to improve user experiences by enabling teams to test new features in production with real workloads and adjust them based on real-time feedback. The meetup highlighted the importance of considering different user personas when implementing progressive delivery strategies, as some users may prefer predictable, stable software while others are more adaptive to change.

The new Flag Archive workflow by LaunchDarkly simplifies the process of removing unused feature flags, ensuring clutter-free dashboards and codebases. This feature brings together important flag information from multiple environments into a single panel for quick assessment before archiving. It helps teams make better decisions faster while maintaining safety and confidence in their features' management. Future updates will continue to enhance this workflow. For more information, contact [email protected] or start a free trial.

Feature flags are a powerful tool for managing software rollouts, allowing developers to control who can see a feature and when. Release management is one of the most common use cases for feature flags, enabling fine-tuning of features before full release. Two common deployment models are ring deployments and percentage-based deployments. Ring deployment involves gradually releasing features to different groups based on attributes or opt-in processes, while percentage-based deployment randomly selects users for each rollout stage. Companies can also combine these methods. When creating feature flags, it's crucial to prioritize them from the start and understand their purpose. Establishing naming conventions is essential for clarity and avoiding technical debt. During implementation, understanding the deployment path and testing the behavior of the feature flag are vital steps. Configuring a fallback value ensures that users experience minimal disruptions in case of failures during evaluation. When deploying release management flags, tracking metrics such as availability, reliability, and response time is crucial to ensure performance does not degrade. Removing short-term feature flags after full rollout helps avoid technical debt.

In June, the Test in Production Meetup took place in London, hosted by Intercom. Jonathan Hare-Winton from Spotify spoke about synthetic monitoring and its use at The Guardian. Synthetic monitoring is a technique that uses emulation or scripted recordings of transactions to create actions for monitoring production systems. It's useful when passive monitoring isn't enough, such as in cases where deployments are frequent and the system is constantly changing. Synthetic monitoring involves taking tests running in production and turning them into synthetic monitoring solutions. Key challenges include ensuring test accuracy, keeping noise levels down, building trust with the system, and managing potential performance effects on primary systems. The main focus should be on performing actions that will trigger passive monitoring to achieve effective monitoring of production systems.