March 2015 Summaries

In Go, the `panic` function serves as an unrecoverable exception that propagates up the call stack until it reaches the topmost function in a goroutine, causing the program to crash. However, for programs structured as asynchronous handler functions like daemons and servers, recovering from panics is crucial to ensure continuous processing of requests. Go's built-in HTTP server package recovers from panics automatically, preventing bugs in handler code from taking down the entire server. However, if a goroutine spawned by a handler function panics, it can crash the server. To address this issue, a utility function called `GoSafely` is introduced to wrap all goroutine creation. This function uses deferred recovery and logging to handle any panics that may occur within the goroutine. While this approach helps protect against naked use of go routines in our web services, it does not offer protection against third-party code spawning a goroutine that panics.

WorkHands, a platform connecting the blue collar workforce, faced challenges in managing two websites and optimizing user registration flow for over 25,000 users. They implemented LaunchDarkly, a continuous delivery platform, which allowed them to launch, test, and optimize features without having to deploy new software versions. LaunchDarkly's toggle sliders provided flexibility for WorkHands' sales team to demonstrate the software at any time. The Rollout feature enabled WorkHands to slowly introduce users to their new website by controlling the percentage of users accessing it. This allowed them to focus on building their product without worrying about infrastructure management.

The text discusses the importance of considering factors beyond just uptime when assessing the reliability of a system that uses a remote API. It highlights how LaunchDarkly, an API provider for feature flags, focuses on maximizing availability by trading off consistency and minimizing the possibility of network partitions. The company leverages CDNs like Fastly to cache content and reduce the number of required network calls, while also implementing caching in their SDKs to further improve reliability. The text emphasizes that measuring reliability should involve considering metrics such as baseline error rates, median + 90th percentile response times, and failure mode under load, rather than just focusing on uptime alone.

In this text, the author shares their experience of bicycling cross-country from Bellingham, Washington to Washington, DC in 2007. They emphasize how resetting their odometer to zero every day was a reminder that they had to start anew each day, regardless of how far they traveled the previous day. This concept is then applied to their experience as CEO and co-founder of LaunchDarkly, where they highlight the importance of continuous effort in growing a startup company. Every day starting at zero also helps them maintain balance and avoid burnout while celebrating small victories along the way.

Downtime is sometimes necessary for upgrades or maintenance in applications. There are different types of maintenance modes such as whole-site, whole-feature, and per-user feature maintenance. Depending on the stack, there may be existing libraries to help with this, but building your own 'maintenance mode' flag into your application can provide more flexibility. LaunchDarkly offers a solution for implementing maintenance mode flags as feature flags, allowing developers to only degrade the site as much as necessary and no more. Automation scripts can also be created to streamline the process of turning on maintenance mode for one user at a time.

LaunchDarkly is a platform that allows developers to control feature access for early adopters or specific users. It enables businesses to separate business logic from code, allowing faster development, feedback collection, and iteration. The platform now offers an easy-to-use dashboard where customers can view which users have access to what features. This eliminates the need for time-consuming integration with user databases. Users can search for individuals by email, first name, or last name, see their current feature access status, and make changes as needed. WorkHands CEO Patrick Cushing praises this feature, stating that it streamlines the process of granting access to new instructors through a simple click on LaunchDarkly's dashboard.