May 2023 Summaries

Unleash 5.1 introduces new features and improvements, focusing on bulk toggling and enhanced login options. The bulk toggling feature allows users to enable or disable multiple flags simultaneously, complementing other bulk actions such as archiving and tagging, with actions applying existing strategies or defaulting to a 100% rollout. Additionally, improvements to SSO and sign-in options include a warning screen for disabling password-based logins and bug fixes on the password change screen, along with updated documentation for setting up Azure AD SSO. These updates reflect user requests and aim to enhance functionality and user experience.

Nick Jiang describes how his team at Codeium tackled a scaling issue in their AI code assistant using Unleash Edge. The problem arose when a buggy release caused their Postgres instance to overload, as Unleash-generated metrics resulted in an excessive number of transactions per second. This was due to the unexpected usage of Server SDKs in their distributed system, which scaled with their users and directly impacted the Postgres database. After consulting the Unleash Community, they deployed Unleash Edge, which acts as a cached read-replica between the SDK instances and the main Unleash server. This solution effectively decreased the transaction load by over 100 times, despite introducing minor complexity and propagation latency. The team found the trade-offs acceptable, as they could adjust the refresh intervals to manage latency, and Jiang recommends Unleash Edge for its scalability benefits.

Mark Fulton, an implementation architect at Unleash, shares his positive experience as a new hire on the Customer Experience and Insights team. He describes his journey from discovering Unleash through a former colleague to being impressed by the company's mature documentation and open-source emphasis. Mark highlights the transparent and well-structured interview process, which included a practical exercise that helped him understand Unleash's platform and customer-focused approach. In his first week, he engaged in a comprehensive onboarding process that included learning about the company's strategies, product knowledge, and customer interactions. He appreciated the supportive and collaborative environment, where each new employee is encouraged to contribute from the start. Mark's initial contribution was developing a tutorial to fill a gap he identified in the new user experience, which also enhanced his understanding of Unleash's SDKs. Overall, he expresses enthusiasm for working with Unleash's customer-centric culture and contributing to the company's future successes.

Flag streaming, which involves using a long-lived connection for real-time updates from a flag service to client software, is often showcased for its immediate change delivery capabilities but lacks practicality outside niche applications like chat apps, stock tickers, and gaming. In most cases, polling at discrete intervals suffices, especially for server-side implementation, while client-side software benefits from fetching flag values during initialization and specific user interactions. Despite the allure of streaming, its necessity is mitigated by the inherent latency and reliability issues of SaaS offerings, making proxy solutions more appealing for service reliability. For client software, unpredictable network performance suggests an alternative approach of initializing, caching values, and avoiding time-bound interval updates, favoring event-triggered refreshes. Streaming, although not essential for most use cases, remains a well-established technique and can be implemented where real-time updates are critical.

Gard Rimestad's article provides an update on Unleash's infrastructure, focusing on improvements in load balancing, observability, and deployment processes following their transition from VMs to Kubernetes. The team addressed challenges with Traefik load balancer pods timing out by adjusting their Kubernetes service configuration to better synchronize with AWS Network Load Balancers, ultimately opting to expose Traefik through host ports for more reliable traffic routing. In terms of observability, they shifted from direct Cloudwatch logging to using Fluent Bit, Loki, and S3 for efficient log management and implemented VictoriaMetrics for enhanced metrics collection. For deployments, Unleash optimized their approach by dividing the deployment responsibilities among different operators, significantly speeding up customer updates and reducing the burden on developers. Looking ahead, they are exploring the use of Spot instances and developing a Kubernetes drain assistant to enable the integration of Karpenter for autoscaling and resource optimization.