December 2024 Summaries

LinkedIn, a global leader in professional networking, faced challenges managing its complex infrastructure, including legacy systems and custom-built technologies. The company's internal synthetic monitoring system struggled with limited visibility and delayed issue detection, resulting in prolonged Mean Time to Detect (MTTD) and frequent service disruptions. To address these challenges, LinkedIn partnered with Checkly, adopting its Monitoring as Code solution to modernize and enhance reliability. By automating TLS monitoring, implementing dynamic API checks, and integrating Playwright-based tests into its CI/CD pipeline, LinkedIn significantly reduced MTTD and empowered its engineers to take ownership of monitoring. The integration provided real-time visibility into user experience, reduced costs through efficient testing processes, and ensured critical user flows were protected during deployments.

The author of this text has experimented with AI tools like ChatGPT and Claude to generate Playwright tests, but found that "normal AI consumer tools" aren't code-focused enough. They then turned to GitHub Copilot, an AI-assisted coding tool, to see if it could help with test generation. The author learned that a good LLM coding prompt consists of multiple building blocks, including role prompting and setting clear code generation boundaries. They also discovered that providing application code and context is essential for generating high-quality Playwright tests. The author found that Copilot has two significant advantages when used for coding: it can embed source code into prompts and provide project-specific roles and instructions. However, they concluded that relying solely on AI to generate end-to-end tests is an "intriguing myth" and that human expertise is still necessary to ensure test quality. The author believes that new Playwright tools may monitor the DOM and feed HTML snapshots to LLMs to generate locators and actions, which could be a better approach than prompting source code.

Kubernetes has enabled more complex microservice architecture, leading to a gap between the state of dashboards and the health of services for users. To ensure uptime, teams need external tools checking on their systems, such as synthetic monitors like Kuberhealthy and Prometheus Blackbox exporter. While both tools monitor endpoint availability, they serve different purposes: Kuberhealthy focuses on internal Kubernetes health checks, while the Blackbox Exporter is designed for monitoring uptime and connectivity of external services. Choosing between them depends on specific needs and use cases.