May 2023 Summaries

The Sentry SvelteKit SDK is now generally available, providing support for monitoring, error tracking, performance analysis, and session replay for SvelteKit applications. The SDK supports all of Sentry's features right out of the box, including error monitoring, performance monitoring, session replay, and profiling. Adding the SDK to a SvelteKit app is simple, with a one-line command that sets up the SDK and guides users through configuration. The SDK was developed on top of existing Svelte and Node SDKs, utilizing hooks exposed by SvelteKit to capture errors and spans on both client and server. Future plans include adding support for more adapters, resolving stack traces in dev-mode, and incorporating community feedback.

The migration of JavaScript and SourceMap processing from Python to Rust in the Symbolicator service at Sentry led to initial challenges, including low throughput and backlog issues due to lock contention. The core problem was identified as the need for a &mut reference for Read and Seek operations, which caused contention when sharing a zip::ZipArchive across multiple tasks. The solution involved giving each task its own Mutex-locked copy of the ZipArchive, resolving the contention and improving system performance. However, the frequent decompression of zip archives has emerged as a new bottleneck, suggesting a need for reconsideration of the use of zip archives for tasks requiring frequent access. Despite this, the system's overall architecture remains high-performing, with decompression now being the primary performance constraint.

The use of web fonts can cause cumulative layout shifts (CLS), which negatively impact user experience. To minimize CLS, it's recommended to avoid using web fonts or use built-in fonts instead. If web fonts are necessary, consider using fewer font files, variable-size fonts, hosting your own fonts, and the `font-display` property with a fine-tuned fallback font. The `font-display` property can be set to various values, including `swap`, which ensures that the user sees a visible fallback font while the correct font is loading, or `fallback`, which hides the text for 100ms before showing the fallback font. Using performance monitoring tools like Sentry can help track CLS scores in production and identify areas for improvement.

Tame the performance of code you didn't write: A journey into stable diffusion` The author embarked on a journey to understand the Stable Diffusion codebase, which is written in Python and uses deep learning and neural networks. They set up Sentry, an error tracking tool, to instrument the code and gain insights into its performance. The author created a Sentry transaction spanning the `txt2img` function and used Profiling flame graphs to visualize the code execution. By analyzing the flame graph, they identified slow-running functions and added them to the list of functions to instrument. This process allowed them to drill down into performance bottlenecks and gain a better understanding of the code's behavior. The author concluded that by using Sentry and Profiling, developers can transform an opaque codebase into one that can be reasoned about, allowing for improved performance and debugging capabilities.

Sentry Profiling is a solution to help users identify performance issues in their applications by pinpointing the lines of code causing bottlenecks. It collects data on execution times and aggregates results for a comprehensive view of application performance across environments. The tool saves hours of manual instrumentation, allowing developers to build performance as a feature from day one. Sentry Profiling complements other Performance features, including Performance Issues, which detects common problems by grouping transaction data. The tool provides code-level visibility, enabling users to see exactly which call stacks occur most frequently during slow spans and find the root cause without adding instrumentation. It also supports UI Frames visualization on iOS and Android, rendering slow and frozen frames inline with profiling data, making it easier to identify performance issues and optimize resource consumption. Sentry Profiling is now available for additional platforms, including PHP, Ruby, and .NET, and has been used by companies like Azos Seguros and SpiffWorkflow to improve application performance and troubleshoot issues quickly.

Sentry aims to streamline the triaging process for developers by improving time to resolution and issue resolution rate through various enhancements such as Suggested Assignees, reduced alert noise, and automated code mappings. These improvements help users identify and assign issues to the right team or developer, resulting in better outcomes including faster issue resolution and reduced un-targeted notifications. Sentry continues to invest in making triaging more efficient for its users with upcoming features such as targeted notifications to Suspect Commits, support for larger CODEOWNERS files, and easier ownership rule setup.

Riot Games' Julius Skripkauskas and Nextdoor's Walt Leung discuss building better mobile experiences, focusing on choosing the right technology, tracking key performance indicators (KPIs), and staying up-to-date with trends in device formats and AI. They share their experiences with cross-platform frameworks like React Native and Flutter, as well as declarative UI frameworks such as Jetpack Compose and SwiftUI. The duo emphasizes the importance of measuring mobile app performance using tools like Sentry to ensure a high-quality user experience, particularly for large-scale apps with millions of users. Additionally, they touch on the future of devices in mobile development, including larger screens, foldable devices, and wearable technology, and how AI can be used in mobile app development, such as with ChatGPT and GitHub co-pilot.