November 2025 Summaries

Interfaces, while essential for data exchange, can impose limitations on user performance, as highlighted in a talk by Joran at Systems Distributed 2025. This concept was put into practice when a team sought to overcome the limitations imposed by the Parquet file format, which was dominating their query CPU time due to the need to convert data into the Arrow format for efficient querying. Despite Parquet's widespread use and design goals of efficient data storage, retrieval, and interoperability, it fell short in handling computational queries efficiently and required costly conversion processes. The team explored Vortex, a file format optimized for decoding and querying data directly from object storage, which resulted in a 70% performance improvement and better storage size efficiency. Vortex's design, which supports general-purpose compute pushdown and offers extensibility for future encodings, provided a more suitable interface for their needs, illustrating the importance of selecting an interface that aligns with specific use-cases for enhanced performance.

The recent development in profiling tools for Node.js introduces the ability to attach custom key/value labels at runtime using the @polarsignals/custom-labels library, enhancing the observability of asynchronous JavaScript and TypeScript code. Initially available only for languages like Go, Rust, C, and C++, this feature allows developers to set labels that annotate stack traces, aiding in performance analysis through tools like Parca and Polar Signals Cloud. The implementation for Node.js utilizes the AsyncContextFrame-based AsyncLocalStorage introduced in Node v22, which avoids the performance overhead encountered with the initial async_hooks-based approach. This improved method efficiently manages label propagation in asynchronous contexts and became the default in Node v24, facilitating better integration with profiling tools that rely on eBPF to track these labels during execution. By leveraging open-source access to Node and v8 internals, the system determines offsets in internal objects to effectively extract label data, enabling more precise profiling without incurring significant performance penalties.

Source maps are crucial tools that map minified JavaScript files back to their original source code, enabling developers to debug effectively by translating errors in production code to precise locations in the original TypeScript or JavaScript files. These maps are generated throughout the JavaScript build pipeline, which includes transpilation, bundling, and minification, and they preserve the connection to the original code at each stage. Source maps use a JSON format with fields like version, file, sources, and mappings, where the mapping data is encoded using Variable Length Quantity (VLQ) encoding to efficiently store position data by representing differences between positions with small deltas. This allows for compact storage of large minified file coordinates as smaller encoded strings. The mapping strings use commas and semicolons to structure lines and segments, where each segment defines the mapping from a position in the generated file to a position in the source file. Understanding these mappings and the encoding process provides insight into how source maps maintain the integrity of the original code during debugging, with future integration into tools like parca-agent and Polar Signals Cloud promising to extend this functionality to performance profiling workflows.