Bitbucket Pull Request Diff Performance

5 posts · Dec 2011 – May 2022

What this trail covers

Bitbucket Cloud is enhancing its pull request experience with performance improvements in diff rendering and merge execution times. This involves optimizing API response times for diffs and diffstats, reducing timeouts, and streamlining backend hooks and locking mechanisms for merges. The platform is also addressing issues with background task scheduling and execution on the new cloud infrastructure to ensure faster and more reliable operations. This post details improvements to the file system

All Atlassian trails

May 2022Improving performance on complex diffsThis post details the technical shift from a '3-way' diff algorithm to a 'three-dot' (2-way) diff algorithm in Bitbucket Cloud's pull request functionality. It explains the algorithmic difference, the performance benefits, and how users can still view destination branch changes by syncing. It also outlines the deprecation of the 'merge' parameter and introduction of the 'topic' parameter in the Diff API to control the diff behavior.
Aug 2021An exciting update on diff and merge performanceThis post details specific optimizations implemented to improve Bitbucket Cloud's diff response times by 40% and reduce timeouts by an order of magnitude through in-memory processing. It also outlines improvements to pull request merge execution times, including streamlining pre- and post-receive hooks (e.g., removing stale ref lock file cleanup from synchronous flow, moving ref packing to a background hook), reducing aggressive locking by refactoring merge code to leverage Git's built-in consistency mechanisms, and fixing a bug in the background task scheduling library that caused worker idleness during peak traffic, leading to wall times tracking closer to execution times.
Apr 2021Extinguishing our performance fires and rebuilding for the futureThis post details significant performance improvements in Bitbucket Cloud, primarily by addressing bottlenecks in the file system layer. Key contributions include: 1. Caching packfiles for Git operations (clones, fetches) to reduce file system I/O by nearly half. 2. Implementing a custom libgit2 ODB backend with a high-performance caching layer between libgit2 and the file system, reducing throughput by over 30% for website and API traffic. 3. Optimizing diff caching by consolidating ad-hoc caching strategies, leveraging known access patterns to warm caches, and improving cache hit rates for APIs like pull request conflicts to nearly 100%. The post also outlines future investments in formalizing performance SLOs, transitioning to a new team ownership model for performance metrics, and building automation for guardrails and circuit breakers.
Jan 2019Exclude files from the new pull request experienceIntroduced the ability for repository administrators to exclude files from pull request diffs using pattern matching. This feature aims to improve code review experience by reducing noise from generated files (e.g., lock files, minified JS/CSS) and enhancing browser performance for large pull requests. Excluded files are replaced with an informational card linking to the file's diff.
Dec 2011Pull requests with side-by-side diffsIntroduced side-by-side diff view for pull requests, compare view, and individual commits, allowing for synchronized scrolling and easier code review. Also added expandable context to unified diffs, allowing users to click skipped sections to reveal more lines of context.
Advertisement