We've made significant performance improvements to Runpod's automated GitHub integration, and we're excited to share the results. For those unfamiliar with our GitHub integration, it's designed to streamline the container deployment process. By connecting your GitHub repository to Runpod, you can automatically trigger container builds whenever you push changes to your codebase. This means less time spent on manual deployment steps and more time focused on what matters most: building great AI applications. However, there was a problem recently where this wasn't working as planned.‍

What Changed

‍Our engineering team identified and resolved a bottleneck in our container image upload pipeline. This was causing an problem where Github builds were proceeding at unacceptably slow speeds (if they finished at all, as this would cause the builds to butt up against our maximum build time and end up timing out.) After a thorough analysis of the build process, we rewrote key components of our registry image uploader to optimize how layers are transferred during the build process.‍

The Results

‍The numbers speak for themselves:

• Over 65% reduction in upload times for container images
• P98 upload performance improved from nearly 3 hours down to under an hour
• Layer uploads now running at multi-gigabit speeds

For developers using our GitHub integration to build and deploy container images, this means significantly faster iteration cycles and reduced wait times when pushing updates.

‍What This Means for You

‍If you've previously experienced slow build times when using Runpod's GitHub builder—particularly for larger images—you should see a noticeable improvement. No action is required on your end; these optimizations are already live.

‍We're Listening

‍Performance is an ongoing priority for us. If you encounter any issues with build times or the GitHub integration, please reach out to our support team. Your feedback helps us identify areas for continued improvement.

Author profile: Brendan McKeag