Inside GitHub’s new Git storage architecture
GitHub details a new Git storage architecture designed to support 5,000 concurrent agents and 7,200 pushes per second on a single repository. The system maintains low-latency read performance while scaling to meet intense write demands.
GitHub · 07 November 2026 · Read the original →
How we halved DNS build delay and cut memory use by 85%
Cloudflare explains how they optimized their DNS propagation pipeline to handle a 3x increase in record changes. By changing how the Zone Builder processes specific event types, they reduced memory usage by 85% and halved build delays.
Cloudflare · 07 November 2026 · Read the original →
How we continuously validate formal specifications in production
Datadog introduces FETCh, a tool used to continuously validate production system behavior against formal TLA+ specifications. This approach allowed them to identify a subtle bug that had remained hidden despite conventional testing methods.
Datadog · 07 November 2026 · Read the original →
F3 Lightning: How Meta Accelerates Ads ML Features to Production at Scale
Meta's F3 Lightning framework evolves their machine learning feature authoring to handle requests requiring over 20,000 features. It addresses the bottleneck of using a single merged graph as a release unit, which previously required full recompilation for any change.
Meta · 07 November 2026 · Read the original →
How OrderServ achieved up to 64% cost savings with multi-tenant ECS
This case study describes how a food-ordering platform migrated from dedicated clusters to multi-tenant ECS to handle traffic surges. The architectural shift resulted in a 64% reduction in infrastructure costs while improving reliability.
AWS Architecture · 07 November 2026 · Read the original →
Safe and Efficient Multi-Epoch Embedding Training
Pinterest engineers discuss the infrastructure requirements for safe multi-epoch training of embeddings in recommendation systems. The post covers the trade-offs and safety mechanisms needed to maintain model stability and efficiency across multiple training cycles.
Pinterest · 07 November 2026 · Read the original →
Scaling Model Evaluation at Pinterest with Ray (Part 4)
This technical deep dive explains how Pinterest integrated the Ray framework into their machine learning pipeline for scaling model evaluation. It focuses on Ray's role in performing 'last mile' data processing efficiently within an existing ML infrastructure.
Pinterest · 07 November 2026 · Read the original →
Lots of talent deployed this week just to make the cloud bill less embarrassing.
7 stories, every Saturday
Published here every week. Follow by RSS to get it as it lands.