Apps & Software

Scaling GitHub's Git Infrastructure for an Agent-Driven Future

· GlobeWire Daily Editorial Desk · Source: github.blog

2 min read

Key takeawaysTo manage exponential growth in Git traffic, GitHub is migrating repositories to Azure while evolving its internal storage architecture. Physical constraints in data centers and regional power capacity necessitated a swift migration to cloud infrastructure, which began in August 2025. Currently, most repositories distribute their replicas across both…
Scaling GitHub's Git Infrastructure for an Agent-Driven Future

Image: Eleanor White, Victoria Dye · Photo source · Resized and letterboxed to 16:9; no AI generation

Addressing Infrastructure Constraints

To manage exponential growth in Git traffic, GitHub is migrating repositories to Azure while evolving its internal storage architecture. Physical constraints in data centers and regional power capacity necessitated a swift migration to cloud infrastructure, which began in August 2025. Currently, most repositories distribute their replicas across both GitHub's data centers and Azure to balance availability and latency.

Optimizing Write Throughput

Moving replicas across regions introduces latency, specifically a 25ms round-trip delay. Under the existing architecture, this latency limits write throughput because the system uses repository-wide locks, forcing writes to queue behind the slowest healthy replica. To address this, GitHub is shifting from repository-level serialization to per-reference serialization.

  • Per-reference serialization: Only writes contending on the same reference are queued, allowing independent updates to proceed concurrently.
  • Result: Internal testing demonstrated a 3–4x improvement in write throughput with lower latency.

Decoupling Read and Write Operations

GitHub is transitioning toward a decoupled model where storage is separated from compute. This involves the introduction of virtual replicas:

  1. Virtual Write Replica: Acts as a bridge to Azure Blob Storage, ensuring writes are durable without requiring read-heavy compute nodes.
  2. Virtual Read Replica: Allows for elastic scaling of read operations, which can be instantiated on demand without joining the core consensus quorum.

This shift allows GitHub to scale read and write resources independently, accommodating the heavy read-skew typical of most repositories. Future developments will focus on further optimizing write throughput and refining commit protocols to support the next generation of repository growth.

For more details, engineers can follow updates on the official GitHub Engineering blog.

Share this article
Read the original source: The bridge to agent-scale Git infrastructure

AI-assisted reporting; source claims are attributed and have not been independently verified.

About the author

GlobeWire Daily Editorial Desk — AI-assisted news based on attributed published sources.

More to read