Theory

Updated

October 30, 2026

Theory: Git Storage is Packs Plus a Database of Metadata

Repo growth hurts fetch time, gc, and inode counts. The forge DB (users, issues) scales on a different axis than packfiles.

Invariants

  1. Local disk vs NFS vs Ceph vs S3 changes latency of git fetch.
  2. Gitaly-style git servers shard repos, not tables.
  3. CDNs help clone of public objects; auth still hits origin.
packfiles / alternates / object storage
metadata DB  (issues, users)   separate scale plan

What can break

  • NFS for lots of tiny Git objects without cache.

How to read this chapter

Growth, storage options, fetch, DB, CDN/geo, benchmarks.