GitHub was down

Original Article ↗ Hacker News Discussion ↗

Outage scope and impact

  • Large, global outage: users report failures across North America, Europe, Australia, New Zealand, Hong Kong, Dominican Republic, etc.
  • Not just the web UI: git operations (push/pull), API, mobile app, GitHub Pages, Actions, Packages, Issues, Copilot, and Pages-hosted sites all affected.
  • Some users continued working locally or exchanged patches via git format-patch, git am, or git bundle. Others simply stopped work or used it as an excuse to end the day.

Status page and communication

  • Status page initially showed all green for several minutes while most users saw “angry unicorn” errors.
  • Many criticize this lag as making the status page “effectively useless” during real incidents.
  • Others note that external status dashboards are usually manually updated, internally approved, and influenced by SLA/liability concerns, so delays are expected.
  • Some argue for automated, faster, end‑to‑end monitoring; others point out false positives, complex routing, and “not 100% fidelity” as reasons for caution.
  • GitHub’s X/Twitter status feed is also criticized as hard to use when not logged in.

Suspected cause

  • Official status later attributes the incident to a “database infrastructure related change” being rolled back.
  • Users joke about misapplied configuration changes, database hot patches without transactions, and kubernetes rollbacks, but details remain unclear beyond the DB-change note.

Centralization, reliability, and alternatives

  • The outage highlights heavy industry dependence on GitHub as code host, CDN for artifacts, package registry backend, CI/CD, and Pages host.
  • Some view GitHub’s recurring incidents as evidence of declining reliability post‑acquisition; others note outages are inevitable at scale.
  • Several advocate mirroring to GitLab, Gitea, Codeberg, self‑hosted forges, or keeping offline mirrors; some report excellent uptime with self-hosted Gitea or GitLab.
  • Others remind that git itself is distributed and works fine locally; the real problem is centralizing collaboration, CI, and distribution on one service.

UI, culture, and reactions

  • The whimsical “unicorn” error page divides opinion: some find it unprofessional during serious outages, others prefer it to sterile corporate language.
  • Thread repeatedly uses HN itself as an ad‑hoc “real” status page, praised for speed but acknowledged as informal.