Stale GET /rate_limit GH API calls returning a frozen "fresh" window answer instead of real usage at some circumstances

[rastislav-bostik-tb]

🏷️ Discussion Type

Bug

💬 Feature/Topic Area

API

Body

What

It seems that GET /rate_limit GH API endpoint, at some circumstances, becomes returning a frozen "fresh" window answer instead of real usage.

What I'm seeing: some rate-limit keys: GET /rate_limit stops reflecting real usage — used: 0, reset pinned to now+3600s (+60s for search), every call, regardless of real spend. Ruled out client/edge caching, a gh CLI bug, different identity (below). Not the documented "stale replica" mode either — that gives an old nonzero value, not a perpetual fresh one. Could be missing something — happy to be corrected.

Reproduction

Two calls, same token, same second:

$ gh api "repos/octocat/Hello-World" -i | grep -Ei '^date:|x-ratelimit'
Date: Tue, 08 Sep 2026 12:07:43 GMT
X-Ratelimit-Limit: 5000
X-Ratelimit-Remaining: 4952
X-Ratelimit-Reset: 1788871522
X-Ratelimit-Resource: core
X-Ratelimit-Used: 48

$ gh api "/rate_limit" -i | grep -Ei '^date:|x-ratelimit'
Date: Tue, 08 Sep 2026 12:07:43 GMT
X-Ratelimit-Limit: 5000
X-Ratelimit-Remaining: 5000
X-Ratelimit-Reset: 1788872863
X-Ratelimit-Resource: core
X-Ratelimit-Used: 0

Same core resource, same instant, same token: one path shows 48 used with a live window, the other 0 used with a window pinned exactly 3600s from whenever you ask.

Ruled out

• Client/local caching: fresh gh/curl each time, no local HTTP cache.
• Edge/CDN caching: /rate_limit carries Cache-Control: no-cache. Two back-to-back calls in the same second return different X-Github-Request-Id — genuinely fresh backend hits, not replays.
• Different identity: both calls' X-Oauth-Client-Id identical.
• gh CLI bug: reproduces on gh 2.87.3 and 2.100.0, and with plain curl (no gh) through 4+ api.github.com front-end IPs.
• Not token-type-specific: reproduces on a personal OAuth token (gho_...) and, per evidence below, a repository's own GITHUB_TOKEN.

Proof set — two other orgs, same ~48h onset window

Identical signature (used: 0, reset=now+3600s) in public Actions job logs of two unrelated repos, starting the same 2-day window:

• tenstorrent/tt-metal, workflow _produce-data.yaml, job produce-cicd-data, authenticating with ${{ github.token }} (i.e. GITHUB_TOKEN):

  • Last live reading: run 33069873597 — used=4814 (2026-08-27 12:00 UTC)
  • Last live reading: run 33128216206 — used=4337 (2026-08-28 00:00 UTC)
  • First broken reading: run 33222127835 — used=0, reset=+3600s (2026-08-28 23:59 UTC)
  • Still broken: run 34137288079 — used=0, reset=+3600s (2026-09-07 15:14 UTC)

• cashapp/hermit-packages, workflow autoversion.yml, using a separate PAT-style secret (not GITHUB_TOKEN):

  • Last live: run 33099456511 — used=326 (2026-08-27 17:48 UTC)
  • First broken: run 33144000415 — used=0 (2026-08-28 05:20 UTC)
  • Still broken: run 34076988342 — used=0 (2026-09-07 02:43 UTC)

Both broke independently within the same ~48h window (2026-08-27 evening – 2026-08-28 night UTC), on two unrelated orgs' tokens — one GITHUB_TOKEN, one PAT-style.

Not universal: a third repo, centreon/centreon (check-status.yml, also GITHUB_TOKEN), shows a healthy, correctly-decrementing counter as of today — not every key, not every GITHUB_TOKEN. Looks key/shard-dependent, not token-type.

A guess at the mechanism (could easily be wrong)

Your own engineering blog ("How we scaled the GitHub API with a sharded, replicated rate limiter in Redis") describes keys sharded across Redis: one primary (writes) + read replicas per cluster; GET /rate_limit reads a replica, real headers reflect the write path. Same post documents a related past failure — a stale replica returning an old used value with a new reset. If that's still how it works, this looks like the read path finding no window at all and synthesizing a fresh one — different (and worse) than the documented failure. Guess from the outside, not confirmed.

Is this worth a look?

/rate_limit is the endpoint your docs recommend when you can't rely on response headers (gated jobs, dashboards). A subset of keys silently, durably reporting "fully fresh, nothing used" for 10+ days seemed worth flagging even if I've got part of this wrong.

Didn't dig much deeper than this — just reporting an observed behavior that looked wrong, not a deep investigation. Am I missing something?

Can't offer a strict MRE — seems per-key/per-shard, not reproducible on demand for any token, only whichever ones are affected. If you can reproduce it reliably (or can't at all), that itself tells me something.

[github-actions[bot]]

💬 Your Product Feedback Has Been Submitted 🎉

Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users.

Here's what you can expect moving forward ⏩

• Your input will be carefully reviewed and cataloged by members of our product teams. 
  • Due to the high volume of submissions, we may not always be able to provide individual responses.
  • Rest assured, your feedback will help chart our course for product improvements.
• Other users may engage with your post, sharing their own perspectives or experiences. 
• GitHub staff may reach out for further clarification or insight. 
  • We may 'Answer' your discussion if there is a current solution, workaround, or roadmap/changelog post related to the feedback.

Where to look to see what's shipping 👀

• Read the Changelog for real-time updates on the latest GitHub features, enhancements, and calls for feedback.
• Explore our Product Roadmap, which details upcoming major releases and initiatives.

What you can do in the meantime 💻

• Upvote and comment on other user feedback Discussions that resonate with you.
• Add more information at any point! Useful details include: use cases, relevant labels, desired outcomes, and any accompanying screenshots.

As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities.

Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐