Skip exhausted Claude accounts during failover

[lawrencecchen]

Summary

• do not retry Claude OAuth accounts with an active request-time exhaustion mark
• keep low-headroom but non-exhausted accounts eligible for failover
• bounds the no-quota case to the current upstream response instead of retrying every known-exhausted profile

Tests

• go test ./...

Follow-up to #40: live Anthropic headers prove the remaining 429s are real quota exhaustion on 7d_oi, but the retry path was still re-probing accounts already marked exhausted.

---

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.}

---

Summary by cubic

Skip OAuth accounts marked exhausted during Claude failover. Prevents retry loops and returns the original 429 when no non‑exhausted alternates are available.

• Bug Fixes
  • Exclude scheduler-marked exhausted accounts from retry candidates; keep low-headroom but non-exhausted accounts eligible.
  • Return clearer error: "no untried non-exhausted OAuth accounts available".
  • Add regression test to verify exhausted alternates are skipped and only one upstream call is made.

^{Written for commit dfc7e49. Summary will update on new commits.}

[coderabbitai]

Warning

Review limit reached

@lawrencecchen, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 53 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 751d2208-a47f-4787-8d20-1b0ed9d12840

📥 Commits

Reviewing files that changed from the base of the PR and between 6878d9a and dfc7e49.

📒 Files selected for processing (2)

• internal/proxy/claude_ratelimit_routing_test.go
• internal/proxy/proxy.go

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/claude-oauth-app-quota

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: dfc7e49297

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Do not drop stale-exhausted failover candidates

When an alternate account has a zero scheduler score from stale usage data but no active request-time exhaustion mark, this continue removes it from failover entirely. In that scenario a Claude request whose current account returns a rejected 429 will return the 429 to the client with no_alternate_account, even if the alternate token is valid and upstream would return 200; the code below still documents that these apparently exhausted accounts should be tried because scores can be stale. Please distinguish actively marked-exhausted accounts from stale zero scores, or keep the previous bounded probe behavior for score-only exhaustion.

Useful? React with 👍 / 👎.