mirror of
https://github.com/coder/coder.git
synced 2026-06-03 04:58:23 +00:00
Jason Barnett
da6e708bd2
<!-- If you have used AI to produce some or all of this PR, please ensure you have read our [AI Contribution guidelines](https://coder.com/docs/about/contributing/AI_CONTRIBUTING) before submitting. --> Fixes https://github.com/coder/coder/issues/17069 Builds on #24332 and #24334 which addressed token persistence and rate limit handling. ## Problem When multiple concurrent requests race to refresh an expiring external auth token, providers with single-use refresh tokens (e.g., GitHub Apps) reject all but the first refresh attempt with `bad_refresh_token`. The losing request caches this transient error in the `oauth_refresh_failure_reason` database column and clears the refresh token, blocking all subsequent refresh attempts until the user manually re-authenticates. This is common for users with multiple terminals, IDE connections, or workspaces open, all of which poll the external auth endpoint and trigger concurrent refreshes when the token nears expiry. Database analysis showed 5 of 7 affected users failed within 5-10 seconds of token expiry, matching the Go oauth2 library's `expiryDelta` window. ## Fix Before caching a `bad_refresh_token` failure, re-read the external auth link from the database. If the refresh token has changed (indicating a concurrent caller already refreshed successfully), return the winner's updated link instead of writing a failure. An empty-string guard ensures a token cleared by another loser isn't mistaken for a winner's successful refresh. Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Garrett Delfosse <garrett@coder.com>