Repository navigation
ci: recover owned order during overflow failover - #15618
teamleaderleo wants to merge 1 commit into
Conversation
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Repository: manaflow-ai/cmux/.coderabbit.yaml Review profile: ASSERTIVE Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
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. Comment |
f787d40 to
4c06d91
Compare
|
Superseded by #15620 after rebasing the dependent change onto the current picker branch. |
What changed
Follow up cmux#15614's outage routing fix for an explicitly Blacksmith-only
CI_PR_POOL_ORDER. During an activeCI_CLOUD_OVERFLOW_SAVEDrecord, the picker now rebuilds the owned order for the pinned PR Xcode when the configured order has no owned labels. It still refuses to emit a dead Blacksmith candidate and keeps the retry route owned.Validation
git diff --checkpass.This is a dependent follow-up on #15609; it contains only the small recovery correction.
Summary by cubic
Fixes owned pool routing during a cloud overflow outage so the picker no longer places or retries jobs onto a dead Blacksmith pool. Follow-up to cmux#15614 for an explicitly Blacksmith-only
CI_PR_POOL_ORDER.Bug Fixes
CI_CLOUD_OVERFLOW_SAVEDis present (and its recorded lane matches the liveMACOS_RUNNER_PR), the picker keeps normal owned placement but never emits a non-owned candidate or retry target.Written for commit 4c06d91. Summary will update on new commits.