Symptom

READY=False, REASON=BudgetExhausted (or, for an already-deployed StageSet, READY=True with a status.budgetFreeze and the message prefixed Deployed;). status.budgetFreeze records the last observed remaining budget, the freeze and resume thresholds, and when the freeze began.

Cause

This is not a failure — it is the error-budget freeze working as configured. The metric source named in spec.errorBudget.source returned a remaining-budget value below freezeThreshold, so new-revision rollouts are held until the budget recovers. This is the Google SRE error-budget policy: when the budget is spent, stop shipping feature changes until reliability recovers.

The freeze holds only new-revision rollouts. Drift on the current revision keeps being corrected — a frozen service still has its declared state enforced — and the freeze self-resumes once the remaining budget reaches resumeThreshold.

Diagnosis

kubectl --namespace <namespace> get stageset <name> --output jsonpath='{.status.budgetFreeze}'
kubectl --namespace <namespace> get stageset <name> --output jsonpath='{.spec.errorBudget}'

Compare status.budgetFreeze.remaining against freezeThreshold / resumeThreshold. Confirm the source query returns what you expect — a value that never matches reality points at a wrong query (the stageset_budget_remaining metric carries the same number). A calendar budget source snaps back to full at the period boundary and unfreezes instantly; a rolling source recovers gradually.

Remediation

Usually none — the rollout resumes on its own once the budget recovers. To ship a reliability or security fix while still out of budget (the policy’s explicit exemption), break the glass once:

stagesetctl reconcile <name> --namespace <namespace> --budget-override

That applies the held rollout a single time without disabling the gate. To change the policy, adjust freezeThreshold / resumeThreshold, or set spec.errorBudget.dryRun: true to observe what would freeze without holding anything while you tune the query.