Rollback
Rollback is a protected change that restores the last known-good approved state. It is not an automatic promise that every provider operation is reversible: data migrations, managed control-plane upgrades, and external side effects require a service-specific recovery plan.
The implemented manual rollback workflow uses platform-apply, the apply
identity, the exact saved plan artifact from a supplied plan run, binding hash
verification, and typed ROLLBACK provider/mode/proposal-id. It has not run.
- declare the incident or failed deployment and freeze competing mutations;
- identify the last known-good proposal, commit, state version, and health evidence;
- verify provider, environment, state lock, and OIDC identity;
- obtain protected reviewer approval with self-review prohibited;
- run the rollback plan against the exact intended state and record its hash;
- verify health, security controls, logs, budgets, and drift; and
- close or escalate the incident with a new proposal for permanent repair.
AWS rollback must consider provider-managed dependencies, EKS node/control-plane compatibility, IAM and security-service continuity, and KMS/Secrets Manager availability. Azure rollback must consider AKS control-plane/node compatibility, Entra role propagation, Key Vault recovery, private DNS/endpoints, and Policy remediation state.