Infrastructure lock failure recovery
Authoritative source: AetherCloud. This page is mirrored into the unified AetherIoT documentation.
Use this runbook after State-lock acquisition times out, lock release cannot be proved, or a Plan result lacks evidence for the exact Deployment Stack State key.
Implemented today
Section titled “Implemented today”- The Plan-only infrastructure engine requires an acquired lease for the exact State key and successful release evidence.
- Lock timeout and lock-release failure are typed failures.
- A successful OpenTofu process exit is rejected when application-level lock evidence is absent or mismatched.
- The worker passes argv without a shell, bounds output, restricts temporary files, and removes its workspace on every outcome.
Not implemented
Section titled “Not implemented”Production remote State, a distributed lock adapter, durable lock ownership, operator lock inspection, force-unlock, State repair, and a public Plan route remain planned. The current port has no Apply, Destroy, Import, or State-repair operation.
Safe response
Section titled “Safe response”- Record the Tenant, Project, provider connection, Deployment Stack, exact State key, Plan identity, lock holder evidence, and lease deadline.
- Treat missing acquisition or release evidence as a failed Plan even when the infrastructure process exited successfully.
- Never switch State keys, bypass locking, edit raw State, or expose saved Plan binaries or raw Plan JSON to an agent.
- A bounded retry is permissible only for an acquisition timeout when the lock backend proves the existing lease remains valid and the request retains the same idempotency identity.
- On release uncertainty, isolate the Deployment Stack from further Plan and future mutation until an operator verifies the authoritative backend.
- After recovery, create a fresh Plan from a verified State snapshot. Do not reuse an artifact produced under uncertain locking.
Escalate
Section titled “Escalate”Persistent acquisition timeout or any release failure requires an infrastructure operator. Automatic force-unlock is forbidden. The operator must identify the holder, backend, lease, and active worker before changing lock state; absent that evidence, the Stack remains isolated.
Read Plan infrastructure safely, multi-cloud fusion, and AI invariants.