# Recover and Troubleshoot a Deployment

Keep the private operator state, signed recovery manifests, independently trusted recovery keys, expected deployment/generation IDs and storage access outside the failure domain of the application host.

## Common failures

| Symptom                             | Next check                                                                                                                        |
| ----------------------------------- | --------------------------------------------------------------------------------------------------------------------------------- |
| Prerequisite check fails            | Resolve the reported host, virtualization, storage or tool requirement before installation.                                       |
| Installation key expires            | Ask the issuing administrator for replacement authority; preserve the same pending enrollment intent.                             |
| Connection or licensing check fails | Check the intended control-plane endpoint, trust, network and current entitlement. A cached receipt does not override revocation. |
| Operation was interrupted           | Read `status`; retain its operation ID and checkpoints, then retry the same supported intent.                                     |
| `inspection_required`               | Keep the host and journal intact and request administrator review; do not erase the checkpoint.                                   |
| Gateway stays closed after update   | Resolve application verification before reopening traffic.                                                                        |

## Inspect trusted recovery metadata

```sh
redbeltctl --state-dir /var/lib/redbelt restore plan   --manifest /private/recovery-manifest.json   --public-key /private/recovery-public-key.txt   --deployment-id DEPLOYMENT_UUID   --generation-id GENERATION_UUID
```

Supply identifiers and the public key from independent recovery custody. This verifies signed metadata and reports the required components. It does **not** read back storage, restore application data or start services.

## Stage and inspect exports

`restore stage` takes those same arguments plus `--storage /private/storage.json` and `--staging-dir /private/recovery-staging`. It retrieves the signed snapshot and checks export hashes, sizes and ownership. Use a new or already owned staging destination. Staging is not an application restore.

`restore operator` takes the same trusted manifest/key/ID arguments plus `--exports-directory /private/recovery-staging` and `--output-directory /private/recovered-operator`. It restores configuration and identity into an inactive protected destination. It does not activate credentials or replace the running deployment.

## Restoring the application

Only use the recovery procedure certified for your exact release and deployment. Do not infer a full database/object/vector/workspace restore from `manifest-verified`, successful staging or an inactive operator export. If the supplied release has no certified activation procedure, stop after preserving evidence and contact Redbelt support.

Rolling application images back does not roll a database migration back. A migration, failed restore or changed generation requires its explicit compatibility and recovery rules. Never edit Flyway history, substitute deployment IDs or delete an unfinished journal to bypass those rules.
