Back up instance state

Create a recoverable copy of configuration, run history, and attempt recovery metadata.

Prerequisites

Access to the instance folder, an existing backup directory on persistent storage, and a new destination file name. The database backup can run while the orchestrator is active.

Procedure

1. From the instance folder, create a backup:

   usine backup /backups/instance.sqlite

2. Confirm the command reports both instance.sqlite and instance.durable.sqlite. An instance without durable storage reports only its runtime backup.

3. Copy .env, .usine/github-app.json, .usine/instance-id, .usine/attempt-connections/, .usine/known-hosts/, and .usine/ssh-workers/ to protected backup storage. Include referenced SSH keys and parent credentials through your credential backup process.

4. If you need to preserve active local checkouts, stop the orchestrator before copying .usine/workers/. Keep the snapshot and credential records together.

Check the result

The backup files exist with owner-only access and the command did not overwrite an existing backup. The runtime database includes configuration revisions, events, runs, and transcripts; the durable database includes Pi parent tasks. The runtime backup omits unused database pages, including pages freed when the orchestrator compresses ended attempts’ transcripts. Both compressed transcript segments and late messages are included. The command syncs the runtime backup to disk before publishing it and syncs the destination directory before reporting success.

An export is useful for configuration review, but excludes history and pause state. Restore instance state describes restoring the databases. Transcripts can contain repository content and tool output; protect them with the same access controls as source code.