-
Notifications
You must be signed in to change notification settings - Fork 0
Deployment and Upgrades
Debuglet supports two separate shapes:
| Profile | Purpose | Network and credentials |
|---|---|---|
| Local TEST | Demo, development, and acceptance checks on one machine | Loopback-only; no credentials or TLS |
| Authenticated deployment | A dispatcher and remote executors on machines you operate | TLS, executor credentials, user authentication |
Use the Ansible deployment files for the authenticated profile. They consume operator-supplied inventory and secrets; inventories, private keys, generated certificates, and host-specific values do not belong in the repository.
- Build or download one versioned release payload and verify its checksums.
- Back up the dispatcher database and retain the previous package.
- Apply the explicit database migration for that release.
- Deploy and restart the dispatcher, then verify
/version. - Deploy executors, one host at a time when practical, and verify each reports ready.
- Submit a small TEST measurement and check its logs and terminal state.
Production executor deployment is intentionally manual. CI validates source and artifacts; a private environment with network access to the hosts should execute Ansible after a human approves the target and version.
Keep a known-good package and an offline pre-upgrade copy of each database and its existing -wal/-shm companions. Use the upgrade procedure from the target version. The current source upgrade playbook stages the selected package, checks state and free space, stops the service, and takes this backup before migration. It activates the new executable only after migration succeeds. If migration fails, leave the service stopped; earlier migrations may already have committed. Diagnose the reported version and retry with the same candidate, or restore the complete offline backup before restarting the previous package. Never combine an old database with newer WAL companions. Foreground local-role state is package-bound; changing its version metadata is not an upgrade procedure. Do not point an older daemon at a newer database unless its release documentation explicitly supports the downgrade. An interrupted executor run is quarantined rather than resumed; inspect its state and logs after the executor returns.
See deployment inputs, state and upgrade reference, and systemd deployment.
For port, executor connection, OAuth, readiness or state-version failures, start with the troubleshooting guide for the package you installed.