Production readiness
A practical checklist for releasing, operating, and recovering production applications on OpenShip.
Use this checklist before routing customer traffic. Record an owner and a verification result for each area. Requirements depend on the runtime, data sensitivity, availability target, and your organization's policies.
1. Choose an operating model
| Decision | Verify |
|---|---|
| Control plane | Self-hosted web, Cloud, or Desktop; identify who maintains it and when it is available. |
| Destination | Server, Cloud space, or supported cluster with capacity for the workload and release process. |
| Build placement | Enough build capacity; use a separate builder where supported to reduce contention with live workloads. |
| Automation | An available controller for scheduled jobs, backups, and webhook-triggered operations. |
For continuous team operations, use an always-on control plane. Desktop remains a useful local control plane; remote workloads run independently, but controller-managed automation depends on its availability. See runtime model.
2. Restrict access and protect secrets
- Review organization membership, server access, and permission boundaries.
- Use scoped credentials for automation. Rotate or revoke credentials when their owners or purpose change.
- Restrict a self-hosted dashboard to the intended network or authenticated public entry point.
- Keep private services private. Publish only required ports and routes.
- Store secrets in the supported environment/credential controls, not in the repository or build logs.
- Preserve control-plane encryption keys with your recovery material; a database copy without its keys may be insufficient.
Authentication · Isolation · Environment variables
3. Verify the release
- Confirm the exact source, branch, build commands, runtime configuration, and destination.
- Test the release in an appropriately isolated preview or staging environment.
- Confirm application health, database connectivity, and an actual user operation.
- Check the public hostname, DNS, HTTPS, and route protections.
- Define who can approve a cutover and how to return to the previous release.
A static 200 OK response proves only that the configured route answered. It does not prove application
or database health. Use a health endpoint that checks the dependencies appropriate to your service.
Preview environments · Custom domains · Route protection
4. Plan for persistent data
| Check | Evidence to retain |
|---|---|
| Persistence | Each stateful service uses the intended persistent volume or external storage. |
| Backup destination | Connection verification succeeded; copies survive loss of the workload's host. |
| Schedule and retention | Policy matches the acceptable data-loss window and storage budget. |
| Consistency | Database dump or application-aware hooks are used where a live file copy is insufficient. |
| Restore | A backup has been restored to a safe target and validated by the application. |
| Control-plane recovery | Installation data, credentials, encryption keys, and access procedures can be recovered. |
Document your recovery point and recovery time objectives. A successful backup job is not a tested restore. See backups and restore and control-plane migration.
5. Set capacity and monitoring
- Set appropriate CPU, memory, storage, and replica limits for the runtime.
- Leave capacity for builds, release replacement, retained rollback artifacts, and backups.
- Configure monitoring and logs and a notification path your team checks.
- Monitor service health, error rate, resource usage, disk growth, and backup failures.
- Check that a sleeping application matches its expected availability and latency requirements.
A cluster alone does not guarantee high availability. Verify replica placement, storage behavior, ingress, and failure recovery for the selected runtime. Stateful Swarm workloads and replacement operations have specific constraints; review clusters and scaling.
6. Rehearse recovery and maintenance
| Scenario | Prepare |
|---|---|
| Bad application release | Known-good artifact, available rollback capacity, and a tested rollback procedure. |
| Database migration fails | Application-compatible migration strategy and a separate data recovery plan. |
| Host or volume is lost | Backup destination, restore target, credentials, and DNS/route cutover plan. |
| Control plane is unavailable | Privileged recovery access and a protected copy of installation data. |
| Upgrade causes a problem | Release notes, pre-upgrade backup, maintenance window, and recovery procedure. |
Application rollback does not reverse schema changes or restore data. Deployments, data recovery, and control-plane recovery are separate procedures.
Rollback and redeploy · Database migrations · Updates
Record the handover
Keep the project owner, on-call contact, destination, domains, backup policy, last restore test, and recovery procedure in your team's operating documentation. Use the SLA for the published managed-service commitment; it is not a guarantee for self-hosted infrastructure or your application's own availability.