Deploy and operate
Plan, deploy, administer, troubleshoot, upgrade, and recover BullSequana AI.
Deployment is the beginning of the service lifecycle, not the end. This area connects installation guidance with the administrative and operational work required to keep BullSequana AI reliable after rollout.
Recommended path
First production deployment
Lifecycle
1. Plan
Decide where the platform will run, who owns each external dependency, and which identity, DNS, certificate, registry, storage, and network services must exist before installation. Record the acceptance criteria and rollback boundary before making changes.
Start with Before you deploy.
2. Prepare
Validate cluster access, storage, registry delivery, DNS, trust, identity, and hardware capacity. Treat prerequisites as testable interfaces between the platform and the target environment.
Use the prerequisites, minimum requirements, and the relevant environment guide.
3. Deploy and verify
Apply configuration through the supported GitOps and platform tooling path. Verify reconciliation, ingress, identity, storage, observability, and the first AI request before handing the installation to users.
Use the deployment sequence, GitOps workflow, and Platform CLI.
4. Administer and observe
Separate platform-wide, tenant-wide, team-level, and service-owner responsibilities. Establish dashboards, alerts, logs, traces, audit evidence, capacity reviews, and a repeatable incident path.
Hand each new tenant over with its limits in place. Create it in Administration → Tenants, set its Quotas and Deployment policy, enable the services it needs, such as private MLflow and inference, and assign its tenant administrators on Administrators. See Provision a tenant and Set a tenant's quotas. On installations with GPUs, enable GPUs and decide whether to share them before tenants deploy models.
Use Administration and operations, Foundation observability, and Troubleshooting.
5. Upgrade and recover
Before an upgrade, confirm compatibility, backup scope, validation steps, and the rollback decision point. Recovery plans must include platform configuration, stateful services, external dependencies, and the evidence needed to know that service has been restored.
Use the upgrade process and backup and disaster recovery.
Responsibility map
| Role | Primary responsibility |
|---|---|
| Infrastructure owner | hardware, network, storage, Kubernetes, and cloud dependencies |
| Platform engineer | platform configuration, reconciliation, health, capacity, upgrades, and recovery |
| Platform administrator | installation-wide settings, tenant lifecycle, tenant quotas (CPU, memory, storage, guaranteed GPUs and GPU limit), deployment policies, and monthly BCU allowances |
| Tenant administrator | federation, groups, members, service accounts, model access, monthly BCU caps, and delegated access inside one tenant |
| Security owner | access review, isolation requirements, evidence, sovereignty, and compliance |
| Service owner | onboarding, entitlements, adoption, consumption, and service outcomes |
For a step-by-step first deployment, use the deploy and operate guided path.