Deploy and operate BullSequana AI
Move from environment decisions to a verified, supportable platform service.
This path is for platform engineers, infrastructure operators, deployment engineers, and SREs responsible for an installation.
Outcome
You can prepare a target environment, deploy through the supported configuration path, validate the platform, establish operational ownership, and recover or roll back safely.
1. Choose the deployment model
Choose a deployment model explains the supported infrastructure shapes and responsibility boundaries. Record which team owns Kubernetes, networking, DNS, certificates, registry delivery, storage, identity, and accelerators.
2. Establish environment context
Use Decisions to make first to capture the target domain, cluster, registry, network, identity, and access assumptions. Select the relevant environment guide before applying generic configuration.
3. Validate prerequisites
Complete the prerequisites as evidence, not as a reading exercise. Each required dependency should have an owner, a validation method, and an escalation path. Confirm minimum requirements against the intended workload rather than the cluster total alone.
For an installation with GPUs, plan Enable GPUs and decide whether tenants share cards through GPU sharing.
4. Configure and deploy
Understand the configuration model, then follow the deployment sequence. Use the supported GitOps workflow and Platform CLI so the desired state remains reviewable and repeatable.
5. Verify service readiness
Do not stop at healthy Kubernetes workloads. Verify:
- reconciliation completes without persistent failures
- ingress, DNS, certificates, and identity work from a client path
- storage and databases persist expected state
- an authorized user can reach the Portal
- an authorized application can make a BSQAI API request
- metrics, logs, traces, and dashboards are populated
- tenant isolation is preserved
6. Establish operations
Assign ownership for incidents, capacity, tenant administration, security review, upgrades, backups, and service communication. Use Administration and operations and Troubleshooting to turn technical checks into a repeatable operating model.
Hand over the first tenant with its limits in place:
- Create it in Administration → Tenants; see Provision a tenant.
- Set its Quotas: GPU allocation, compute quota, and monthly BCU allowance; see Set a tenant's quotas.
- Set its Deployment policy: the models, serving runtimes, and resource profiles it may deploy; see Set what a tenant may deploy.
- Enable the services it needs, such as private MLflow and inference.
- Assign its tenant administrators on Administrators.
7. Prepare change and recovery
Use Upgrade 1.4.0 to 2.0.0 for compatibility and rollout decisions. Use Backup and disaster recovery to define what is protected, how restoration is validated, and when rollback becomes recovery.
You are done when
- the deployment is reproducible from reviewed configuration
- user and application acceptance paths succeed
- monitoring and incident ownership are active
- the first tenant has quotas, a deployment policy, and tenant administrators
- capacity and backup assumptions are documented
- rollback and recovery have explicit decision points