Using Berkelium

Berkelium offers multiple interfaces tailored to different scientific workflows.

1. The Web User Portal

The Berkelium User Portal is the fastest way to:

Monitor live resource quota consumption (CPU Cores, RAM, NVIDIA GPUs, Ceph storage).
Provision new namespaces for grants and research groups.
Download customized kubeconfig connection files.
Inspect active pod statuses and decommission stale namespaces.

2. The `kubectl` Command-Line Interface (CLI)

The standard tool for interacting with the Kubernetes API server.

# Check current namespace and cluster context
kubectl config current-context

# List running pods in your scientific namespace
kubectl get pods -n sci-myproject

# View real-time resource utilization
kubectl top pods -n sci-myproject

# Describe pod events and errors
kubectl describe pod <pod-name> -n sci-myproject

3. Helm & Kustomize

Science IT provides pre-packaged Helm charts for common scientific software stacks:

JupyterHub / Dask Gateway
RayCluster (KubeRay)
PyTorch Training Operator
PostgreSQL / TimescaleDB Clusters
MLflow Model Registry

Example launching a Ray Cluster:

helm repo add kuberay https://ray-project.github.io/kuberay-helm/
helm install my-ray-cluster kuberay/ray-cluster \
  --namespace sci-myproject \
  --set head.resources.limits.cpu=8 \
  --set worker.replicas=4 \
  --set worker.resources.limits."nvidia\.com/gpu"=1

4. Headless & Automated CI/CD

For automated pipelines (e.g. beamline data ingestion triggering upon detector acquisition):

Science IT supports Kubernetes Service Accounts with tightly scoped RBAC role bindings.
Submit a consultation ticket at Science IT Support to obtain a machine robot token for your automated acquisition server.