Skip to main content

Datalayer on Kubernetes

This section details the steps to take to deploy the Datalayer services on a Kubernetes (K8S) cluster.

The following pre-requisites are needed to deploy the Datalayer services.

☸️ A Kubernetes cluster with administrator access.

🐳 Docker Images available in a Docker Repository.

⚓ Helm Charts available in a OCI Repository.

🫧 Configuration and Environment Variables.

🔒 Secrets available in a Vault.

title: Cluster​

Cluster

Clouder creates and manages Kubernetes clusters using Kubeadm on cloud VMs, with built-in support for CRIU checkpoint/restore. This enables snapshotting running pods (e.g., Jupyter notebook sessions) and restoring them later on the same or different hardware.

Architecture​

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   Azure      │     │    OVH       │     │  Bare Metal  │
│   VMs        │     │    VMs       │     │  Servers     │
└──────┬───────┘     └──────┬───────┘     └──────┬───────┘
       │                    │                    │
       └────────────────────┼────────────────────┘
                            │
                 ┌──────────▼──────────┐
                 │     Kubeadm         │
                 │  (cluster bootstrap)│
                 └──────────┬──────────┘
                            │
                 ┌──────────▼──────────┐
                 │   Kubernetes +      │
                 │   CRIU enabled      │
                 └─────────────────────┘

Quick Start​

# Install Clouder
pip install clouder

# Check version and available commands
clouder --version
clouder --help

Context Management​

# Initialize a new context
clouder ctx init

# List and select contexts
clouder ctx ls
clouder ctx show
clouder ctx set azure <context-id>
clouder ctx set ovh <context-id>

Cloud Provider Setup​

Configure your cloud provider credentials:

# Azure (interactive setup)
clouder azure configure

# View Azure configuration
clouder azure

See the Azure Setup guide for detailed instructions.

Cluster Operations​

# Create a cluster
export CLUSTER_NAME=my-cluster
clouder k8s create $CLUSTER_NAME

# Get kubeconfig
clouder k8s kubeconfig $CLUSTER_NAME

# Set as active cluster
clouder k8s use $CLUSTER_NAME

# List clusters
clouder k8s ls

Node Pools​

# Create worker node pools
clouder k8s create-nodepool $CLUSTER_NAME \
    system --flavor Standard_D4s_v5 \
    --min 1 --desired 2 --max 10 \
    --roles system --xpu cpu

# Create GPU node pool for Jupyter
clouder k8s create-nodepool $CLUSTER_NAME \
    jupyter-gpu --flavor Standard_NC6s_v3 \
    --min 0 --desired 1 --max 5 \
    --roles jupyter --xpu gpu-cuda

# Scale a node pool
clouder k8s update-nodepool $CLUSTER_NAME \
    jupyter-gpu --min 0 --desired 0 --max 5

Virtual Machines​

# Azure VMs
clouder azure vm-ls
clouder azure vm-create
clouder azure vm-delete <name> --resource-group <rg>

# OVH VMs
clouder vm ls
clouder vm create

Storage and SSH​

# S3 buckets
clouder s3 create my-bucket
clouder s3 ls

# SSH keys
clouder ssh-key ls
clouder ssh-key create my-key

Operator and Server​

# Start/stop the Clouder Kubernetes operator
clouder operator start
clouder operator stop

# Start the Clouder Jupyter server extension
clouder server

System Info​

# Current context info
clouder info ctx
clouder info me

Documentation​

SectionDescription
Azure SetupConfigure Azure credentials, create VMs, manage resources
Clouder K8SHow Clouder uses Kubeadm to bootstrap Kubernetes clusters
CRIUCheckpoint and restore pods with CRIU
CLI ReferenceFull command reference for all Clouder CLI commands

Once you've completed the guides, you'll possess the fundamental knowledge needed to configure and oversee the services enabling Jupyter clients to scale their data analysis.