JupyterHub

JupyterHub provides multi-user notebook environments for data scientists and ML engineers, with configurable spawners for resource allocation.

Architecture

+------------------+     +------------------+
| JupyterHub Proxy |     | JupyterHub       |
| (Port 80/443)    |---->| (Spawner)        |
+------------------+     +------------------+
                                 |
                          +------+------+
                          | User Pods   |
                          | (Notebooks) |
                          +------------+

Spawner Configuration

Profile	CPU	Memory	GPU	Use Case
Small	2	4Gi	0	Data exploration
Medium	4	16Gi	0	Feature engineering
Large	8	32Gi	0	Model training (CPU)
GPU	4	16Gi	1	Deep learning

Authentication

JupyterHub authenticates users through the MATIH IAM service using OAuth2/OIDC:

auth:
  type: oauth2
  oauth2:
    authorizeUrl: "https://api.matih.ai/api/v1/auth/oauth2/authorize"
    tokenUrl: "https://api.matih.ai/api/v1/auth/oauth2/token"
    userInfoUrl: "https://api.matih.ai/api/v1/auth/oauth2/userinfo"

Pre-installed Packages

User notebook environments include pre-configured connections to all MATIH data infrastructure: Trino, Spark Connect, MLflow, Feast, and S3/MinIO.

Triton Overview