Requirements

System Requirements

All resource specifications are approximate and will likely vary with your specific use case.

  • Virtual Network x 1

    • Private Subnet x 2 (recommended at least 50 IP addresses per subnet)

    • Public Subnet x 2 (recommended at least 10 IP addresses per subnet)

      • (can optionally be left private)

  • Object Storage x 1

    • (can add more as needed)

  • Virtual Machine x 7* (recommended with auto-scaling for up to at least 20 instances)

    • 6 for control plane services

      • recommended at least 16 GiB RAM (such as the AWS t3.xlarge template) across 2 autoscaling groups (3 minimum, 3 desired, 5 maximum)

    • 1 for data plane services

      • recommended at least 64 GiB RAM (such as the AWS m5.4xlarge template) in an autoscaling Group (1 minimum, 1 desired, 10 maximum)

  • Network Load Balancer* x 1

    • (NOTE: this is auto-provisioned by the Ingress LoadBalancer service)

  • TLS Certificate x 1

  • Domain x 1

External Dependencies

Connectivity to these external dependencies is needed for various components, such as DNS services for routing client connections to load balancer endpoint(s).

Cloud Provider Services

  • Kubernetes Services

  • Object Storage

    • Default storage medium for datasets and model artifacts

  • IAM Services

    • Manage privileges for internal services, which enable:

      • Reading and writing data from object storage

      • Automatically scaling cluster node groups

      • Automatically configuring DNS records for external access

      • Building images and pushing them to ECR (AWS only)

  • DNS Services

    • Manage records for external access to the API layer, validated by TLS certificates

  • Certificate Manager (AWS only)

    • Manage TLS certificates

  • Secrets Manager (AWS only)

    • Used during installation to load setup credentials

3rd Party Services

  • Robust Intelligence Container Registry

    • Retrieve Robust Intelligence Docker images (can be mirrored to your private registry if desired)

  • Velero (recommended)

    • Provides backup and disaster recovery

  • DataDog (recommended)

    • Provides cluster health monitoring

  • Prometheus Node Exporter (recommended)

    • Provides node health monitoring

  • OIDC Authentication (optional)

    • Integrates with your identity provider

  • SMTP Service (optional)

    • Integrates with your SMTP server to send email notifications for model testing