GCP Deployment Architecture

Important: You can currently use the GCP deployment template only with Automation Suite 2023.10. We therefore recommend referring to the Automation Suite 2023.10 documentation.

This page offers insight into the deployment architecture on GCP, the required components, and all known limitations.

Architecture Diagram

Components

• Virtual network
  • A subnet where all nodes reside.
  • A NAT Gateway for outbound connectivity (a Cloud Nat resource attached to a Cloud Router);
  • Firewall Rules to secure subnet traffic.
• A DNS private zone needed for installation. For more details, check the Known limitations section.
• 3 Managed Instance Groups. You can choose the instance type for server, agent, and GPU agent nodes. Make sure to check Multi-node HA-ready production machine requirements and Single-node evaluation machine requirements to meet the hardware requirements. Each VM has a 128 GiB OS and 256 GiB cluster binaries and state disk. Server nodes have an additional 512GiB/2048GiB data disk, depending on whether the AI products are installed.
  • Server nodes (cluster control plane). Server nodes also run workloads.
  • Agent nodes. Designed to only run workloads (they have no control plane services). If the number of desired agent nodes is 0, an empty Managed Instance Group is created.
  • GPU nodes. Nodes used specifically for ML models that have video cards. If the number of desired GPU nodes is 0, no Managed Instance Group is created.
• Public load balancer used to balance HTTPS traffic from port 443 to nodes.
• 2 Internal load balancers and a Managed Instance Group needed for forwarding node registration requests. The VMs have the smallest instance size possible.
• Task Mining node deployed as a separate VM. Its instance type is n2-standard-32.
• Bastion instance used to access the other nodes. It has a public IP and SSH enabled.
• SQL Database Instance:
  • 8 cores and 32 GiB RAM
  • 1000 GiB disk size that can be manually extended
  • The databases are created by the installer
• Secret Manager used to store auto-generated credentials for SQL server, Automation Suite Platform, and ArgoCD console.

Known Limitations

DNS

• Due to the fact that a DNS cannot be automatically attached to the LB:
  • The steps for configuring DNS can be completed only after the installation, while they are needed during installation. The private DNS zone solves this issue and can be safely deleted after the installation is completed. Alternatively, for testing purposes check the Step 4: Configuring the DNS.
  • Core DNS upstream servers must be forced to match the node’s nameservers. This could lead to the corruption of the rke config file ( /etc/rancher/rke2/config.yaml) at a VM restart, which can impact the post-installation process of upgrading.

RHEL

• Google may update the RHEL version without notice, thus leaving Automation Suite deployments outside of support. Currently, manual deployments using custom RHEL images are the only way to and stay in support.