Google Professional Cloud Architect: Shared VPC Architecture Patterns
Shared VPC lets teams deploy resources in separate Google Cloud projects while using network resources centrally controlled from a host project. It is an organizational and security design pattern: application teams can own their service projects, while network administrators manage approved subnetworks, routes and firewall policy. For the Google Professional Cloud Architect exam, the challenge is deciding when centralized networking supports autonomy and when it becomes a bottleneck or unwanted shared dependency.
On this page
- Separate the host project’s responsibilities from service projects
- Grant the minimum subnet access that teams need
- Design CIDR allocation for future environments
- Keep connectivity and security controls reviewable
- Account for ownership changes and operational failure modes
- Test the model with a realistic deployment
A service project does not automatically receive unrestricted access to every resource in the host project. It must be attached and granted appropriate permissions. Designing these boundaries explicitly is more important than memorizing a diagram that places every environment inside one giant VPC.
Separate the host project’s responsibilities from service projects
The host project owns the Shared VPC network and its network-level controls. Service projects contain workloads such as eligible Compute Engine resources using subnetworks from that network. An organization administrator or authorized Shared VPC administrator establishes the relationship and delegates access. This architecture can make it possible for multiple business units to work independently without duplicating corporate networking policy.
Project boundaries still matter for billing, resource administration, quotas and audit records. The service project’s workload operator should not necessarily control routes or firewall policy. The host project’s network owner should not need permission to deploy every application’s code. A clear separation of these roles reduces both accidental outages and administrative escalation requests.
Grant the minimum subnet access that teams need
A service-project administrator typically needs the appropriate Network User permission for subnetworks it is allowed to use, as well as permissions for creating the actual workloads in the service project. A broad project-level grant may expose more subnet choices than necessary; narrower subnetwork permissions can reinforce environment and trust separation. Validate IAM at the actual resource scope rather than assuming a project Owner title is required.
Consider a payment application and an internal analytics service that both need shared corporate DNS. They may use different subnetworks with distinct firewall treatment and permitted identities. A single shared network does not imply that every workload should be able to contact every other workload by default.
Design CIDR allocation for future environments
Central network management should assign nonoverlapping address blocks with room for growth. A VPC whose subnet ranges were selected arbitrarily can become difficult to connect to existing on-premises networks or acquisitions. Plan for regional subnet expansion, workload clusters, private service connectivity and address consumption under scaling events.
Document which subnetworks are shared with which service projects. For delegated teams, predictable names and clear location boundaries are useful. Avoid creating an application dependency on a particular subnet simply because it happens to have spare addresses today.
Keep connectivity and security controls reviewable
Network administrators can centrally apply routing and firewall controls. Workload owners should still know which inbound and outbound flows their application requires and how service identities govern access. The network becomes easier to audit when firewall rules reflect service purpose and sources, rather than wide-open allowances for all address ranges in the organization.
Shared VPC is not the same feature as VPC Network Peering, Private Service Connect or a hub-and-spoke routing product. Those solve different network connectivity problems. Choose Shared VPC for common VPC resources across projects in the applicable organization structure; choose other connectivity constructs when independent VPCs need to communicate or access managed services under specific controls.
Account for ownership changes and operational failure modes
Who can change the host network in an incident? A global administrative mistake in the host project can affect many service-project workloads at once. Establish change review, policy as code where appropriate, route and firewall backups, monitoring and an emergency rollback process. Centralization reduces duplicated work but concentrates change impact.
Project migration needs planning, too. Moving a workload into a different service project is not equivalent to copying its VM identifier; dependencies on identities, subnet permissions and service policies must be reconstructed. During organizational restructuring, check whether participating projects still satisfy the Shared VPC organization and folder requirements.
Test the model with a realistic deployment
Before scaling to dozens of teams, validate one complete service lifecycle: an application team creates an allowed resource on an approved subnet, can reach permitted dependencies, is blocked from restricted segments and can view enough network logs to diagnose failures. A failed deployment might indicate missing compute.networkUser permissions, wrong project association or a subnet in the wrong region.
The examination’s architectural choice is about division of responsibility. Shared VPC fits when centralized network governance and separate service-project autonomy are both required. A good answer also addresses least privilege, address planning, incident blast radius and the mechanisms by which teams prove their connectivity works.