Network Virtual Appliances and Policy-Based Routes in Google Cloud

GCP Study Hub
November 16, 2026

Network virtual appliances, such as third-party firewalls or inspection VMs, sit in the traffic path of a VPC network, and Google Cloud steers traffic through them with routes. Custom static routes can point at the appliances directly or at an internal passthrough Network Load Balancer in front of them, and policy-based routes can redirect matching packets to that load balancer before subnet routes apply. How the route points at the appliances decides whether the design is active-passive or active-active and whether a failed appliance is taken out of service automatically. Those distinctions are worth knowing well for the Professional Cloud Network Engineer exam.

Three ways to route through appliances

There are three ways to point custom static routes at a group of appliances, and each behaves differently.

One route per appliance, with different priorities. Only the appliance on the lowest-numbered route carries traffic. The others wait until that route stops being usable. This is an active-passive design, and the standby appliances sit idle.

Equal-priority routes, one per appliance. Traffic spreads across the appliances, but routes that name instances have no health checking. A hung appliance that is still running can keep receiving its share of traffic.

One route whose next hop is an internal passthrough Network Load Balancer. New connections are hashed across all healthy appliances behind the load balancer, so the design is active-active. The load balancer's health check removes a failed appliance from rotation, and the remaining appliances keep carrying traffic. This is the standard pattern for running appliances in a highly available setup.

Which load balancers can be a next hop

Only internal passthrough Network Load Balancers are supported as route next hops. Proxy load balancers, which only handle specific application protocols, and external load balancers, which are built for traffic arriving from outside Google Cloud, don't fill that role.

When an internal passthrough Network Load Balancer is the next hop, the route sends all supported protocols to the backends, including TCP, UDP and ICMP, as long as the forwarding rule uses TCP or UDP rather than L3_DEFAULT. One load balancer can therefore carry the full mix of traffic for the destination range.

Building the appliance tier for availability

A few choices around the load balancer complete a highly available design:

  • Use a managed instance group. Build the appliances from an instance template in a managed instance group, so failed appliances are recreated automatically and manual work stays low.
  • Spread the appliances across zones. Two appliances in one zone are lost together in a zonal outage.
  • Account for return traffic. If the appliances apply source NAT, replies return straight to the appliance that handled the connection, so no route is needed in the return network.
  • Leave clients unchanged. Because the route lives in the VPC network, client VMs keep sending traffic normally and need no configuration changes.

Policy-based routes

A policy-based route sends matching packets to an internal passthrough Network Load Balancer, and Google Cloud evaluates policy-based routes before subnet routes. That ordering is what makes them necessary in one common case. A static route can't take over traffic that matches a subnet route, so it can't steer traffic between two subnets of the same VPC network through an inspection appliance. A policy-based route can, because it is applied before the subnet route is considered.

Scoping a policy-based route

A policy-based route needs a careful scope. Limiting it to a network tag on the source VMs redirects only the VMs that carry that tag. Every other VM, including the appliance itself, still follows the subnet route. That lets the appliance forward inspected packets on to the destination subnet without the policy-based route sending them back to it.

A policy-based route without a tag reaches every VM in the network. That pushes unrelated workloads through inspection, and it also catches the appliance's own forwarded traffic, which then loops back to the load balancer. Tagging the source workloads, and leaving the appliance untagged, avoids both problems.

Policy-based routes only control the next hop for packets inside the VPC network. They are a tool for steering traffic within the network, not for changing how clients on the internet reach it.

Static routes or policy-based routes

For the Professional Cloud Network Engineer exam, the two tools divide along route selection. A custom static route can't take over traffic that matches a subnet route, so it suits destinations outside the network's own subnets, and pointing it at an internal passthrough Network Load Balancer adds health checking and active-active behavior. A policy-based route is evaluated before subnet routes, so it is the tool when the traffic to be redirected stays inside the network, and its scope deserves as much attention as its next hop.

Our Professional Cloud Network Engineer practice exams cover network virtual appliances and policy-based routes alongside VPC route selection and internal load balancing, with detailed explanations that drill these distinctions.

Get tips and updates from GCP Study Hub

arrow