Microsoft AZ-104: Azure Route Server for Hybrid Networks
Azure Route Server exchanges BGP route information with supported network virtual appliances and Azure routing. It simplifies dynamic routing only when route ownership, failover, propagation and security inspection paths are designed deliberately. Route Server can simplify dynamic route exchange, but it does not replace a correct transit architecture or every gateway function. Know which peers advertise which prefixes, how route propagation affects spokes, and what happens when the advertised path disappears.
On this page
- Choose dynamic routing for a real change problem
- Keep the dedicated Route Server subnet correct
- Peer appliances with the right failure behavior
- Control which prefixes are advertised and learned
- Understand peering and hybrid-route propagation
- Coordinate Route Server with Azure Firewall
- Detect failures with more than session state
- Keep the routing design understandable to future operators
Choose dynamic routing for a real change problem
Static routes are understandable in small topologies, but a hybrid network with many spokes, branch prefixes and virtual appliances can become difficult to operate through manually maintained route tables. Every prefix change may require several coordinated updates, and a failed appliance can leave routes pointing toward an unavailable next hop. Dynamic routing can reduce that burden when the participating devices and Azure services support the intended behavior. A useful Microsoft AZ-104 hybrid-routing validation records route exchange peers and confirms the routes Azure Route Server actually learns and advertises.
Azure Route Server is a managed service that exchanges routing information using BGP between supported network virtual appliances (NVAs) and the Azure software-defined network. It does not inspect packets or act as the data-plane router carrying all traffic itself. Its role is control-plane route exchange, so deployed NVAs and Azure network resources still need working forwarding paths.
The first design question is what routes must be learned dynamically, from which peers, and where they should be advertised. An organization adding one isolated spoke may not need Route Server. A central routing environment with multiple NVAs and frequent branch changes may benefit substantially if route propagation is planned and verified.
Keep the dedicated Route Server subnet correct
Azure Route Server requires its documented dedicated subnet and has product-specific deployment constraints. The subnet is not a convenient location for ordinary workloads, firewalls or service endpoints added later. Microsoft’s Route Server FAQ describes the required subnet name and restrictions, including unsupported NSG and user-defined route associations on that subnet.
Reserve adequate address space, place the service in the intended hub VNet and verify support and limits before deployment. A seemingly harmless change to subnet allocation or filtering can break the routing control plane. The network diagram should show the Route Server resource and BGP peers separately from the packet-forwarding devices.
Do not infer that deploying Route Server makes a hub-spoke environment automatically secure. If an application subnet sends traffic directly toward a destination without traversing the intended NVA, dynamic route exchange alone does not guarantee inspection. The appliance topology, route selection and any central firewall requirements still determine actual traffic paths.
Peer appliances with the right failure behavior
NVAs participating in BGP can advertise reachable prefixes and learn routes according to supported policy. Multiple appliances may provide resilient forwarding when the design handles peer state, route preference and return paths correctly. A BGP session reporting Established does not prove that application traffic is forwarded or inspected as intended.
An appliance may advertise a route while its internal firewall engine or downstream link is unhealthy. The platform could still see a valid next hop, depending on how the device handles withdrawal and health. Build health and failover behavior into the NVA design, including controlled withdrawal or preference changes. Test what routes Azure sees after an appliance is intentionally removed from service.
Be cautious with route asymmetry. One direction might travel through NVA-A and the reply through NVA-B or a direct VNet peering route, undermining stateful inspection. Compare effective routes from both endpoints under normal and failed-peer conditions. A successful ping through one path is not evidence of symmetric behavior for every protected application flow.
Control which prefixes are advertised and learned
A BGP route can be valid yet unwanted. Overly broad advertisements can redirect traffic toward an appliance that should not handle it; overly specific prefixes can consume route table capacity and complicate troubleshooting. Use the supported route filtering and route-map capabilities where necessary, and document why a prefix should propagate to each network segment.
Route summarization reduces repeated routes when the address plan permits it, but aggregation can hide reachability gaps. An aggregate advertised during partial failure might attract traffic to destinations whose more specific underlying networks are unavailable. Test the forwarding behavior behind any summary and avoid announcing a broad default route without a clear purpose.
Microsoft documents current Route Server capabilities. Those limits matter during large network expansions: an appliance advertising too many prefixes can affect BGP sessions and service behavior. Monitor route count and growth as part of capacity planning rather than waiting for a failure at a documented boundary.
Understand peering and hybrid-route propagation
Spoke VNets, transit hubs and on-premises connections require intentional route exchange. VNet peering does not behave as a full mesh of automatic third-party routing simply because address spaces are visible. Gateway transit, route-server exchange, branch-to-branch features and NVA forwarding must be reviewed within the supported design.
For example, an on-premises branch prefix may be learned through ExpressRoute and then advertised toward an NVA, but whether traffic can traverse from one connected network to another depends on route propagation and the data-plane topology. Do not assume every route can be safely re-advertised in every direction. Avoid route loops and unintended access between networks with different trust levels.
Hybrid networks often combine BGP with DNS and firewall requirements. A private service can resolve correctly while taking an unintended path if new advertisements change next-hop selection. Include application flow and security inspection checks in routing change validation instead of relying only on BGP peer status.
Coordinate Route Server with Azure Firewall
Azure Firewall and third-party NVAs serve different operational purposes and can coexist in some supported architectures, but the route plan must state which component inspects each flow. Dynamic routes that bypass an intended inspection point can undo carefully configured policy without changing the firewall rules themselves.
In a complex hub, a platform team might use Route Server for NVA route exchange while Azure Firewall handles some controlled egress or inter-spoke traffic. That requires explicit route selection and compatible service integration. Do not reuse sample topologies without checking current routing behavior, service versions and limitations for the actual network.
Security owners should review route changes as part of policy changes. If an NVA announces a new default path, verify whether public egress, private endpoints and on-premises networks remain within the approved inspection model. Effective route views and appliance logs can provide the data-plane evidence that the BGP control plane alone cannot.
Detect failures with more than session state
Monitoring should cover BGP peer state, learned and advertised prefixes, route changes, appliance health and representative application connectivity. A session that stays up while the appliance silently drops traffic is an operational failure even if Route Server appears healthy. Conversely, a brief peer reset that does not affect traffic may deserve a different severity from a real outage.
Create a baseline of expected routes and peers. After a deployment or maintenance window, compare the current route picture to that baseline. Unexpected new prefixes deserve as much scrutiny as missing ones because they can redirect traffic or expose unintended paths. Preserve configuration changes and timestamps to explain why the route graph changed.
A safe lab can use a supported BGP-capable NVA to advertise one test prefix, observe it in Azure’s routing information, then withdraw it intentionally. Verify both the control-plane route change and the resulting traffic behavior toward a test destination. Do not run route advertisement experiments in production without an approved blast-radius assessment.
Keep the routing design understandable to future operators
Document the Route Server subnet, BGP peers, expected prefixes, route maps, NVA failover model, and the network flows whose security depends on the design. A table describing which routes are advertised in each direction is often more useful during an incident than a diagram containing unlabeled arrows.
Review growth and change cadence. Dynamic routing reduces manual updates but increases the need to understand route policy and failure propagation. The strongest design keeps policy intent explicit while using BGP to distribute only the routes that should change dynamically.
A live BGP session proves that a routing relationship is established, not that each application can use the advertised path safely. Check the prefixes the peers exchange, the route selected at the source subnet, the return route, and any inspection devices between them. The operational test is a successful authorized transaction across the intended hybrid path, including when one peer or appliance is deliberately unavailable.