Microsoft AZ-104: Azure Private Endpoints and DNS
A private endpoint gives an Azure service a private network entry point, but successful application access still depends on correct DNS, routing, service permissions, and a clear separation between public and private paths. A private endpoint changes the name-to-address path as well as the traffic route. If the application resolves a public address or the wrong private zone, a restrictive network design may be technically correct while the connection still fails.
On this page
- Understand what becomes private and what does not
- Keep the ordinary service hostname in the application
- Decide where private DNS zones are authoritative
- Route DNS queries from hybrid and spoke networks
- Keep service authorization and packet reachability distinct
- Plan more than one private endpoint carefully
- Rehearse an endpoint migration without weakening security
- Make private service access maintainable
Understand what becomes private and what does not
A private endpoint creates a network interface with a private address in a VNet for access to a supported Azure service. Applications can use that path to reach the service without traversing its public endpoint in the same way. The endpoint does not move the underlying platform service into the customer’s subnet, and it does not automatically disable the service’s public endpoint or grant application authorization. When preparing a Microsoft AZ-104 networking change, test both private endpoint routing and the DNS name clients actually resolve; a private IP alone is not enough.
Consider a storage account serving a private application. Creating a blob private endpoint gives the account a private network path for the blob subresource from properly connected clients. The application still needs valid storage data authorization. The owner must separately decide whether public endpoint access should remain available, be restricted by firewall rules, or be disabled after all required clients have migrated.
This distinction matters in architectural reviews. A diagram showing a private endpoint icon is not proof that every consumer uses the private address. A developer’s laptop, an Azure build agent and a workload VM can resolve the same hostname differently if their DNS configurations differ. Evidence must be collected from the networks that actually make requests.
Keep the ordinary service hostname in the application
Azure services often use a public-looking service hostname that follows a DNS alias chain to private-link naming when a private endpoint is configured. The application should generally continue using the documented service hostname rather than hardcoding an endpoint’s private IP. Correct DNS then chooses the private network path for clients intended to use it.
Hardcoded addresses complicate replacement, failover and service compatibility. They also prevent clients from benefiting from DNS changes that accompany controlled private endpoint management. If a workload was configured to call a raw private IP as a quick incident workaround, document the deviation and design a return to supported name-based access.
For an Azure Storage blob endpoint, the private DNS zone and record pattern differ from those of other Azure services or even other storage subresources. Confirm the actual service’s documented private-link DNS mapping. Do not assume one privatelink zone is sufficient for blob, file, queue and other APIs accessed by the same storage account.
Decide where private DNS zones are authoritative
A private DNS zone needs to be reachable through links or resolver forwarding paths from the consuming VNets. An endpoint and zone created inside one workload group may work locally while failing for peers or on-premises systems. Large environments often benefit from a managed zone and resolver architecture rather than separate copies of overlapping zones in every subscription.
Creating multiple private zones with the same namespace can produce conflicting or incomplete answers when clients depend on different links and forwarders. The design should specify which team owns records, how endpoint lifecycle updates them, which VNets are linked, and how hybrid clients resolve the private names. A central zone is only useful when every intended consumer actually reaches an authoritative resolver for it.
Microsoft’s Private Link DNS guidance discusses private zone patterns and the interactions of public and private resolution. Follow the current service-specific tables rather than copying records from an unrelated product. DNS zone groups can automate record associations for supported configurations, but administrators must still account for custom DNS architectures and multiple endpoints.
Route DNS queries from hybrid and spoke networks
A VM using Azure-provided DNS in a linked VNet may resolve the private endpoint correctly. An on-premises machine using corporate DNS needs a forwarding path into Azure’s private zone resolution. Azure DNS Private Resolver can provide supported inbound and outbound endpoints and forwarding rules, depending on the network architecture and required directions.
A hub-and-spoke design may link a central resolver and zone to the hub while workload spokes use the approved resolver path. Peering alone does not guarantee that private DNS zones linked only to the hub are resolved by every spoke’s configured DNS server. Verify the actual VNet DNS settings and forwarding rules, including routes and firewall permissions to the resolver endpoint.
Beware of forwarding loops. If an Azure resolver forwards a private suffix to on-premises DNS while the on-premises server forwards it back to Azure, lookups may time out or behave inconsistently. Test the selected suffix and resolver targets with traceable evidence before adding broad wildcard forwarding rules that can affect unrelated public DNS dependencies.
Keep service authorization and packet reachability distinct
A correct private IP answer does not guarantee a successful storage or database transaction. An NSG, user-defined route, central firewall or endpoint policy may prevent the network connection. The destination service can also reject an authenticated client with an insufficient role or token. Diagnose each gate separately rather than granting Contributor to repair a network timeout.
Start with the client’s name-resolution result. If the intended private address appears, test the required protocol and port, review effective network rules and routes, and confirm service endpoint health. If the transport works but the API returns a denial, inspect service-specific identity and authorization. VNet and DNS troubleshooting provides a complementary packet-path diagnostic approach.
An intentionally blocked public endpoint can make a DNS error very visible. A client that resolves the public address fails while another client using private DNS succeeds. That is not evidence that public access must be reopened; it is evidence that the failing client lacks the intended private resolution or route. Restoring private access preserves the organization’s security decision.
Plan more than one private endpoint carefully
A large estate may have several application VNets, regions or environments that need private access to one service. Each endpoint has a network location and potential DNS implications. Decide which clients should connect to which endpoint, and how records and routing change during a failure or environment migration. Do not let automatic zone records accidentally send every client toward an endpoint reachable only from one network.
Cross-region and hybrid arrangements require additional review of service support, name resolution, routing and recovery behavior. Some platforms allow private endpoints to connect to a service in a different region, but that does not make cross-region latency, failover or cost irrelevant. Test the actual client topology and avoid assuming that one successful endpoint deployment validates every future region.
Infrastructure-as-code deployments should declare endpoint dependencies, required zone groups or records, and ownership. When an endpoint is removed, check whether associated DNS records and network permissions are cleaned up appropriately. Stale private records can redirect clients to an address with no working endpoint long after a migration appears complete.
Rehearse an endpoint migration without weakening security
In a disposable test environment, deploy a supported Azure service with a private endpoint and an appropriately linked private zone. From a test VM, resolve the documented service hostname and confirm the private answer. Test the intended operation with its narrow service permissions. Then remove the test zone link and observe the lookup change; restore the link before testing connectivity again.
Next, deliberately deny the operation through service authorization while leaving DNS and the network path intact. The change should yield a different symptom from a missing zone link or blocked route. Record that distinction, along with the control responsible for each outcome. It teaches a recoverable diagnostic process rather than encouraging blanket network rules.
This exercise demonstrates why a private endpoint alone cannot prove application reachability. Network placement, DNS answers, workload authorization and service health each require their own evidence; when one fails, changing the other three can make the incident harder to explain.
Make private service access maintainable
Private endpoints have recurring operational concerns: IP address management, DNS ownership, endpoint approval or connection state, security policy, monitoring and cleanup. An endpoint created with no owner becomes a long-term dependency that few people feel confident changing. Record the associated service, consumers, network scope, DNS zone, and recovery route.
Security teams should review whether public access is still needed after the private path is verified. Application teams should know how to test from their own networks rather than asking the platform team to temporarily reopen broad access whenever a deployment fails. When these procedures are explicit, private connectivity strengthens the architecture without turning ordinary operations into a series of unexplained connection failures.