Microsoft SC-500: Private Link Security Patterns
Private Link can reduce the exposed network path to Azure platform services, but DNS, resource authorization and endpoint ownership determine whether that private boundary is actually secure and operable. Engineers should trace the client’s hostname resolution, subnet reachability and service-level authorization separately. Creating a private endpoint does not automatically remove the public path, repair DNS in every connected network or give an application permission to read protected data.
On this page
- Decide which service must become privately reachable
- Choose between workload-local and centralized endpoint placement
- Treat private DNS as security architecture
- Disable unintended public access as a separate control
- Combine private connectivity with Entra and service permissions
- Understand subnet policy, routes and inspection limitations
- Plan hybrid connectivity and multi-region recovery
- Make endpoint creation and approval repeatable
- Prove both expected access and prohibited access
Decide which service must become privately reachable
A private endpoint creates a network interface with a private IP address inside an approved virtual network for a supported PaaS resource or subresource. A client can use the normal service hostname while DNS returns the endpoint’s private address in the networks that should take the private path. That reduces public reachability, but it does not alter the caller’s identity or automatically grant data access to the target service.
Start with a threat model and data-flow map. Which application needs storage, SQL, Key Vault or a model endpoint? Which network and identity should be permitted? Does a partner or operations tool also require access? Avoid creating a private endpoint simply as a generic best-practice token when no private-network boundary is required, because every endpoint adds DNS, ownership and failure dependencies.
Identify the exact target subresource. For Azure Storage, blob, file, queue, table and DFS paths may require different private endpoints depending on the application. One endpoint to the storage account’s blob subresource does not automatically make every storage service private. An engineer should enumerate actual API dependencies before declaring that a resource has been isolated.
Choose between workload-local and centralized endpoint placement
A private endpoint in the workload spoke keeps its IP address, DNS relationship and resource lifecycle close to the application team. This can make blast radius easier to understand, especially when services belong to different security zones or when production and nonproduction must remain separate. The tradeoff is that several spokes accessing one shared service may each need endpoint and DNS management.
Centralizing endpoints in a hub may reduce duplication for genuinely shared services. It also concentrates dependency on shared routing and DNS and requires clarity about which spokes should be allowed to reach the same resource. A hub IP address is not a security authorization model: every connecting application still needs an appropriately scoped role or service credential.
The right pattern follows ownership and acceptable failure domains. For a sensitive finance datastore used by only one application, a workload-local endpoint may be easier to protect. For a centrally owned enterprise service with many approved consumers, a shared endpoint may be reasonable when routing, inspection, change management and access control are explicit. See Azure network security at scale for the wider topology tradeoffs.
Treat private DNS as part of the security boundary
Private Link deployments frequently fail because DNS was not designed with the endpoint. A client that should use the private path may resolve the public service address through an unlinked zone, an on-premises recursive resolver or an inconsistent conditional-forwarding rule. The endpoint itself can be healthy while the application experiences timeouts or unexpected public routing.
Document which private DNS zone serves each supported service, which virtual networks link to it, which clients forward queries to Azure and how hybrid resolvers find the record. Use the normal service hostname when possible instead of hardcoding endpoint IP addresses in application configuration, because addresses and placements change during migration or recovery.
Validate DNS from multiple locations: the application spoke, a prohibited spoke and an on-premises client if hybrid access is required. Compare the resolved address with the intended endpoint and inspect actual routes. If a client resolves the public address, do not compensate by opening an unnecessarily broad service firewall rule. Fix the resolution path and confirm the policy’s intended public-network behavior.
Disable unintended public access as a separate control
Adding a private endpoint does not guarantee that the public endpoint or service firewall is disabled. Resource settings differ by service, and management/control-plane behavior may also have separate access requirements. State whether public access is deliberately allowed for any approved administration or partner use, and identify compensating controls where it must remain.
For a private-only workload, stage the new endpoint and DNS first, test representative consumers and then restrict public access according to the service’s supported controls. Keep a rollback plan that restores authorized connectivity without leaving a permanent public bypass. Check not only that application traffic succeeds privately but also that an unauthorized external client cannot reach the service through the old path.
A useful negative test starts from an Internet-based or unapproved test source and attempts the service’s public endpoint. Record whether it is denied at network or service authorization layers. Network denial and authentication failure are different evidence; the team’s security objective should specify which path must be unreachable and what an authorized caller may still do.
Combine private connectivity with Entra and service permissions
Private access constrains which network path reaches the target. Microsoft Entra ID and service authorization determine what a principal can do after reaching it. A compromised workload in an allowed VNet may still abuse excessive Azure RBAC, SQL permissions or Key Vault roles. That is why a private endpoint is defense in depth, not a reason to weaken identity controls.
For example, an API running in a private application spoke may need read access to a particular storage container. Give its managed identity the narrow data role, route it to the approved private endpoint and confirm a second unrelated workload cannot read the same container merely because its packets can reach the storage IP. Both identity and network tests belong in acceptance evidence.
Where applications still use account keys or other shared secrets, consider whether a managed identity or federated authentication path can reduce credential exposure. The managed identities and least privilege discussion explores this authorization layer, which is independent of endpoint placement.
Understand subnet policy, routes and inspection limitations
Private endpoint network-policy support can be enabled for supported scenarios in which NSGs, user-defined routes or both are intended to apply at the endpoint subnet. That does not mean an arbitrary central inspection route will behave correctly. Routing specificity, return path and DNS must be tested together, especially when traffic flows through a stateful firewall or network virtual appliance.
Decide which flows genuinely require shared inspection. Forcing traffic through a central firewall by default may add latency and operational fragility while the service already authenticates every request and the network is restricted to one application. Conversely, cross-zone or regulated data flows may require explicit central visibility. Document the rationale instead of applying one blanket route pattern.
Use effective-route and security-rule diagnostics when behavior differs from the diagram. A 403 from the target service often requires an authorization investigation, while a timeout may indicate a route or firewall denial. Avoid widening several controls simultaneously during troubleshooting: change one suspected layer, record the result and preserve the intended negative test.
Plan hybrid connectivity and multi-region recovery
On-premises clients can use private endpoints over supported hybrid connections when both routing and name resolution are configured correctly. Review VPN/ExpressRoute traffic paths and DNS forwarding from the corporate resolver. The fact that an Azure VM can reach the endpoint is not proof that an on-premises business system will resolve and route to the same private address.
High-availability design must include the endpoint dependency. A workload deployed in two Azure regions may need region-appropriate endpoints, DNS records, routes and authorization. During failover, a secondary instance that can start but cannot find its private datastore is not operationally recovered. Test failure without reopening the target’s public endpoint as a permanent emergency path.
Maintain an endpoint-and-zone dependency map. Teams should know which applications would fail if one hub resolver, private DNS zone link or endpoint connection is removed. Include endpoint approval and service-owner contact details in runbooks so operations can distinguish a revoked endpoint connection from a broken route during an outage.
Make endpoint creation and approval repeatable
Private endpoints consume subnet address space and may require approval by a service owner when requester and resource are in different administrative boundaries. Plan IP capacity, naming, environment segmentation and operational ownership before a large rollout. In a busy platform, abandoned endpoints and DNS records can obscure which path an application still uses.
A standard provisioning workflow should capture target service/subresource, application owner, environment, endpoint subnet, expected DNS zone, public-network decision, authentication principal and recovery region. Automation can validate these parameters and create approved endpoint/zone relationships without asking network teams to reconstruct the design for each workload.
Lifecycle management is just as important as creation. When an application migrates or retires, review the associated connection approval, endpoint network interface, DNS records and service firewall exceptions. Remove unused resources after dependency validation so the platform does not accumulate confusing private paths whose original business justification no longer exists.
Prove both expected access and prohibited access
A production readiness exercise should start with a table of allowed and disallowed callers. From the application VNet, resolve the normal service name, verify the private IP path and perform one authorized operation. From an unapproved network, verify the endpoint is not reachable through a permitted route or the service cannot be accessed contrary to policy. From an approved VNet but unauthorized principal, confirm the service denies the operation.
Record the outcome separately for DNS, network connectivity, TLS/service connection and resource authorization. A single successful API call proves only one path works; it does not prove public fallback is closed, unrelated VNet access is denied or privileged identities are restricted. Repeat the tests after shared DNS, route, firewall, endpoint and recovery changes.
For Microsoft SC-500, Private Link design is best understood as a combined network, identity and operations discipline. Choose the endpoint boundary intentionally, treat DNS as critical infrastructure, turn off unwanted public access, maintain least privilege and prove the actual traffic outcome. That is more durable than memorizing where a portal button creates a private endpoint.