Microsoft AZ-104: The Troubleshooting Trail Across Azure
A customer reports that an internal inventory application stopped loading after a routine Azure deployment. The virtual machine is running, there is no apparent platform-wide incident, and the support chat is already debating whether to restart the VM, open the network security group or assign the service a broader role. None of those changes follows from the evidence yet. First establish the failing request, its source and destination, and the layer responsible for the observed behavior.
This fictional incident is a practical Microsoft AZ-104 troubleshooting exercise rather than a generic study schedule. It follows one application across Azure identity, Blob Storage, DNS, routing, compute and monitoring, illustrating why several plausible fixes can be wrong. These are original educational examples, not claims about an actual customer outage or recalled Microsoft questions. For the tested administration context, use the Microsoft AZ-104 exam page and the current issuer objectives. Build any demonstration only in an authorized, cost-controlled test environment.
Suppose the application runs on a small VM and reads reports from a storage account through a private endpoint. A managed identity is intended to have the necessary Blob data permission. The account’s network policy restricts unintended public access. A private DNS zone is supposed to return the endpoint’s private address to application clients. A deployment modifies configuration shortly before users notice timeouts.
These details create several competing hypotheses. A recent deployment might have changed an application setting. The managed identity might lack the data-plane role. The VM might resolve the wrong endpoint address. An NSG or user-defined route might block traffic. The application process could also be unhealthy despite the VM’s Running status. Investigating them in a reasoned order matters more than memorizing the Azure services involved.
Before changing anything, create a small incident packet: resource IDs, subscription, client network, exact hostname, requested port and operation, affected users, first failure time, current configuration revision and the rollback owner. “Azure storage is broken” is not a testable statement. “This VM’s request to this Blob hostname fails from this subnet after this revision” is one.
Two permission questions are easily confused. Can an operator change the Azure storage account configuration? Can the application read a blob? Management-plane roles and data-plane roles are distinct. A human can view a storage resource in the portal but lack object read permission. An application with a correct Blob data role can still fail because its endpoint is unreachable. A successful Microsoft Entra sign-in does not establish either complete authorization or connectivity.
Record the principal used by the application, its relevant role assignment and the actual scope. If the failed operation is a blob read, inspect a suitable Storage Blob Data Reader or Contributor role, as applicable. If the deployment itself was rejected, inspect the failed deployment operation: Azure Policy, a resource lock and an RBAC denial each point to a different control. Do not “solve” an unknown by assigning Owner across a subscription.
The existing Entra ID and governance explanation clarifies identity, policy and resource scope. The RBAC least-privilege article develops the distinction between principal, role and scope. In this incident the existing identity grant is provisionally valid, so widening it without a denied-access event would be a speculative and risky change.
Use the exact storage hostname the application requests and perform the lookup from the affected VM or its runtime. If that client receives a public address when the design requires a private endpoint, the immediate investigation is name resolution. The private endpoint can be Approved in the Azure portal and still be unreachable to a client that resolves the wrong destination. Testing a different machine’s DNS result may hide this difference.
Microsoft’s private endpoint DNS troubleshooting guidance describes common causes: the private DNS zone is not linked to the VNet that issues the query; its A record is missing or stale; or a custom resolver’s conditional forwarding does not reach the intended Azure resolver path. Peered VNet and on-premises clients may use different paths. Record which resolver answered and which IP address was returned instead of assuming all clients share the same DNS configuration.
In this fictional incident, suppose the private DNS zone has the correct endpoint record but the application VNet link was lost during a configuration deployment. That would explain why an affected VM gets a public address. This is the hypothesis chosen for the example, not a verified diagnosis of any actual environment. The administrator should inspect zone linkage before authorizing its restoration, then re-run the same lookup from the application host.
If the correct private address is returned but requests still fail, do not assume DNS is the whole issue. The next boundary is routing, filtering, host behavior or the target service itself. A corrected FQDN answer is an important intermediate result, not evidence of a successful Blob read.
A user-defined route determines which next hop Azure selects for a destination. A network security group determines which defined flows are admitted or denied at its attached scope. The host’s firewall can impose additional controls, and a service may be unavailable even when a TCP connection succeeds. Changing all of these settings together would conceal which layer was actually responsible for the failure.
Network Watcher Next hop tests Azure’s forwarding decision for a destination address. A result of None may reveal a route that discards packets. Effective security rules shows the combined applicable inbound and outbound policies for a VM NIC, including the subnet and NIC NSGs and relevant admin rules. IP flow verify or NSG diagnostics can evaluate an appropriately defined packet, giving narrower evidence than a topology diagram.
For a one-time source-to-destination test, connection troubleshoot can report reachability, observed latency and indications of routing, security or guest-firewall problems. Confirm resource prerequisites, especially when using a documented preview capability. A successful probe establishes only that the tested connection was reachable in those conditions. It does not prove the application used the correct credentials or returned useful content.
The existing AZ-104 virtual networking and private endpoint DNS guides explain these technologies individually. This incident adds the decision order: address resolution first when the hostname is wrong; network path and policy when the address is right but traffic fails; application and authentication if the network appears usable.
A failed Blob request can produce a similar symptom for unrelated reasons. A managed identity may lack the necessary data role. A shared access signature may be expired, not yet valid or scoped to the wrong operation. The storage firewall may deny a client carrying otherwise valid authorization. A private endpoint may also be unreachable from the caller. The administrator needs the actual response and the identity or token type to tell these failures apart.
For Microsoft Entra authentication, inspect an appropriate Storage Blob Data Reader or Contributor assignment at the intended scope. If using SAS, examine its permissions, resource, start and expiry and never paste a complete valid token into an incident ticket. Do not broaden the service principal’s management role merely because it cannot read a blob. And do not turn off account network restrictions merely because a correctly authorized application still cannot reach the approved endpoint.
In this worked case, the identity role was correct, and the DNS observation pointed to the application VNet’s private-zone link. An administrator should therefore preserve the existing permission policy and address the proven resolver condition. After that, conduct a harmless test read under the application’s own identity. A successful command run as a different administrator cannot establish that the application principal can perform the same operation.
The Azure storage operations guide provides broader access, replication and object lifecycle context. This case demonstrates why a data-access error must be traced through independent network and authorization boundaries before anyone grants a wide role.
The recent deployment still deserves investigation even if DNS proves to be the primary failure. A completed ARM or Bicep operation means Azure accepted its defined changes; it does not demonstrate that the deployed application configuration is useful. Inspect what changed in connection settings, resource identifiers and service dependencies. Preserve credentials outside the incident record. Check whether the application process is listening, whether its configuration points at the intended hostname and whether the new revision introduced another failure.
If DNS is corrected yet errors persist, compare the current application health and request logs with the prior state. A restart may temporarily reload a stale setting, but treating restart as the default response can destroy evidence and conceal a repeatable deployment mistake. A virtual machine, App Service and container deployment each have different operational controls; use only the troubleshooting paths appropriate to the actual workload. The existing Azure compute administration discussion explains those lifecycle differences.
An incident timeline should connect what users experienced to changes and telemetry, without pretending all signals contain the same kind of evidence. Azure Monitor platform metrics can show resource pressure, but a CPU curve does not explain a denied Blob request. The Activity Log can reveal management-plane changes, while resource and application logs require suitable collection configuration before their details can be queried. An empty Log Analytics result could mean missing ingestion or the wrong query time, not an absence of errors.
The AZ-104 Azure Monitor incident guide describes metric selection, log collection, action groups and Network Watcher escalation. For this case, correlate the deployment timestamp, the first failed request, the DNS answers seen by the VM, any authorized DNS configuration correction and the first successful end-to-end application transaction. Preserve resource IDs and UTC timestamps so observations can be compared accurately, and distinguish a documented fact from a suspected cause.
Suppose the evidence in this case confirms that the application’s VNet was no longer linked to the private DNS zone. The authorized correction is to restore the missing link—not to grant Owner, disable an NSG or recreate an otherwise healthy private endpoint. Before the change, document the old state and the observation that supports the action. Use the smallest supported configuration change and have a rollback plan if the expectation is wrong.
After restoring the link, resolve the same hostname from the same application VM. If the intended private address appears, test the actual network path and then the Blob read using the application’s managed identity. Finally, perform the inventory request that originally failed. This last result is essential: a corrected DNS lookup and even a successful TCP probe are intermediate tests, not evidence that users can complete the application’s intended work.
Check that monitoring reflects recovery, identify whether the configuration regression could recur, and add the relevant test to the deployment process. If a broader disaster-recovery event is involved, verify the usable recovery point and application dependencies rather than relying only on a green backup status. The Azure Backup and recovery resource covers that separate failure objective.
In an authorized lab, recreate only the meaningful elements of this case: a small workload, a storage dependency, a suitable data identity, an approved private endpoint and a DNS path you can safely inspect. Predict the effect of a wrong DNS record, insufficient Blob data permission or an NSG denial before running any test. Record the observation that distinguishes each cause. Restore the original configuration before leaving the lab, and review costs and resource cleanup.
The broader AZ-104 hands-on preparation article covers a wide range of exercises. This troubleshooting trail has a narrower goal: follow one failure to its controlling boundary, change only what the evidence supports and confirm a real user-facing outcome. That sequence remains useful even when an exam scenario changes the resource names or combines domains differently.
On AZ-104, several answers may contain valid Azure tools, but the correct administrative response depends on the symptom, permitted scope and requested result. Knowing which observation comes next is more valuable than reflexively adding a role, restarting a resource or selecting a familiar dashboard. Compare current behavior with Microsoft’s documented controls whenever the platform evolves.
