Preparing for the Azure Administrator Associate exam is not mainly a matter of remembering service names. You need to be able to change an Azure environment, explain the consequences, and diagnose the layer that failed. An administrator can know the definition of an Azure role but still assign it at the wrong scope. They can configure a backup and never verify that an application will actually run after restoration. These are the gaps a useful study plan should uncover.
Microsoft’s AZ-104 exam objectives, effective April 17, 2026, define the dated exam scope and domain weights. Preparation for the Microsoft AZ-104 exam is stronger when each domain is practiced through an observable administrative task, with a result that can be checked rather than merely memorized.
Turn each exam domain into evidence of an administrative skill
The current blueprint measures five areas, and their percentages describe relative emphasis rather than independent modules. Troubleshooting a single application can cross networking, identity, storage and monitoring. Use the table below as a study map, not as a reason to stop learning a topic at the edge of its weight band.
| Microsoft skill domain | Weight | Meaningful practice result |
|---|---|---|
| Manage Azure identities and governance | 20–25% | Authorized principals at correct scopes, predictable Policy decisions and responsible resource ownership |
| Implement and manage storage | 15–20% | Working access controls, defensible redundancy and tested object recovery |
| Deploy and manage Azure compute resources | 20–25% | Reproducible deployment, understood dependencies and a safe change or rollback |
| Implement and manage virtual networking | 15–20% | Verified packet paths, secure intended access and correct name resolution |
| Monitor and maintain Azure resources | 10–15% | Useful collected signals, actionable alerts and restorable workloads |
For every objective, ask three questions: Can I configure this? Can I explain why the alternative would be less suitable? Can I identify which boundary failed when the result is wrong? A working screenshot alone answers at most the first question. Keep a notebook of your predicted result and observed result, including the case where the environment cannot support a live test and the work remains a documented design exercise.
Start with identity and governance: permission is not a property of a username
In a disposable training subscription, create two test principals and a resource group. Assign a narrowly scoped built-in Azure role and test an allowed resource-management operation and one disallowed operation. Change the role assignment scope only after predicting which additional resources it will affect. The distinction between a role definition, principal and scope is more useful than memorizing a list of role names. Azure RBAC role scope determines which resources and actions a principal may access.
Practice Microsoft Entra user and group properties, external access and self-service password reset under a supported tenant setup. Microsoft Entra authentication and Azure resource authorization are related but not interchangeable: successful sign-in does not prove a caller has a storage data-plane role.
Then apply a harmless tag policy and a resource lock in your test group. Attempt a change that should be blocked. Determine whether the reason is an Azure Policy assignment, missing RBAC action, ReadOnly or CanNotDelete lock, quota, or a service prerequisite.
Resist the temptation to grant Owner whenever something fails. Add a budget alert and inspect an Advisor recommendation, noting that a budget alert does not itself prevent all spending. Remove any unused resource after the lab and verify that there are no retained billable dependencies.
Separate storage authorization, resilience and recoverability
Create a test storage account and upload a non-sensitive file. Compare access through a suitable Microsoft Entra data role with a tightly scoped shared access signature. Deliberately expire the SAS, remove a role assignment, and test a service-side network restriction one at a time. All may surface to users as an upload failure, yet the repair belongs at different layers. A firewall rule does not grant permission; a valid token does not guarantee that the network route is open.
Use blob versioning and soft delete on disposable objects to practice recovering from an overwrite and a deletion. For Azure Files, test a supported share snapshot or share soft-delete scenario separately. Blob soft delete and Azure Files snapshots provide distinct recovery options for different storage failure scenarios. Know why LRS, ZRS and geo-redundant choices address infrastructure failure domains but do not independently undo every mistaken application write.
Transfer a small authorized dataset with Storage Explorer or AzCopy and confirm the actual destination content. Distinguish a single transfer from a continuously configured object replication policy. When choosing storage tiers and lifecycle rules, consider retrieval time, access patterns, retention and the cost of data that can no longer be accessed quickly. The design should follow the recovery objective, not merely the lowest advertised storage price.
Practice compute as an operational change, not a creation wizard
Deploy a small VM with a data disk and a narrowly scoped managed identity. Record the initial size, region, disk tier, identity, network interface, guest service status and backup coverage. Check whether it runs the intended application, not just whether Azure says the VM is Running.
A controlled resize should begin with measured CPU, memory or I/O pressure, available SKU capacity, quota and an interruption plan. Azure VM disk encryption, availability, and resource moves require separate checks because recovery, encryption, and migration constraints differ.
Modify an existing ARM template or Bicep file in a lab and examine the change before applying it. If the predicted deployment touches an unexpected resource or security setting, correct the source before release. Practice interpreting the template as well as editing it, because repeatability is not the same as blindly applying infrastructure definitions.
Compare the operational purpose of Virtual Machine Scale Sets, Azure Container Registry, Container Instances, Container Apps and App Service. A replicated VM definition cannot make a stateful application available without considering shared data and traffic routing. For App Service, configure a suitable plan, custom DNS and TLS in a disposable scenario, distinguish inbound restrictions from outbound VNet integration, and verify a staging slot or supported recovery process. A deployment slot does not automatically roll back an irreversible database change.
Trace networking from client request to service response
Create two subnets and document their address spaces. For a given client-to-application request, write down the DNS answer, route, effective NSG rules and destination service requirement. An NSG filters traffic; a user-defined route chooses a next hop. Neither substitutes for the other. When a connection fails, confirm the source and target IP, protocol, port and effective rule before changing permissions.
Use a simple VNet and subnet exercise to explore explicit egress and safe address allocation. Peer two nonoverlapping test networks, then explain why peering does not automatically give every spoke access to every other spoke. For PaaS connectivity, distinguish service endpoints from private endpoints: the former extend supported subnet access controls and route traffic through the Azure backbone, while private endpoints introduce a private network interface and require correct private name resolution and authorization.
Introduce one failure: a DNS zone link is missing, a lower-numbered deny rule matches first, or a user-defined route points to an unavailable appliance. Diagnose it with effective security rules, effective routes or Network Watcher/Connection Monitor, depending on the symptom. When comparing Azure Load Balancer with Application Gateway, tie the selection to transport versus HTTP routing and health checks instead of using the services as interchangeable labels.
Make monitoring lead to a successful recovery
Collect one VM performance metric and a relevant diagnostic log into an appropriate destination. Run a Log Analytics query against an actual collected record, then explain how you would spot a missing data collection rule if the expected event were absent. Metrics show numeric trends; logs support event investigations; Azure Activity Log shows management operations. An alert rule should identify a condition worth acting on, and an action group should reach somebody who can respond.
Use Azure Monitor Insights for a VM, storage account or network when configured telemetry supports it. Treat the dashboard as an investigative starting point rather than a definitive diagnosis. If a service is slow, compare the resource metric to guest processes, dependency latency and time-aligned logs. An alert that fires continuously without identifying an actionable state should be revised, not merely added to a larger list.
For data protection, identify the workload, relevant Recovery Services or Backup vault capability, policy and restore path. Distinguish Azure Backup from Azure Site Recovery: a recoverable backup is not the same as a replicated workload ready for planned or unplanned failover. Restore a nonproduction test resource when possible, then check data, identity, network access and service behavior. Record expected versus observed outcomes. A successful backup job does not establish a tested recovery time.
Use an integrated lab to test your weakest assumptions
After a domain-by-domain first pass, connect a few inexpensive resources into one scenario: an authenticated user submits an allowed request; a web or VM workload accesses a secured storage object; networking restricts the path; monitoring reports a deliberate failure; and an operator restores the affected state. The exact services depend on what your authorized sandbox supports. You do not need a large production environment to demonstrate the boundaries.
Keep a short incident record for each failure: expected operation, affected principal and resource, observed error, first hypothesis, data used to test it, correction and final verification. If you cannot run the exercise because it needs an unavailable tier or tenant capability, document it as an untested design and use current first-party documentation to evaluate the decision. Do not describe a hypothetical lab as a completed deployment.
Adapt study time to the evidence. One administrator may know storage thoroughly but repeatedly overlook DNS. Another may be comfortable with routing but confuse Microsoft Entra roles with Azure resource roles. A uniform number of hours per topic would waste the first person’s time on the wrong material. Revisit weak decisions until you can explain the result without looking at a menu.
Judge exam readiness by the decision you can defend
Practice questions are useful for discovering unfamiliar terms and weak areas. After each wrong answer, identify the missing distinction and verify it against the current Microsoft guide or a controlled lab. A score alone is weak evidence if the same scenario keeps failing when the wording changes. Avoid assuming that older blog posts reliably describe current exam length, question formats, service names or the official scheduling process; check those details with Microsoft when booking.
Before scheduling, answer five questions aloud: Why would a resource-group role fail to authorize Blob data access? Which recovery control would reverse a deleted file share rather than a deleted blob? What changes when a VM moves to another resource group, and what remains in the same region?
Why can a private endpoint exist while a client still resolves a public IP? Which test proves a restored application can actually serve users? The quality of the diagnostic explanation matters more than merely naming a product.
Finally, close the training resources, reconcile any unexpected costs and revisit Microsoft’s active AZ-104 objectives for changes. The goal is durable Azure administration judgment: appropriate authority, correct configuration, observable failures and verifiable recovery.