An Azure resource can deploy successfully while remaining insecure, unavailable or impossible to recover in the time the business needs. Portal screenshots and completed tutorials therefore prove less than they seem. Effective AZ-104 practice starts with a declared operating requirement, makes one controlled change, captures the evidence and tests both what should succeed and what should remain prohibited.
These original exercises support the skills measured by Microsoft AZ-104 Exam across identities, governance, storage, compute, virtual networking and monitoring. Use a training subscription or another environment you are authorized to administer, only synthetic data and explicit cost limits. Clean up every resource, role, rule and credential after the exercise. The goal is a small portfolio of observable administrator decisions, not a collection of expensive Azure products.
Create a test environment that makes outcomes repeatable
Record the subscription, resource group, region and temporary resource names before any deployment. Keep test account identities separate from normal administration wherever practical. Assign a budget alert or set clear consumption limits and choose small resource sizes that satisfy the exercise. Avoid creating unexpected ongoing costs through public IPs, gateways, reserved capacity or retained disks.
For each lab use a test card with five elements: required security or operational property, known baseline, one planned change, expected positive and negative outcomes, and cleanup procedure. A positive test confirms a permitted behavior; a negative test confirms the intended boundary still blocks an unacceptable one. Record the evidence rather than relying on “the wizard said success.”
Preserve the relevant configuration, time and error messages. Do not copy access tokens, secrets or real customer information into the notebook. When a test fails, keep the first result; it may reveal a mistaken assumption. Repair one cause, rerun the affected case and explain how the evidence changed.
Lab 1: demonstrate identity, permission and governance as separate controls
Create a disposable resource group and a small test resource. Define two identities: one should inspect the resource and the other should be permitted a specific management action at a narrowly defined scope. Record the expected permissions before applying the assignments. Where account setup is restricted by organizational rules, a supported simulation of effective role scope may replace live identity creation.
Sign in or act using authorized test principals and attempt one permitted and one prohibited operation. A successful sign-in does not prove that every Azure management action is allowed. Verify that the role assignment’s scope explains the observed result. A role intended only for one resource group should not unintentionally grant modification rights across the whole subscription.
Next use a safe demonstration Azure Policy rule that restricts a chosen configuration, such as a location or tag requirement. Test a deployment that complies and another that violates the rule. Record whether policy evaluation denied the request even when a principal otherwise had sufficient deployment rights. This shows that governance and authorization are different enforcement systems.
Restore original role and policy settings. The principle behind Azure Resource Manager is clearer once a learner can name the resource scope, decision authority and observation for each failed or successful operation.
Lab 2: prove data access and network access independently
Create a small Blob Storage container holding synthetic files. Define an intended read-only data identity and a second identity that should not have access. Test the permitted read and the prohibited read while the network path is known to work. Do not grant broad subscription management rights merely to simplify a data-plane test.
Then change a controlled network condition. If using private endpoints is within the training budget, create the supported endpoint and configure the relevant private name resolution and VNet access. Otherwise, use a restricted public-access rule in the disposable environment to demonstrate the same distinction. A client with correct identity permissions may still fail because its network path is blocked or its DNS target is wrong.
Capture the source, resolved endpoint, request result and authorization decision. The presence of a private endpoint icon is not evidence that every client is using the private route. Use a client in the relevant network and verify the expected address and traffic path instead of relying solely on a test from an administrator workstation.
The Azure Storage Explorer workflow can help inspect a test object, provided the access method used is recorded and appropriately scoped. Remove temporary shared links and credentials at the end. Your lab passes when the permitted identity succeeds through the intended path and the prohibited case remains denied.
Lab 3: trace traffic through a virtual network
Build one small virtual network with two appropriate subnets and an inexpensive disposable service. Record the allowed traffic flow from a test client and one flow that should be blocked. Establish a baseline before adding or modifying a network security group. If the baseline already fails because the service is stopped or the route is missing, fix that unrelated issue before treating the result as evidence about filtering.
Apply a narrowly scoped rule and test both cases. Capture the source address, destination, protocol, timestamp and whether the enforcement point reports an allow or deny. An application timeout alone is ambiguous: DNS, routing, guest firewall and service health can produce similar symptoms. Correlate observations rather than declaring success because the application became unreachable.
Use network security group behavior to explain which rule applied and why. Do not open the service broadly to the internet just to make a demo easier. Remove the test rules and verify normal allowed traffic again during cleanup.
Lab 4: connect compute deployment to workload verification
Deploy a minimal VM or managed application using an authorized, low-cost configuration. Before deployment, write the intended operating state: the application should serve a specific synthetic request; one administrative action should be permitted; an unauthorized client should not gain management access; and the service should emit enough evidence to diagnose failure.
Deploy the workload and verify configuration separately from application behavior. The resource may show that provisioning succeeded while the guest service, container process or application endpoint still fails. Check the right layer before changing capacity or networking. A storage dependency problem cannot always be corrected by resizing the compute service.
Repeat the deployment with a small declarative template if appropriate, capturing expected resource properties and comparing them with actual state. Avoid embedding secrets in a Bicep or ARM template. The demonstration is useful when it shows which settings are reproducible and which runtime behaviors still require independent tests.
Lab 5: turn a metric into an actionable operational signal
Select a harmless test failure, such as temporarily stopping a disposable application or causing a synthetic request to return a known error. Determine which log entry, application event or resource metric can reveal the condition. Set up an appropriate temporary diagnostic configuration and record the collection settings so a missing result can be interpreted correctly.
Use Azure Monitor to inspect the events and, where suitable, create a test alert. A CPU threshold may not identify a failed customer request. An Activity Log entry may explain a management change without proving how an application behaved. State the question the signal answers and the important failures it would miss.
Generate a positive and a negative case. A test failure should produce the intended observation, while a healthy synthetic request should not generate the same alert without justification. Record any ingestion delay and avoid interpreting no alert as proof that all services are healthy. Disable temporary diagnostics and alerts that would create cost or noise after the lab.
Lab 6: prove recovery rather than backup existence
Create a small disposable data artifact, document its expected contents and configure a permitted backup or recovery method in the lab. Record the restoration point and protection settings. Then alter or delete only the synthetic test data and restore it to a separate safe target if possible.
Verify that restored content matches the expected state and that an authorized test workload can access it. A successful backup operation is not enough if the restored artifact is inaccessible because of a missing key, identity, network path or dependent configuration. The Azure Backup concepts should be evaluated against actual recoverability and the exercise’s recovery objective.
Record which failures the chosen protection would not address. Replication may carry a deletion forward; a snapshot may protect a prior state but require additional restoration work. The exercise should end with a clear statement of achievable recovery time and possible data loss under the tested conditions, not an invented guarantee of business continuity.
Integrate the labs into one administrator incident
For a final controlled exercise, modify a disposable storage access condition so a test application cannot read a needed object. Provide the learner with identity, application, route and diagnostic observations. Ask them to establish the expected transaction, identify the failing layer, make a narrowly approved correction and retest both access and security.
The learner should explain why an application might fail even when its resource deployment is complete; why a successful sign-in does not establish data-plane authority; and why a working private endpoint still needs correct name resolution. These are AZ-104 operating principles, not shortcuts derived from remembered exam items.
A useful lab portfolio consists of the requirement, evidence, change, outcome, limitation and cleanup record for each exercise. Practicing this way improves administrator judgment without requiring a production incident or claiming that one successful test proves mastery of every Azure service.