Skip to content
THE EXAMTOPICS EDITORIAL Explore practice exams

MICROSOFT

Microsoft AZ-104: Running Azure Well — Administrator Playbook

A practical Azure operations playbook for access reviews, change management, incident evidence, network troubleshooting, monitoring, recovery and cloud operating routines.

Administering Azure is easiest on the day an environment is deployed. The deeper work begins afterward, when access changes, services are upgraded, traffic grows, alerts fire, and a team needs to recover data without guessing what was protected. Mature cloud operations are built around reliable routines, good evidence and controlled changes rather than individual administrators remembering where every setting lives.

Microsoft AZ-104 covers much of this operational ground, including identity and governance, storage, compute, virtual networking, and monitoring or maintenance. Those subjects form a useful learning foundation, but the work of running Azure well extends beyond passing a test. This article focuses on how an administrator applies those skills day after day: protecting access, coordinating changes, diagnosing failures, managing capacity and proving that recovery will work.

Start with an operating inventory, not a portal dashboard

An administrator needs to understand which resources exist, why they exist, who owns them and what would fail if they disappeared. A resource inventory should show subscriptions, resource groups, important applications, network dependencies, data stores, business owners and the associated recovery requirements. Tags can help with organization and cost reporting, but they are only as useful as the conventions that keep them accurate.

Different resources have different operational importance. A disposable development VM may be acceptable to replace from a template. A production identity dependency or storage account supporting regulated data has more stringent change and recovery considerations. Treating all resources equally creates unnecessary effort in some places and underinvestment in others. The point of an inventory is to make those differences explicit.

A good initial question is, “What would the service owner notice if this resource stopped working?” That answer helps identify which availability checks, access controls, alerts and recovery procedures deserve priority. It also reveals the scope of a change. A seemingly harmless NSG update can affect multiple applications if it is applied to a shared subnet.

Make identity and governance routine maintenance

Access is not a one-time provisioning step. People join teams, change roles and leave; automation identities are created for new workloads; and temporary permissions can remain after a project ends. Build a regular process for reviewing Azure RBAC role assignments, administrative groups, privileged accounts and the scopes at which access is granted. An Owner assignment at subscription scope has consequences far beyond a single resource group’s requirements.

Microsoft Entra administration and Azure resource authorization are related but separate. A user may exist in the directory yet lack rights to operate a storage account or VM. Conversely, a powerful Azure resource role does not automatically permit every tenant-wide identity action. Explain the difference whenever investigating an access failure instead of casually increasing permissions.

Governance should constrain unsafe changes without stopping legitimate work unnecessarily. A policy that denies deployment in an unapproved region may be justified by a data-residency requirement. A resource lock can guard against accidental deletion, but it can also interrupt maintenance if applied without understanding the service’s update behavior. Record why a control exists, what its scope is and how exceptions are reviewed. Evidence matters because an administrator who cannot explain a restriction cannot safely maintain it.

Plan changes so their consequences are observable

Uncontrolled portal edits are difficult to reproduce and compare across environments. Use ARM or Bicep definitions where appropriate, keep parameters and deployments under review, and test changes against representative nonproduction resources. Infrastructure as code reduces certain configuration differences, but it does not automatically detect poor resource sizing or protect against every incorrect parameter.

Before a change, write down the expected effect, the affected resources, the signals that will confirm success and a rollback or mitigation approach. A deployment slot can reduce risk for supported App Service releases, while a template redeployment may require a different rollback plan. A virtual machine disk change also has different consequences from changing a network route. Choosing the mechanism depends on the resource and the failure model, not on a universal deployment checklist.

After the change, compare observed state to intended state. Did the resource accept the configuration? Did the application continue to serve requests? Did permissions, DNS or upstream dependencies change? This is where Bicep, ARM and administrative automation become operational tools rather than exam vocabulary. A repeatable deployment whose impact is not checked is still a risky deployment.

Monitor for decisions, not merely for data collection

Metrics and logs have different strengths. CPU and latency metrics can help reveal resource pressure quickly; log records may explain why a request failed or which change preceded the failure. Azure Monitor can collect these signals, and alert rules can connect them to action groups and operating processes. The challenge is deciding which signal actually matters to the service owner.

For a customer-facing application, response errors and service availability are often more meaningful than a raw host CPU value. For a batch import, completion time and data quality may matter more than a momentary resource spike. For storage, unauthorized access attempts, capacity and operational errors can be important indicators. Configure alerts around service outcomes and known risks, then test their delivery path. An alarm nobody receives or understands is not an effective operational safeguard.

When a problem occurs, inspect evidence across components. An application can fail because its managed identity lost access to data, its private endpoint name resolves unexpectedly, its VM is out of resources, or a deployment altered configuration.

Use a layered approach to connectivity incidents

Suppose a VM can reach one internal service but cannot access a storage account. There are several possible explanations, and changing the NSG first may be wrong. Record the source and destination, resolve the intended hostname, inspect routes and effective security rules, determine whether access uses a public endpoint or private endpoint, and confirm the identity and permissions used for the request.

Name resolution and network reachability must be tested separately from authorization. A request denied with an explicit permission error should send you toward role assignments or tokens; a timeout might indicate an endpoint, routing, firewall or DNS problem. A load-balanced application may be healthy at the VM but considered unhealthy because its probe points to a failing port or path. The operator’s aim is to isolate the first failing layer and make the smallest justified correction.

Azure Network Watcher and Connection Monitor can provide evidence for selected connectivity investigations. They do not replace understanding the route or guarantee that an application is authorized to use a downstream service. Keep a short incident timeline: symptom, suspected cause, test, result, correction and confirmation. A consistent method is faster in the long run than a series of broad network openings made in desperation. DNS, peering and routing provides the relevant technical background.

Recovery confidence comes from a restore test

Backup policies should be designed from the actual recovery requirements. How much recent data could the business afford to lose? How long may a service remain unavailable? Which components must be brought back in which order? A successful scheduled backup is useful evidence, but it does not by itself answer these questions.

Practice recovery with an appropriately disposable workload. Check whether the required restore point exists, perform a supported restore, verify the data, and identify dependencies such as identities, networks, DNS records and application configuration. Distinguish the capabilities of Azure Backup and Azure Site Recovery. Backup can support recovery from data loss under its supported scenarios, while Site Recovery addresses supported disaster recovery and failover workflows. Redundancy and replication are not comprehensive substitutes for item-level restoration.

Record what an operator actually did during the restore and what required manual intervention. Those details inform the runbook more than a screen showing a green backup job. A service may technically be recovered but remain unavailable to users because a private endpoint, DNS zone or secret was not reconnected correctly.

Give operations a sustainable cadence

Not every control needs to be reviewed at the same frequency. A practical cadence depends on business risk, rate of change and automation. The following is an example, not a Microsoft-prescribed schedule.

Cadence Sample operational work Evidence of completion
Daily or per shift Review high-priority alerts, failed jobs, service health and important deployment changes Resolved incidents, acknowledged exceptions and escalation records
Weekly Check recent access or configuration changes, storage protection failures, cost trends and open capacity risks Reviewed findings with owners and assigned remediation actions
Monthly or by risk Validate scoped permissions, test sample restores, review recurring alerts and examine policy exceptions Restore evidence, access-review decisions and revised operating documentation
Before major releases Review dependencies, change scope, available recovery options and post-deployment signals Approved change plan, observed result and mitigation record

These routines should be adapted to service criticality. A platform team running sensitive production applications may need continuous monitoring and automated policy evaluation; a training environment may not. Good administration means making frequency and ownership proportional to risk instead of creating paperwork that no one uses.

From Azure skills to professional responsibility

An Azure administrator should be able to explain why a policy exists, why one hosting option was chosen, what evidence indicates a problem and how a change will be validated. Communication matters because decisions involve other teams and business priorities. The person who can translate “the application is broken” into a precise hypothesis about DNS, permissions or deployment state helps everyone respond more effectively.

The archived AZ-104 skills outline remains useful as a technical checklist, subject to the objectives effective for your exam date. In practice, continue beyond the assessment: keep deployments understandable, review access, learn from incidents, test recovery and improve the quality of operating evidence after each change.

Mastery is not the ability to remember every Azure setting. It is the ability to protect an environment’s intended behavior while requirements and technologies change. That is the professional skill the certification can help develop, and it is the daily operational habit that keeps cloud services useful.

Continue exploring the technologies and certification concepts that matter most to you.