Microsoft AZ-104 Exam Dumps, Practice Test Questions

Process environment variables are set when the container starts. Updating the shared secret definition does not automatically replace their in-memory values.

Traffic splitting changes request routing across revisions; it does not restart the existing container process or refresh an already-read environment variable.

Changing the maximum replica limit does not force all existing containers to restart or reload their process environment. After rotating an app secret used via environment variables, restart the active revision or deploy an appropriate new revision.

A zero traffic allocation stops HTTP routing to a revision but does not necessarily stop its active containers or background workloads. Deactivate that specific revision to stop its replicas while retaining rollback ability.

Ingress settings are application-scoped and changing them doesn't selectively deactivate one particular revision's containers. It could alter routing to every active revision while leaving the unused revision active; use the revision lifecycle control instead.

Revision labels are routing metadata, not its lifecycle state. A revision left active may keep replicas running even if it no longer has a label or production traffic.

A higher maximum only permits additional replicas when events arrive. With the minimum still zero, scaling can return the app to no running replicas during quiet periods. Set minReplicas to at least one to maintain the warm connection.

A zero maximum cannot provide an always-on worker and may be invalid when configuring scaling bounds.

Removing the trigger does not establish a guaranteed always-running replica. The minimum value controls that requirement explicitly.

Changing the traffic weights would expose unapproved code to ordinary users. The requirement is a targeted QA route.

Application-level ingress affects availability of endpoints. Disabling it does not create the selective test path requested.

A query-string label on the shared Container App URL is not a revision-routing primitive. The shared hostname still follows configured production traffic weights; an assigned revision label exposes its own stable hostname for QA to reach A without redirecting ordinary traffic.

The container-app revision mode, not unrelated storage network rules, governs which revision is active after image deployment.

Container image settings belong to the revision template. Image changes are revision-scoped and can cause a new revision.

Multiple revision mode exists for controlled concurrent versions and traffic splitting. The stated app is in the narrower single revision mode.

A startup probe protects slow initialization from premature liveness restarts. If readiness already passes while the model is still loading, that signal alone does not prevent requests from reaching a replica that cannot yet answer them. Readiness must reflect application serving capability.

A delayed liveness check can avoid restarting a healthy startup, but it is not a traffic-admission signal. A replica can still receive and fail user requests while liveness is delayed if readiness reports success too early.

Liveness tests detect unrecoverable unhealthy processes; an aggressive probe can create restart loops instead of waiting for normal application startup.

More replicas repeat the same listener/target mismatch. Horizontal scale cannot make a process listen on an unconfigured port.

EXPOSE is image metadata and does not cause the process to listen on 8080. With the application actually bound to 5000, ingress targetPort must target 5000 or the process itself must be reconfigured to listen on 8080.

Disabling ingress removes the intended public routing path. It does not expose the container process on the correct port.

A startup probe validates initialization and stops running once startup succeeds. It does not monitor a later deadlock; a liveness probe must continuously check whether the process remains healthy.

A readiness failure typically removes a container from serving requests; it does not express the same restart-on-unhealthy behavior as liveness.

Traffic weighting distributes requests among revisions and does not establish a failed-process restart policy. A liveness failure is the signal to restart an unresponsive container.

Managed identity supports resource authentication and is not the cost ceiling for scaling. Disabling it may break access to the queue rather than bound replicas.

A minimum defines a floor, not an upper bound. The app could still scale above the permitted instance count during large backlogs.

Message expiration changes workload data behavior, not the number of Container App replicas allowed. It can discard messages while scaling remains uncontrolled.

Scaling up moves to a larger per-instance plan SKU, which changes worker resources rather than the requested number of similarly sized instances.

More threads can change concurrency inside a single worker but do not add the independent App Service plan instances that the capacity requirement specifies.

A larger worker can help memory-bound operations but is a scale-up change, not additional workers of the current SKU. That differs from the requested horizontal capacity.

More workers spread requests across instances but each instance still has the original memory limit. A request that is too large for one worker can continue failing.

Autoscale adds parallel instances of the same SKU. A memory-heavy request still runs within one worker and can exhaust its unchanged RAM allocation.

The process architecture can alter address-space limits, but does not guarantee the underlying plan has more physical RAM. The case identifies the per-worker memory SKU as insufficient.