Microsoft AZ-104 Exam Dumps, Practice Test Questions

A probe path determines backend health, not the URL path used to dispatch each client request.

Two backend pools do not supply the missing HTTP path matching feature. A Layer 7 gateway is needed to choose by URL path.

Azure Load Balancer matches transport fields, not decrypted HTTP host or URL paths; it cannot use those values for backend selection.

Health probes only assess backend health and cannot forward the requested SSH session.

A load balancing rule distributes eligible flows among backend instances, which is the opposite of reaching one designated maintenance VM.

Application Gateway URL routing operates at HTTP/HTTPS Layer 7, not arbitrary SSH TCP port forwarding.

An outbound rule governs SNAT of connections initiated by backend instances; it does not distribute inbound traffic to a healthy NVA backend pool. An internal Standard Load Balancer HA Ports inbound load-balancing rule targets all supported TCP/UDP ports.

A conventional load-balancing rule targets configured protocol/port mappings. It does not automatically include all the NVA’s dynamic ports.

Inbound NAT rules forward to selected instances and still require port-specific planning; they do not implement a single all-port balanced backend service.

A TCP probe verifies connectivity to the port, not HTTP response semantics. It cannot conclude that an application route succeeds end-to-end.

TCP port probes are supported. The issue is that the selected protocol provides a weaker signal than an HTTP/HTTPS application health endpoint.

Load balancer health depends on the configured probe, which may be limited to port reachability. Application errors can remain invisible if the probe does not exercise them.

HA Ports is an inbound transport rule that covers multiple TCP/UDP service ports for applications such as network virtual appliances. It does not itself expand the ephemeral SNAT port allocation of a Load Balancer outbound rule. The observed failure is outbound port exhaustion and requires changes to outbound translation capacity, not extra inbound port matching.

A health probe determines whether a backend can receive new incoming load-balanced flows. More frequent probes add health checks and cannot create new ephemeral outbound source ports. The fleet is already reaching outbound limits while sending Internet traffic; tune the SNAT allocation and frontend capacity rather than probe timing.

Client IP affinity changes how inbound flows from one source address are distributed across backend servers. It has no guarantee of increasing outbound translation port inventory. Outbound SNAT sizing is configured separately via outbound rules and available frontend public IPs, not through the inbound load-distribution hash mode.

Client IP affinity uses network-layer source/destination address information, not an HTTP Cookie header. When many users share a proxy or NAT IP, the load balancer sees them as the same client address and cannot distinguish individual browser sessions. It may offer IP-based persistence but does not satisfy the specified per-user cookie-aware behavior.

Floating IP controls packet destination address translation for scenarios such as clustered listeners and port reuse. It does not turn a Layer 4 load-balancing rule into a Layer 7 parser or inspect cookies from application requests. The manager has confused direct-server-return mapping with application session affinity; the selected backend cannot be cookie-pinned that way.

An HTTPS health probe checks whether an application endpoint is healthy enough to receive new flows; it does not inspect each user request or maintain a cookie-to-backend lookup. Selecting a secure probe protocol changes backend eligibility checks, not the data-plane Layer 4 distribution algorithm or its lack of application-layer cookie parsing.

Diagnostic logs may support deeper investigations but impose log collection and query effort when a native transaction count metric already supplies the required aggregate time series.

Cost forecasts project billing trends rather than count requests in five-minute intervals. The chart would answer a different operational question.

Activity Log records management operations like configuration changes, not the data-plane transaction-count metric needed for interval utilization.

A subscription name is management metadata and not the advertised metric dimension identifying the storage operation category.

Turning off logs does not isolate a dimension value in metrics. It also removes other useful investigative evidence.

Replication configuration changes durability and may create risk but does not provide the missing dimensional breakdown of the existing latency metric.

Activity Log tracks management-plane events, not the continuous CPU utilization series needed to inspect the current workload.

The Perf table depends on appropriate guest performance collection into the workspace. With no guest collection configured, it is not the direct source of a current platform VM CPU series. Azure emits the VM Percentage CPU platform metric independently; the Metrics chart is the immediate appropriate data source for the requested live workload overview.

Guest memory counters describe memory use inside the VM operating system; they are not the direct platform CPU utilization signal requested. The CPU Percentage metric is already emitted by Azure for the VM without custom guest-event collection and can be charted immediately. Guest collection adds a different data source rather than replacing the CPU chart.

Averaging across long windows will smooth short spikes and may hide the very saturation episodes the team needs to identify.

Count reports how many metric samples are present, not the highest utilization in the interval. It cannot reveal whether a brief CPU peak reached saturation.

Summing percentage observations across time generates a cumulative number that does not represent the maximum instantaneous utilization. It may mask or distort the peak the team needs.

Changing aggregation may show a total count but does not separate the response-code categories. The cause remains hidden in a combined series.

A resource filter narrows which resources contribute data but does not divide one API gateway’s observations by HTTP response code. The diagnostic dimension is the discriminating signal.

A duplicate unsegmented chart still mixes failures of different response codes. It cannot reveal whether 401, 429 or 500-class responses dominate.