AZ-400 Question 293
Single answerYou are managing a microservices-based application running on Azure Kubernetes Service (AKS). Each microservice is instrumented using Azure Monitor and Application Insights. Recently, users have reported increased latency for specific features during peak usage. After viewing basic metrics such as CPU and memory usage on the cluster, you notice no obvious resource bottleneck. Which approach should you adopt next to identify the root cause of the latency?
- A
Enable automatic horizontal pod autoscaling in AKS to add more nodes when CPU usage spikes
- B
Use Application Insights to enable distributed tracing and correlate requests end-to-end
- C
Add custom performance counters in the container's base image to track OS-level resource usage
- D
Configure a pipeline task that restarts all slow-responding pods in the cluster
Show answer and explanation
Correct answer: B
Explanation
When investigating latency issues in a microservices application, you need visibility into which service or operation contributes to the slowdown. Distributed tracing in Application Insights offers an end-to-end view of requests through each microservice, capturing detailed performance data and correlation logs. This is a recommended best practice as described in the official Azure documentation for Application Insights (https://learn.microsoft.com/azure/azure-monitor/app/distributed-tracing), enabling you to pinpoint performance bottlenecks and address them effectively.
- A. Incorrect.
Incorrect. While autoscaling can help handle increased load, it does not directly provide deeper insights into which microservice or component is causing latency. This option focuses on resource scaling rather than root-cause analysis.
- B. Correct.
Correct. Distributed tracing in Application Insights allows you to trace a request across multiple services, identifying performance bottlenecks in specific microservices or external dependencies. This is the best next step for diagnosing latency issues.
- C. Incorrect.
Incorrect. Custom performance counters at the OS level might provide some insight, but they often do not offer a clear view of distributed call paths or service-level bottlenecks. OS-level metrics alone may not reveal which microservice is slowing down requests.
- D. Incorrect.
Incorrect. Automatically restarting pods might temporarily alleviate symptoms but does not help you analyze or diagnose the underlying root cause. It is more of a reactive approach and does not address the core issue.