PCDOE Optimizing service performance Practice Test 9 — 15 Questions

Question 1

Your team runs a Java microservice on GKE. The service's p99 latency is high, and you suspect that long GC pauses are the cause. You want to identify the specific heap allocation patterns causing the pauses without restarting the running pods. Which approach should you take?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Enable Cloud Profiler with the Heap profile type for the service, then inspect the flame graph to identify the allocation hot spots.. Cloud Profiler's heap profile type is designed to collect allocation data from live Java services with minimal overhead, showing which functions allocate the most memory. This directly identifies the allocation hot spots that drive GC pauses, enabling targeted code fixes without restarting pods. Other tools either pause the JVM, lack allocation detail, or only show symptoms.

Answer

Configure a liveness probe with a longer timeout and increase the pod's memory limit to reduce GC frequency.

Answer

Run a jmap -histo:live command against each pod to capture a heap histogram, then compare the object counts across pods.

Answer

Enable Cloud Trace on the service and analyze the span durations for the slowest requests to pinpoint the GC pauses.

Question 2

A financial services company runs a microservices application on GKE. They use Cloud Monitoring to track performance. During peak hours, the frontend service experiences high latency, and the backend service shows increased error rates. You need to identify the root cause efficiently. Which approach should you take?

Accepted Answer

Use Cloud Trace to analyze the latency distribution across microservices and identify the bottleneck.. Cloud Trace provides end-to-end latency tracing across microservices, allowing you to see the exact path of requests and identify where delays occur. It is the most efficient tool for diagnosing performance bottlenecks in distributed applications, especially during peak load when multiple services interact.

Answer

Set up an uptime check for the frontend service and configure alerting policies for error rates.

Answer

Enable Cloud Profiler on all services to collect CPU and heap profiles.

Answer

Create a custom dashboard in Cloud Monitoring that displays CPU and memory usage for all pods.

Question 3

You manage a GKE cluster running a latency-sensitive service. You notice that the Horizontal Pod Autoscaler (HPA) frequently oscillates between scaling up and down, causing performance degradation during scale-down periods. The HPA is configured with a CPU utilization target of 80%. Which action should you take to stabilize the HPA behavior?

Accepted Answer

Increase the HPA's --horizontal-pod-autoscaler-downscale-stabilization flag value on the kube-controller-manager.. The downscale stabilization window is the HPA parameter that controls how long the autoscaler waits before scaling down after a decrease in load. Increasing this value makes the HPA less reactive to short-lived dips, reducing oscillation. Other options either do not affect HPA behavior or address unrelated concerns like disruption budgets or resource recommendations.

Answer

Add a Vertical Pod Autoscaler (VPA) in recommendation mode to adjust CPU requests based on actual usage.

Answer

Set the HPA's targetCPUUtilizationPercentage to 50% to provide more headroom for scaling decisions.

Answer

Configure a PodDisruptionBudget with minAvailable set to 80% to ensure enough pods remain during scale-down.

Question 4

A DevOps team manages a GKE cluster that runs a stateful application with persistent volumes. They observe that during scaling events, new pods take a long time to become ready, impacting performance. Which two actions should they take to optimize pod startup time? (Choose two.)

Accepted Answer

Pre-pull container images on nodes using a DaemonSet to reduce image pull time.. Pre-pulling container images with a DaemonSet eliminates image download time during pod startup, and using high-performance PersistentVolumes reduces I/O latency during application initialization. Both actions directly target common bottlenecks in stateful pod startup, leading to faster readiness and improved scaling performance.

Answer

Use a StatefulSet with a headless Service to ensure stable network identities and ordered deployment.

Answer

Increase the CPU and memory requests for the pods to ensure they are scheduled quickly.

Answer

Configure readiness probes with appropriate initialDelaySeconds and periodSeconds to avoid premature traffic.

Question 5

Your team runs a web application on Compute Engine instances behind an external HTTP(S) Load Balancer. Users report slow response times for static assets. You want to reduce latency by caching content at Google's edge. Which service should you enable?

Accepted Answer

Cloud CDN with the external HTTP(S) Load Balancer.. Cloud CDN is the Google Cloud service that caches HTTP(S) content at edge locations. It is enabled on external HTTP(S) Load Balancers and reduces latency for static assets by serving them from close to users. The other services provide connectivity, security, or different load balancing, but none offer edge caching for web content.

Answer

Cloud Armor with a rate-based ban rule to block high-latency clients.

Answer

Cloud Interconnect to connect on-premises users directly to the VPC.

Answer

Cloud Load Balancing with a TCP proxy load balancer to optimize packet routing.

Question 6

A startup runs a web application on GKE. They notice that during traffic spikes, the application becomes unresponsive, and they want to automatically scale the number of pods based on CPU utilization. Which Kubernetes resource should they use?

Accepted Answer

Horizontal Pod Autoscaler (HPA). The Horizontal Pod Autoscaler (HPA) is designed to automatically scale the number of pods in a deployment based on metrics like CPU utilization. It directly addresses the need to handle traffic spikes by adding more replicas when demand increases, ensuring the application remains responsive.

Answer

Pod Disruption Budget (PDB)

Answer

Cluster Autoscaler

Answer

Vertical Pod Autoscaler (VPA)

Question 7

Your team runs a Java-based microservice on Google Kubernetes Engine (GKE). The service uses a Java Virtual Machine (JVM) heap of 4 GiB and is configured with a container memory limit of 5 GiB. During load testing, the pods are frequently OOMKilled. The JVM flags do not include any container awareness settings. What is the most likely cause of the OOMKilled events?

Accepted Answer

The JVM is not container-aware and sizes its heap based on the node's total memory, causing it to exceed the container limit.. The JVM, without container awareness, often allocates heap based on the host's total memory, not the cgroup limit. This can cause the JVM to exceed the container's memory limit, leading to OOMKilled. Setting explicit JVM flags like -XX:MaxRAMPercentage or -Xmx to a value below the limit ensures the JVM respects the container constraints. The other options do not directly cause OOMKilled in this context.

Answer

The GKE node has insufficient memory, causing the kernel to kill the container to free up resources.

Answer

The Kubernetes liveness probe is misconfigured, causing the kubelet to restart the pods and report OOMKilled.

Answer

The container memory limit is set too low because the JVM heap is 4 GiB, leaving only 1 GiB for non-heap memory, which is insufficient.

Question 8

A company runs a GKE cluster with a mix of latency-sensitive and batch workloads. They want to ensure that latency-sensitive pods are prioritized during resource contention. Which approach should they take?

Accepted Answer

Assign higher priority to latency-sensitive pods using PriorityClasses and enable preemption.. PriorityClasses with preemption allow you to assign higher priority to latency-sensitive pods. When resources are insufficient, the scheduler can evict lower-priority pods to accommodate higher-priority ones. This ensures that critical, latency-sensitive workloads are prioritized during contention, maintaining their performance.

Answer

Use node affinity to schedule latency-sensitive pods on dedicated nodes.

Answer

Configure taints and tolerations to reserve nodes for latency-sensitive pods.

Answer

Set higher CPU and memory requests for latency-sensitive pods to ensure they get scheduled.

Question 9

Your organization runs a stateful application on Google Kubernetes Engine (GKE) using a StatefulSet with a PersistentVolumeClaim (PVC) backed by a Compute Engine persistent disk (pd-standard). Users report high latency during write-heavy operations. You want to improve write performance. What should you do?

Accepted Answer

Change the StorageClass to use pd-ssd and recreate the StatefulSet.. The write latency is likely due to the use of pd-standard, which has lower IOPS and throughput. Upgrading to pd-ssd (SSD persistent disk) provides higher IOPS and lower latency for write-heavy workloads. Because the StorageClass of a PVC cannot be changed after creation, you must recreate the StatefulSet and its PVC with the new StorageClass. Other options either do not address write performance or could increase latency.

Answer

Increase the size of the persistent disk to improve IOPS.

Answer

Use a regional persistent disk instead of a zonal persistent disk.

Answer

Enable read cache on the persistent disk to speed up writes.

Question 10

Your team manages a high-traffic API on Google Kubernetes Engine (GKE). You want to reduce latency for users by caching responses at the edge. The API responses are highly dynamic and personalized per user. Which approach should you take?

Accepted Answer

Deploy a Memorystore for Redis instance and cache API responses in the application layer.. For dynamic, personalized API responses, edge caching via Cloud CDN is ineffective because the content varies per user. Instead, caching at the application layer with Memorystore for Redis allows storing per-user data with low latency, reducing backend load and improving response times. Session affinity and CDN with authorization headers do not provide caching benefits for personalized content.

Answer

Use Cloud CDN with cache keys based on the user's session cookie.

Answer

Enable Cloud CDN with a short TTL and vary on the Authorization header.

Answer

Use GKE Ingress with a backend service that has session affinity enabled.

Question 11

Your team runs a microservices application on GKE. You want to reduce inter-service latency by ensuring that traffic between services stays within the same zone and avoids unnecessary hops. Which GKE feature should you configure?

Accepted Answer

Topology-aware routing for Services.. Topology-aware routing is a GKE feature that directs traffic to endpoints in the same zone as the client, minimizing latency and cross-zone costs. It is configured per Service and is the correct choice for keeping inter-service traffic local. Network policies, Cloud CDN, and Istio ingress gateway serve different purposes and do not address zone-local routing for internal services.

Answer

Cloud CDN to cache responses between services.

Answer

Istio ingress gateway to manage incoming traffic.

Answer

Network policies to restrict traffic between namespaces.

Question 12

A DevOps team manages a high-traffic API on GKE. They use Cloud Monitoring to track performance. They want to detect subtle latency regressions after each deployment and automatically roll back if the 99th percentile latency exceeds a threshold. Which approach should they use?

Accepted Answer

Use Cloud Deploy with a canary deployment strategy and configure an analysis job that queries Cloud Monitoring for latency metrics.. Cloud Deploy provides canary deployment strategies with automated analysis. By defining an analysis job that queries Cloud Monitoring for the 99th percentile latency, Cloud Deploy can automatically compare the canary's performance against the baseline and trigger a rollback if the threshold is exceeded. This is a fully managed, integrated solution that requires minimal custom code. The other options either require manual intervention, lack automation, or do not provide rollback capabilities.

Answer

Create a Cloud Monitoring alerting policy on the 99th percentile latency metric and trigger a Cloud Function to roll back the deployment.

Answer

Implement a custom CI/CD pipeline that uses Prometheus to scrape latency metrics and triggers a rollback via kubectl.

Answer

Configure a GKE Horizontal Pod Autoscaler based on the 99th percentile latency metric.

Question 13

A media company serves video thumbnails from a Cloud Storage bucket through an external Application Load Balancer with Cloud CDN enabled. Analytics show a cache hit ratio near 20 percent for image objects, even though the same images are requested repeatedly by millions of users. Response headers show Cache-Control: private, max-age=600 on the objects. You need to raise the hit ratio without changing application code. What should you do?

Accepted Answer

Change the object metadata to Cache-Control: public, max-age=86400 and invalidate the cached entries.. Cloud CDN honors origin Cache-Control semantics, and a private directive explicitly forbids shared caches from storing the response. Thumbnails are identical for all users, so they should be publicly cacheable with a long max-age. Updating object metadata and invalidating the previously stored entries makes the objects eligible for edge caching and immediately reflects the new policy, lifting the hit ratio without touching application code.

Answer

Increase the backend bucket's capacity and set a longer default TTL in the load balancer's backend configuration.

Answer

Move the thumbnails into a regional bucket in the same region as the load balancer to reduce origin latency.

Answer

Enable negative caching on the backend bucket so misses are remembered and revalidated less often.

Question 14

A company runs a stateful application on GKE using persistent disks. Users report slow database queries during peak hours. The operations team suspects disk I/O bottlenecks. They want to identify the root cause and optimize performance. Which action should they take first?

Accepted Answer

Analyze disk I/O metrics in Cloud Monitoring to identify the bottleneck.. The first step in optimizing performance is to diagnose the issue using metrics. Cloud Monitoring provides detailed disk I/O metrics that can confirm whether the bottleneck is indeed disk-related and pinpoint the specific cause. Only after analysis should the team consider actions like increasing disk size or changing disk type. This approach avoids unnecessary changes and ensures that resources are used effectively.

Answer

Increase the size of the persistent disks to improve IOPS.

Answer

Enable GKE node auto-provisioning to add more nodes with local SSDs.

Answer

Switch to a higher-performance disk type such as SSD persistent disks.

Question 15

Your team runs a microservices application on Google Kubernetes Engine (GKE). You notice that some services experience increased latency due to frequent garbage collection pauses in the Java Virtual Machine (JVM). You want to optimize JVM performance in the containers. Which two actions should you take? (Choose two.)

Accepted Answer

Set the JVM's initial and maximum heap sizes to the same value to avoid heap resizing overhead.. To reduce GC pauses, setting -Xms and -Xmx to the same value avoids heap resizing overhead, and using G1GC with a tuned pause time target helps achieve low latency. Increasing memory limit alone may not help without JVM tuning, disabling JIT is harmful, and Serial GC is unsuitable for larger heaps. These two actions directly address the garbage collection pauses.

Answer

Set the JVM's garbage collector to Serial GC to minimize overhead.

Answer

Disable the JVM's Just-In-Time (JIT) compiler to reduce CPU usage.

Answer

Increase the container memory limit to allow the JVM to use more heap space.