Examkin

Soru

Zorluk: Çok zorCapacity Planning and Infrastructure Workload Scaling Optimization

A financial technology enterprise processes high-throughput transaction streams using Google Kubernetes Engine (GKE) and Cloud Spanner. During quarterly settlement windows, incoming event volume surges suddenly by 20×20\times baseline load within a 3-minute window. During recent events, the platform experienced packet drops and elevated latency because the GKE Horizontal Pod Autoscaler (HPA) reacted too slowly to CPU load, and worker node scaling was constrained by regional Compute Engine quota limits. Which TWO capacity planning and workload scaling optimization strategies should you implement to resolve these performance bottlenecks? (Select TWO.)

  1. Deploy low-priority pause pods to maintain a pre-allocated buffer of node capacity, and reconfigure HPA to autoscale based on unconsumed Pub/Sub message queue depth.Cevap
  2. Request regional Compute Engine CPU and API quota adjustments in advance of scheduled traffic spikes, and implement client-side connection pooling to handle Cloud Spanner session growth.Cevap
  3. C
    Decrease the GKE HPA target CPU utilization threshold to 20% across all worker deployments to force faster pod creation during traffic bursts.
  4. D
    Rely on dynamic cluster autoscaling to automatically expand GCP regional resource quotas during demand spikes without submitting advance quota increase requests.

Cevap

The correct strategies are to deploy low-priority pause pods with custom Pub/Sub queue-depth HPA metrics, and to request regional Compute Engine quota increases in advance while implementing database connection pooling.
To absorb sudden 20×20\times traffic spikes, capacity planning requires both compute readiness and proactive quota management. Implementing low-priority pause pods creates a pre-provisioned pool of VM resources that high-priority application pods can immediately reclaim without waiting for new VM startup times. Reconfiguring HPA to scale on queue depth (such as Pub/Sub unconsumed message count) ensures pod scaling initiates as soon as work accumulates. Concurrently, submitting regional quota requests before peak events guarantees that underlying GCP compute limits do not block scale-out, while database connection pooling protects Cloud Spanner from session exhaustion.

Adım Adım Çözüm

1
Analyze pod autoscaling bottlenecks during rapid ingestion bursts.
Identified that CPU utilization is a lagging indicator for sudden queue-based traffic spikes.
Scaling microservices based on unconsumed Pub/Sub message queue depth triggers HPA before CPU load accumulates.
2
Address GKE node provisioning latency for rapid scale-out.
Implemented overprovisioning via low-priority pause pods.
Pause pods reserve compute capacity on VM nodes; when actual work pods need resources, pause pods are evicted instantly, bypassing GKE VM creation delays.
3
Evaluate infrastructure quota boundaries.
Identified that Compute Engine regional quotas act as hard ceilings during automated scaling.
Quota increases must be requested proactively from Google Cloud prior to predictable peak demand events.
4
Optimize database connectivity scaling.
Added connection pooling and session management for Cloud Spanner clients.
Prevents downstream database connection saturation as worker pod count expands rapidly.

Anahtar Kavram

Capacity Planning, Preemptive Overprovisioning, and Custom Metric Autoscaling in GCP
Bu soruyu puanla

Konular

Capacity Planning and Infrastructure Workload Scaling Optimization