Examkin

Question

Difficulty: HardAuto-scaling and Capacity Planning

A financial services organization operates a daily batch processing pipeline on a Compute Engine Managed Instance Group (MIG) consuming messages from Cloud Pub/Sub. Every morning at market open, incoming message rates surge by 10×10\times within 30 seconds. Because instance initialization and software warmup take approximately 4 minutes, dynamic auto-scaling policies based on Pub/Sub queue depth or CPU load experience severe queue backlogs and SLA violations during the first 10 minutes of the spike. Which architecture strategy should a Cloud Architect recommend to eliminate initial latency spikes while maintaining cost efficiency during off-peak hours?

  1. Implement scheduled auto-scaling rules on the Managed Instance Group to scale out baseline capacity prior to market open, combined with dynamic target tracking scaling based on Cloud Pub/Sub unacknowledged message depth.Answer
  2. B
    Configure the Managed Instance Group auto-scaler to use a very low CPU utilization target threshold (15%15\%) so scaling actions trigger immediately when traffic begins to rise.
  3. C
    Migrate the processing workload to a Google Kubernetes Engine (GKE) cluster with Cluster Autoscaler and Horizontal Pod Autoscaler (HPA) configured on default node pools.
  4. D
    Automate daily Cloud Quota API requests to temporarily increase regional vCPU limits 15 minutes before market open and release them afterward.

Answer

Implement scheduled auto-scaling rules on the Managed Instance Group to scale out baseline capacity prior to market open, combined with dynamic target tracking scaling based on Cloud Pub/Sub unacknowledged message depth.
Combining scheduled auto-scaling with queue-based dynamic auto-scaling solves both the predictable instant spike and off-peak elasticity needs. Scheduling ensures instances are provisioned and warm before market open, avoiding the 4-minute boot delay, while Cloud Pub/Sub queue depth scaling dynamically adjusts capacity for unexpected workload variations throughout the day.

Step-by-Step Solution

1
Analyze the workload scaling bottleneck
Identified that the 4-minute instance initialization delay causes latency spikes during sudden 30-second traffic surges at predictable times.
Reactive auto-scaling cannot provisions compute resources fast enough when workload initialization time exceeds traffic surge speed.
2
Evaluate proactive capacity planning strategies
Scheduled auto-scaling pre-provisions capacity before known spikes, guaranteeing warm instances are available at market open.
Pre-warming compute resources removes startup latency from the critical processing path.
3
Select metric for dynamic off-peak scaling
Cloud Pub/Sub unacknowledged message count provides an accurate measure of pending work for asynchronous queue processors.
Queue depth metrics reflect actual work backlog better than CPU utilization for I/O-bound or message-driven processing.

Key Concept

Combining scheduled auto-scaling for predictable spikes with custom metric queue-depth scaling for asynchronous workloads.
Estimated Time:2m 0s
Rate this question

Topics

Auto-scaling and Capacity Planning

More questions from this topic