mediumMultiple Choice
Autoscaling Cold Starts on Vertex AI
A company deploys a model on Vertex AI Prediction with autoscaling enabled. They notice that during a traffic spike, new instances take several minutes to become available, causing high latency. What is the best solution?
Quick Answer
The answer is to set a higher min replicas to maintain a baseline of warm instances. This is correct because autoscaling cold start solutions on Vertex AI hinge on the fact that new instances require several minutes to initialize and load the model, creating latency during traffic spikes. By raising the minimum number of replicas, you ensure a pool of pre-warmed instances is always ready to serve requests instantly, absorbing the initial surge while new instances spin up in the background. On the Google Professional Machine Learning Engineer exam, this question tests your understanding of Vertex AI Prediction’s scaling behavior and the trade-off between cost and latency. A common trap is to assume that reducing the scaling window or enabling faster provisioning solves the cold start problem, but those options do not eliminate the inherent initialization delay. Remember the memory tip: “Warm the pool, don’t cool the spike”—keeping a baseline of warm instances is the only direct way to prevent latency from cold starts.
⚠ Common exam trap
Google Cloud often tests the misconception that increasing max replicas or decreasing machine type solves cold-start latency, when the real solution is maintaining a warm baseline via min replicas.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Set a higher min replicas to maintain a baseline of warm instances
Setting a higher min replicas ensures that a baseline number of instances are always warm and ready to serve traffic. During a traffic spike, new instances still take time to provision (cold start), but the warm instances handle the initial surge without latency spikes. This directly addresses the observed high latency during spikes.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Disable autoscaling and use a fixed number of replicas
Why it's wrong here
Fixed replicas remove the elasticity that absorbs the spike, so latency persists and capacity is wasted off-peak; Vertex AI autoscaling already adds replicas, and the delay stems from slow instance start-up, which min replica counts or larger machines address. It tempts because predictable capacity avoids scaling lag, and it would be correct for steady, well-forecast load.
- ✗
Increase the max replicas setting
Why it's wrong here
Raising max replicas permits more instances but does not shorten the minutes-long cold start each new replica needs to load the model; pre-warming or a minimum replica count addresses that provisioning delay. It is tempting because max replicas caps burst capacity, which is the right lever when traffic exceeds the configured ceiling rather than when scale-out itself is slow.
- ✗
Decrease the machine type to reduce provisioning time
Why it's wrong here
A smaller machine type still requires the same container image pull and model load during provisioning, so cold-start latency persists; it may even worsen throughput per replica. Reducing machine size is genuinely useful for trimming cost on low-traffic, latency-tolerant endpoints, not for accelerating scale-out during a spike.
- ✓
Set a higher min replicas to maintain a baseline of warm instances
Why this is correct
Autoscaling adds instances reactively, and cold-start provisioning plus model loading takes minutes, so latency spikes before capacity arrives. Keeping a higher minimum replica count maintains pre-warmed instances that absorb the initial burst, directly addressing the stem's slow scale-out constraint.
Go deeper
Related to this question
About these practice questions
One of 775 original PMLE practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
Same concept, more angles
1 more way this is tested on PMLE
These questions test the same concept from different angles. Work through them to make sure you can recognise it however the exam phrases it.
Variation 1. A team deploys a model using Vertex AI Endpoint with automatic scaling. They observe that during traffic spikes, new instances take a long time to become ready, causing high latency for some requests. What should they configure to reduce this startup time?
medium- A.Increase the max replicas
- ✓ B.Use a custom container with a smaller footprint
- C.Enable predictive autoscaling
- D.Set a higher target CPU utilization
Why B: A custom container with a smaller footprint reduces image pull time and container initialization overhead, which are the dominant contributors to Vertex AI Endpoint replica startup latency during scale-out. Smaller images pull faster from Artifact Registry and start faster, so new replicas become ready sooner and absorb traffic spikes with less queuing delay.
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This PMLE practice question is part of Courseiva's free Google Cloud certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the PMLE exam.