ExpertMinds LogoExpertMinds
Hard1 markMultiple Choice
Domain 4.2: Managing GKE resourcesGKENode PoolsGPUsKubernetes Scheduling

GCP ACE · Question 35 · Domain 4.2: Managing GKE resources

Your GKE cluster currently runs on a node pool of e2-standard-2 machines. Your new machine learning workload requires nodes with attached GPUs. You cannot attach GPUs to the existing e2-standard-2 nodes.

Which TWO steps should you take to run the new workload without disrupting existing applications? (Select TWO)

Answer options:

A.

Edit the existing node pool to change the machine type to an n1-standard-4 and attach GPUs.

B.

Create a new node pool in the cluster using a machine type that supports GPUs (e.g., N1 series) and attach the required GPUs.

C.

Delete the existing GKE cluster and recreate it with GPU-enabled nodes.

D.

Configure your ML workload's Kubernetes Deployment with a nodeSelector or node affinity to ensure it is scheduled on the new GPU node pool.

E.

Use kubectl to manually SSH into the existing nodes and install the Nvidia drivers.

How to approach this question

Understand that node pools are immutable regarding machine types. To add new hardware, add a new pool. Then, use Kubernetes scheduling features to direct the right pods to the right pool.

Full Answer

Create a new node pool with GPU-enabled machine types, and use node selectors or taints/tolerations to schedule the ML workload on the new node pool.
In GKE, the machine type of an existing node pool cannot be changed. To introduce nodes with different hardware (like GPUs), you must create a new node pool within the existing cluster. This allows the cluster to have a mix of standard nodes and GPU nodes. To ensure that only the machine learning workload uses the expensive GPU nodes, you configure the workload's Deployment manifest with a `nodeSelector`, `nodeAffinity`, or `tolerations` to target the specific labels applied to the new GPU node pool.

Common mistakes

Thinking you can edit the machine type of an existing node pool, or forgetting that you need to explicitly route the workload to the new nodes.
34All questions36

Practice the full GCP Associate Cloud Engineer Practice Exam 6

50 questions · hints · full answers · grading

Sign up freeTake the exam

More questions from this exam

Q01What is the primary purpose of a Google Cloud project?EasyQ02Your development team needs to manage Compute Engine instances in a specific project. They need t...MediumQ03You are automating the setup of a new Google Cloud project using a bash script. You need to enabl...EasyQ04Your startup has a strict monthly cloud budget of $500. You want to be notified immediately if yo...MediumQ05Your finance team wants to perform granular analysis of your Google Cloud spending using SQL. The...Hard
View all 50 questions →