Compute resource management in MLOps - Practice Problems & Coding Challenges

Challenge - 5 Problems

💻 Command Output

intermediate

Output of Kubernetes resource request command

What is the output of the following command when run on a pod named ml-training-pod in Kubernetes?

kubectl get pod ml-training-pod -o jsonpath='{.spec.containers[0].resources.requests}'

A {"cpu":"500m","memory":"1Gi"}

B {"cpu":"1","memory":"512Mi"}

C Error: resource requests not found

D {"cpu":"100m","memory":"256Mi"}

💡 Hint

Resource requests define the minimum compute resources a container needs.

The command extracts the resource requests for the first container in the pod. Option A shows a typical request of 500 millicores CPU and 1 GiB memory.

🧠 Conceptual

intermediate

Understanding GPU resource allocation in MLOps

In an MLOps pipeline, why is it important to specify GPU resource limits for training jobs?

A To allow the training job to run without any CPU resource limits

B To prevent a training job from using more GPU memory than allocated, avoiding interference with other jobs

C To increase the training speed by automatically adding more GPUs when needed

D To disable GPU usage and force training on CPU only

💡 Hint

Think about resource sharing in a multi-tenant environment.

Specifying GPU limits ensures that a training job does not consume more GPU resources than allocated, which helps maintain fairness and stability when multiple jobs run simultaneously.

🔀 Workflow

advanced

Order the steps to configure autoscaling for compute resources in a Kubernetes cluster

Arrange the following steps in the correct order to enable autoscaling of pods based on CPU usage in Kubernetes.

A 1,3,2,4

B 3,1,2,4

C 1,2,3,4

D 3,2,1,4

💡 Hint

Metrics collection must be ready before autoscaler can use metrics.

First, metrics server must be deployed to provide CPU usage data. Then resource requests must be set so autoscaler knows what to measure. Next, create the HPA to define scaling rules. Finally, verify the scaling behavior.

❓ Troubleshoot

advanced

Identify the cause of pod scheduling failure due to resource constraints

A pod in Kubernetes fails to schedule with the message: 0/5 nodes are available: 5 Insufficient cpu. What is the most likely cause?

A The pod's node selector does not match any node labels.

B The pod's container image is too large to download on nodes.

C The pod lacks a resource limit for memory.

D The pod requests more CPU than any node currently has available.

💡 Hint

Focus on the error message about CPU availability.

The error indicates no node has enough free CPU to meet the pod's CPU request, causing scheduling failure.

✅ Best Practice

expert

Best practice for managing compute resources in multi-tenant MLOps environments

Which practice best ensures fair and efficient compute resource usage among multiple teams running ML workloads on shared infrastructure?

A Disable resource requests and limits to maximize scheduling flexibility.

B Allow all teams to request unlimited resources and rely on manual monitoring.

C Implement resource quotas and limit ranges per namespace to control resource consumption.

D Use a single shared namespace without resource restrictions for simplicity.

💡 Hint

Think about automated controls to prevent resource hogging.

Resource quotas and limit ranges enforce boundaries on resource usage per team, preventing any single team from exhausting cluster resources and ensuring fair sharing.