Why scaling requires different strategies in MLOps - Challenge Your Understanding

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why horizontal scaling differs from vertical scaling

Which statement best explains why horizontal scaling requires different strategies than vertical scaling in machine learning operations?

A Horizontal scaling adds more machines, so it needs coordination between them, while vertical scaling upgrades a single machine's resources.

B Vertical scaling is cheaper and faster than horizontal scaling, so it always uses the same strategy.

C Horizontal scaling only works with cloud services, while vertical scaling is for on-premises only.

D Vertical scaling duplicates data across machines, while horizontal scaling increases CPU speed.

💡 Hint

Think about how adding more machines versus upgrading one machine affects system design.

Horizontal scaling means adding more machines that must work together, requiring strategies for coordination and data sharing. Vertical scaling means making one machine stronger, which doesn't need coordination.

💻 Command Output

intermediate

Output of Kubernetes pod scaling command

What is the output of this command when scaling a deployment named 'ml-model' to 3 replicas?

MLOps

kubectl scale deployment ml-model --replicas=3

A deployment.apps/ml-model scaled

B Error from server (NotFound): deployments.apps "ml-model" not found

C deployment.apps/ml-model created

D No resources found in default namespace.

💡 Hint

Scaling a deployment changes the number of pods, not creating or deleting the deployment.

The command scales the existing deployment to the specified number of replicas and confirms with 'deployment.apps/ml-model scaled'.

🔀 Workflow

advanced

Order the steps to implement auto-scaling for a machine learning service

Put these steps in the correct order to set up auto-scaling for a machine learning model deployment.

A 3,2,1,4

B 3,1,2,4

C 1,3,2,4

D 1,2,3,4

💡 Hint

Think about deploying first, then monitoring and configuring scaling.

You first deploy the service, then define what to monitor, configure scaling rules, and finally test the setup.

❓ Troubleshoot

advanced

Troubleshooting failed scaling in Kubernetes

You tried to scale a deployment but the pods remain at the old count. Which issue is most likely causing this?

A The deployment YAML file is missing the container image.

B The pods are stuck in CrashLoopBackOff due to code errors.

C The Kubernetes cluster is offline.

D The deployment has a resource quota limiting pod count.

💡 Hint

Think about cluster limits that prevent scaling.

A resource quota can limit how many pods can run, preventing scaling even if the command succeeds.

✅ Best Practice

expert

Best practice for scaling stateful machine learning services

Which approach is best when scaling a stateful machine learning service that stores user session data?

A Store session data only in pod memory to improve speed.

B Scale pods freely without session management, relying on pod restarts.

C Use sticky sessions with a load balancer to keep users on the same pod.

D Disable scaling to avoid data loss.

💡 Hint

Think about how to keep user data consistent when scaling.

Sticky sessions ensure users connect to the same pod, preserving session data and avoiding data loss during scaling.