Kubeflow Notebooks Quiz
Baseline: Notebooks 1.11.0 / Community Distribution 26.03.1.
Multiple Choice Questions
What does the Notebook controller reconcile?
- A) A browser process on the user laptop
- B) StatefulSet, Service and configured routing resources from a Notebook CR
- C) An EC2 instance for every user
- D) Only an HTML dashboard
Show Answer
Answer: B) StatefulSet, Service and configured routing resources from a Notebook CR
The StatefulSet controller creates Pods and Kubernetes schedules them. The dashboard is a UI entry point.
What is the precise version baseline here?
- A) All components are Workspaces GA
- B) Notebooks v1.11.0; 26.03.1 calls Workspaces beta while its images are v2.0.0-alpha.3
- C) Notebook and Workspace are identical APIs
- D) v1 has a confirmed end-of-support date in this chapter
Show Answer
Answer: B) Notebooks v1.11.0; 26.03.1 calls Workspaces beta while its images are v2.0.0-alpha.3
Release descriptions and image tags differ. Verify actual API/migration support instead of inferring GA or a v1 retirement date.
Does a Profile automatically isolate every notebook from every other user?
- A) Yes, including AWS and storage
- B) No; Profiles can be shared and network, storage, IAM and app authorization remain separate
- C) Yes, because namespaces block network packets
- D) Yes, because RBAC cancels all unrelated grants
Show Answer
Answer: B) No; Profiles can be shared and network, storage, IAM and app authorization remain separate
The full UI selects a Profile namespace. The Notebook CRD itself does not require a Profile object in every namespace.
What survives replacement of a notebook Pod?
- A) All process memory
- B) Every package installed anywhere in the container
- C) Data on retained persistent volumes; container-layer packages and kernel memory do not
- D) Every attached EC2 instance
Show Answer
Answer: C) Data on retained persistent volumes; container-layer packages and kernel memory do not
Check mount locations, PVC/volume lifecycle and backups. ReadWriteOnce is a single-node access mode, not a single-Pod guarantee.
What are the inspected idle-culling defaults?
- A) Enabled, with a one-minute idle threshold
- B) Disabled; idle threshold 1440 minutes, check period 1 minute
- C) Enabled for every RStudio and shell process
- D) Disabled only for GPU notebooks
Show Answer
Answer: B) Disabled; idle threshold 1440 minutes, check period 1 minute
The culler uses Jupyter kernel activity. Failed/empty API results leave old activity unchanged and can still lead to stopping. Test the actual image and access path.
How does v1.11.0 represent a stopped Notebook?
- A) spec.replicas: 0
- B) Presence of kubeflow-resource-stopped; the controller sets StatefulSet replicas to zero
- C) Annotation value false means running
- D) Deleting its PVC
Show Answer
Answer: B) Presence of kubeflow-resource-stopped; the controller sets StatefulSet replicas to zero
NotebookSpec has no replicas field. Resume by removing the annotation. Even a false string still counts as present.
What does a custom image digest guarantee?
- A) All users have identical complete runtime environments
- B) The referenced image content; mounted data and runtime changes can still differ
- C) Automatic compatibility with all GPU drivers
- D) That UI image restrictions cannot be bypassed through the API
Show Answer
Answer: B) The referenced image content; mounted data and runtime changes can still differ
Use tested server-prefix/port/UID behavior, dependencies and architecture. A mutable tag alone does not pin image bytes.
Short Answer Questions
- Why does stopping an idle GPU notebook not guarantee immediate cost savings?
Show Answer
Pod requests can be released, but other workloads, PDBs, NodePool limits/disruption policies and capacity management affect node termination. EC2 charges can continue while the node remains running.
- How do RBAC, Istio authorization and NetworkPolicy differ for notebooks?
Show Answer
RBAC governs Kubernetes API actions. Istio authorization controls requests handled by configured proxies and policies. NetworkPolicy governs allowed Pod network traffic when enforced by the CNI. None alone guarantees storage/IAM/application isolation.