Quiz 2 of 5: Control plane, scheduling and failures
4
A team runs a 4-member etcd cluster. What is the best conclusion?
Why: Quorum is floor(n/2)+1, which is 3 for four members, so losing two breaks it. A fourth member adds cost and replication latency without adding fault tolerance. Use 3 or 5.
5
On a self-managed cluster every kubectl write fails but reads work, and etcd reports a NOSPACE alarm. What is the correct recovery?
Why: NOSPACE means the database hit its quota, so etcd rejects writes. Compaction frees space logically, defragmentation returns it to disk, and the alarm stays raised until explicitly disarmed. Defragment members one at a time to avoid losing quorum.
6
A Pod is Pending with the event '0/5 nodes available: Insufficient cpu'. What does this mean?
Why: Scheduling is driven by requests, not limits or live usage. This is also the signal that cluster autoscalers, Karpenter and GKE Autopilot use to add nodes.