Rightsize This ClusterSpark · hard · ~14 minThe 'streaming-adjacent' batch cluster runs 24/7: 24 nodes at $0.60/node/hour. A cost review pulls the utilization graphs: CPU averages 19%, memory 24%, and the workload is bursty — a nightly 4-hour heavy window, near-idle otherwise. The owning team's position: 'we sized it for the nightly peak, and downsizing risks the SLA'.Both things can't stay true at these numbers. Rightsize with arithmetic, respecting the peak.Evidence

Utilization

Numbers

Checkpoints

What do the utilization graphs actually say?

Scheduled shape: 24 nodes for 4h + 4 nodes for 20h, at $0.60/node/hr, 30 days. Monthly cost in DOLLARS? ((24x4 + 4x20) x 0.6 x 30)

How do you answer the 'SLA risk' fear honestly?

Wrap it for the cost review: savings, the residual risks you'd monitor, and when spot nodes enter the conversation. Interview-style.