Rightsizing
Rightsizing means moving a resource that is genuinely running to a smaller size or a lower tier, because its measured peak demand leaves a lot of capacity paid for and unused. Unlike turning something off, the resource keeps doing its job — it just stops carrying headroom it never touches.
What we look for
Section titled “What we look for”A rightsizing candidate has to clear a peak-usage bar, not an average one — an average hides the spike that actually sizes the resource:
- Sustained p95 utilization (CPU, memory, DTU, IOPS, throughput, RU/s) sits comfortably below the next size down, across a multi-week window.
- Enough coverage and headroom that known spike windows (month-end jobs, batch loads, cache warm-ups) still fit after the resize.
- The target size is a vetted, concrete SKU — the candidate carries the exact step-down, not a vague “make it smaller”.
The estimated saving is capped at the resource’s actual bill, so a rightsizing claim never exceeds what you currently pay for it.
How to act
Section titled “How to act”Rightsizing is usually an online or short-maintenance-window change and is reversible — you can resize back up by re-running the same command with the original SKU. leancosts drafts the pre-check (the metrics to confirm), the exact resize command, what to watch afterwards, and the rollback. Follow Act on a cost finding.