Observability · Capacity design
Traffic grows 15% a month. When do you run out?
A service's peak traffic grows 15% a month. When does it run out of capacity, and what do you provision and alert on?
Capacity design · search-apiIllustrative scenario, not a real incident
Finance asks for next year's instance budget for the search API, and nobody has a forecast
The search API runs on reserved hardware. An order for more takes two months from request to instances in service, so the autoscaler can move load around the fleet and cannot make the fleet larger.
Peak traffic has grown about 15% a month for a year, each month on top of the last. The team load-tested one instance last week, and today's daily peak uses half of the fleet.
What the dashboards and logs show
- Daily peak today
- 3,000 requests a second
- One instance
- 250 requests a second, measured with p99 inside its target
- Fleet
- 24 instances
- Growth
- 15% a month, compounding
- Lead time for an order
- 2 months
Estimate · scored under "Numbers"
Do the arithmetic first
Work each figure out from what is given, on paper or in your head. Round freely: an estimate counts when it is close, and the working is shown once you check.
- Daily peak today
- 3,000 requests a second
- One instance handles
- 250 requests a second
- Instances in service
- 24
- Growth
- 15% a month, compounding
- Target utilisation at the peak
- at most 70% of capacity
- Lead time for more capacity
- 2 months
- months
- months
- instances
- %
Your rubric score, the model answer, the follow-up questions and the related lessons appear once you have answered the last step.