Latency tiers & cost — seconds, minutes, hours

3 h
requires STH-07

From your world. Daily SLA trade-offs: what a nightly batch costs vs. a near-real-time pipeline, in compute, storage, ops and people

Practice quiz with instant feedback. For a scored run that counts toward your local progress, use uv run de quiz STH-08.

2The seconds tier stores roughly the same retained data as the hours tier. Why does its STORAGE cost still rise (mildly)? AStreaming data is encrypted, and encryption doubles the size BFrequent commits write many small files (and old snapshots). This adds small-file overhead until compaction or expiry CObject stores charge more per GB for fresh data DStorage does not rise at all. The premise is false

4Two engineers model the same workload and tiers independently. What makes their totals land within ±20% of each other? AThey copy the spreadsheet of each other BA deterministic model that formulas drive: fixed inputs (volume, retention, per-unit prices) and explicit formulas for each component. The tier changes only a few multipliers CThey both round everything up to the nearest thousand DThey each guess a round number

5The SLA of a mart is 'fresh by 9 a.m.' The business asks for second-level freshness anyway. What is the right engineering recommendation? AAlways choose the freshest tier. Freshness is free BPrice all three tiers and recommend the cheapest tier that meets the SLA. Seconds gives nothing here and costs several times more CRefuse to model cost and let the business decide without information DMove every other mart to seconds too, for consistency

7A cost model shows the HOURS tier costing more than the SECONDS tier for the same mart. What is the most likely modelling error? AThe model correctly reflects reality BThe tier multipliers are inverted (for example, the model applies always-on compute or on-call to batch and not to streaming) CThe storage price is too low DThe dashboard query is too fast