Section 2: $4.2 Million a Minute - Scaling & Non-Functional Requirements (NFRs)

Executive Summary: This section derives every number the module cites: ~1,200 checkouts/s at peak, $600M of daily GMV, and the ledger fan-out from 10M orders to 30M journal transactions to 120M postings. We turn them into testable SLOs, sort the data into strongly consistent and five-seconds-behind, and find the real scaling problem: not total throughput, but the platform's own accounts, which every capture touches.

Section 1 ended on a question: how much money, and how many writes? Round on purpose, say each step aloud, and make every number change the design.

Orders, Peaks and PSP Calls

You: "I'll assume 60M monthly active buyers, 2M active sellers and 10M orders a day."

orders/day        10M
average           10,000,000 / 86,400 s          ≈ 116 orders/s
peak (10x)        116 x 10 = 1,157               → design for ~1,200 checkouts/s
payment attempts  1.25 per order → 12.5M/day     ≈ 145/s avg, ~1,500 auth calls/s peak
seller sub-orders 1.5 sellers per order          = 15M/day (one capture each)

Interviewer: "Why ten times? Most traffic curves peak at two or three."

You: "Daily curves do. Black Friday and flash sales squeeze a day's buying into minutes, and under-provisioned payments lose revenue and trust. The 1.25 attempts cover declines, retries and 3-D Secure re-tries, each a real PSP call, so the PSP adapter is sized for 1,500/s."