Challenge 4: A sharded relational ledger
The user ledger takes 30,000 requests a second, 30% of them writes: 9,000 writes a second. It must stay in a relational database for its transactions, and a single leader takes 2,500 requests a second. Survive the loss of any instance with p99 under 250 ms, for at most 8,000 USD a month.
Traffic demand30,000 RPS70% reads, 30% writes
p99 latency target≤ 250 msSlowest 1% of requests
Availability≥ 99.99%N-1 failure survival (0 drops)
Monthly cloud budget≤ 8,000 USDA month, everything included