Hands-on design challenge

Challenge 4: A sharded relational ledger

The situation

The user ledger records every balance and every transfer, in one relational database with a standby. It has grown until the writes alone are more than its leader can take, so the leader is full and requests are dropped.

Traffic

30,000 requests a second.

OperationShareSteps
Read70%Read data
Write30%Write data
  • Read70%Read data
  • Write30%Write data

Data

A cache answers 85% of the reads that reach it.

What you are proving

Writes can outgrow any one machine: the ledger takes more writes a second than a single leader can, and keeps its transactions.

Rules

  • The ledger stays in a relational database, and every write is stored in it.Moving money needs transactions: a transfer is stored whole or not at all.
  • A CDN cannot answer these requests.Each balance belongs to one user, so no two users get the same response.

Builds on

Targets

  • Serve all 30,000 requests a second
  • At least 99.99% available, with no single point of failure
  • At most 8,000 USD a month
  • Every write kept in a relational database
  • Any tier loses one instance: nothing dropped, no tier over 80% busy

Hints