Part 7 of 8 · 2 min read
Trade-offs
What to say out loud.
Trade-offs and what to say
| Choice | Gain | Cost |
|---|---|---|
| Two-phase commit | Database enforces atomicity | Locks, slow, coordinator risk |
| TC/C | Parallel, database independent | Application must write cancel logic |
| Saga | Fits microservices | Ordered steps, compensations to write |
| Event sourcing | Audit and replay | Learning curve, more storage |
| Local files | Very fast | State on each node, needs replication |
| Raft group | Safe log | Limited by the leader, more nodes |
| Sharding | Scale | Cross-group transfers need a coordinator |
| Snapshots | Fast rebuild | Storage and a schedule |
Interview angle
Answer in your head first, then open it.
Why not Redis alone?
A transfer needs two writes, and a crash between them loses money.
How do you make a cross-shard transfer atomic?
A saga or TC/C with compensations, and a phase status table for recovery.
Why event sourcing?
Reproducibility: any balance can be rebuilt and checked.
How do you keep the event log safe?
Replicate it with Raft over 3 or 5 nodes.
How do you reach 1 million a second?
Append-only local files, snapshots and many Raft groups.
One follow-up to try. "Add currency conversion." A good reply: a transfer may span two currencies, so an exchange step is added, with a quoted rate that is stored in the event so replays give the same result. Amounts are kept in the smallest unit of each currency, as integers. The audit trail records the rate source and time.
Check yourself: 1. Why store the exchange rate inside the event? 2. Which data is rebuilt, and which is protected? 3. What does the coordinator record while a saga runs?
Words you will meet
| Word | Meaning |
|---|---|
| 2PC | Two-phase commit: prepare, then commit or abort |
| TC/C | Try, confirm or cancel, as application-level steps |
| Saga | A chain of local transactions with compensations |
| Command | A request that may be refused |
| Event | A validated fact, never changed |
| State machine | Code that validates commands and applies events |
| Snapshot | A saved copy of the state at one moment |
| Raft | A consensus algorithm that replicates a log |