March 12, 2026
Your balance isn't 'eventually' money
"Design Me a Highly Resilient Database"
No context, “wrong” answer: commenters torch broken interviews
TLDR: An engineer failed a job interview for choosing a bank-safe design over the buzzword the interviewer wanted. Commenters slammed the context-free question, backed strict accuracy for money, and challenged the “30,000 servers” claim—calling it proof that hiring is broken and priorities are upside down.
An engineer says he flunked an interview for answering “Design a highly resilient database” with a bank-grade plan, not the buzzword the interviewer was fishing for. He picked Postgres (link) for strict, provable accuracy; the interviewer wanted Cassandra (link). The comments went nuclear. One crowd blasted the process as broken interviewing theater, cheering his pushback and calling the question adversarial. Another thread dove into why money needs hard guarantees: eventual consistency (systems “agree later”) is fine for cat pics, not your paycheck. Think ACID as “every change is correct, isolated, and truly saved.” Meanwhile, folks linked the classic tradeoff explained by the CAP theorem.
The spiciest subplot? The claim about Apple Pay running on 30,000 servers. Skeptics swarmed with “30k for millions of transactions?” energy, while others shrugged: scale is messy and context matters. Supporters praised the author for writing a calm, educational piece instead of rage-posting—“they didn’t deserve you” vibes all around. Meme weekend arrived early: jokes about “your bank balance will eventually be correct,” comparisons to a surgeon asked to build a random organ, and the eternal refrain: you’re better off dodging an employer who wants magic words over real-world tradeoffs.
Key Points
- •The author argues that designing a “highly resilient database” requires product context, as resilience depends on specific data, queries, and failure modes.
- •They propose a production-tested PostgreSQL architecture: CloudNativePG on Kubernetes with S3-compatible storage for WAL archiving, offering automated failover, PITR, and restoration from backups.
- •Cassandra is recognized as highly available with tunable consistency, suitable for large-scale, write-heavy workloads like IoT telemetry, but not ideal for financial ledgers.
- •For regulated finance, ACID compliance and auditability (e.g., PCI DSS) are mandatory; eventual consistency is unacceptable for systems of record.
- •The article emphasizes CAP theorem tradeoffs: under partition, systems must choose consistency or availability, and database choices should reflect use-case priorities.