Stripe Backend Engineer Interview Questions

I would decide which data must be strongly consistent and which can tolerate eventual consistency. Money movement and ledger updates need stricter guarantees, while analytics or dashboards may be eventually consistent. The key is to isolate the critical path and make compensating processes available for noncritical workflows.

I would combine request deduplication, transactional safeguards, and unique constraints at the database layer. If two requests race, only one should be able to create the charge record, and the other should resolve to the same result or fail safely. This is one of those problems where application logic alone is not enough.

Retries are necessary because network failures are inevitable, but retries must be safe. I would make every retried operation idempotent or attach enough context that the system can distinguish a retry from a new intent. Without that, retries can silently turn into duplicate financial actions.

I would keep the ledger append-only, use immutable records, and avoid updating historical entries in place. Scaling comes from partitioning, replication, and read models, while auditability comes from preserving a complete event trail. If corrections are needed, I would record compensating entries rather than rewriting history.

I would compare the source ledger entries, the aggregation logic, and the timing of writes versus reads. Mismatches often come from missing events, duplicate processing, or stale read replicas rather than arithmetic errors. I would also verify whether refunds, reversals, or pending states are being counted correctly.