Topic drills

Pick a topic and run the same fixed format every time: name it → explain out loud (mic) → code it → edge cases → say the trade-off (mic) → follow-up → catch the bug. Never start from a blank slate.

• Idempotency
  A retried request must not double-apply.
• State machine (status lifecycle)
  A status change is a guarded business event, not a field write.
• Optimistic locking
  Concurrent edits: detect the conflict instead of clobbering the winner.
• Pessimistic / row-level locking
  Serialize a read-modify-write across instances.
• Distributed lock
  One scheduled job runs exactly once across many servers.
• Event processing (dedup + order)
  Per-entity final state from an at-least-once, reordered stream.
• Reconciliation
  Compare two sources that should agree; flag or repair the drift.
• LRU / TTL cache
  A bounded cache that evicts least-recently-used and expires entries.
• Queue / worker / retry / DLQ
  Async work that survives failure instead of losing it.
• System design
  Lead with the default architecture, then survive the deep-dive probes.