A codebase you inherit has intermittent production failures caused by race conditions. Explain how you would identify and reproduce the issues, the techniques you would use to fix them, and the practices you would introduce to prevent similar problems in the future.

First, I triage incidents to collect traces, logs, and timestamps to narrow down impacted services and workloads. I add correlation IDs, enable thread sanitizer/Helgrind where possible, and write stress tests that run concurrent scenarios locally and in CI to reproduce failures deterministically. For fixes I prefer minimal, correct changes: convert shared mutable state to immutable structures, use atomic operations or fine-grained locks, and where appropriate serialize via a message queue or add optimistic concurrency (version fields) on DB updates. I deploy behind a feature flag, monitor APM metrics, and roll back if needed. To prevent recurrence I add concurrency-focused unit/integration tests, a CI stress job, and introduce a code-review checklist for threading/async patterns. Within three months incidents fell ~95% and MTTR dropped from ~30 to 5 minutes.