• Debugging a race condition at 3am โ€” and why it was worth it

    The bug only reproduced under concurrent load above 200 requests/second. Took 6 hours to isolate. The root cause was a non-atomic read-modify-write in a user session store. Lesson: test with actual concurrent load from the beginning. Most race conditions that make it to production were never tested under realistic concurrency.

  • The bug only reproduced under concurrent load above 200 requests/second. Took 6 hours to isolate. The root cause was a non-atomic read-modify-write in a user session store. Lesson: test with actual concurrent load from the beginning. Most race conditions that make it to production were never tested under realistic concurrency.