Hi Maxim,
Thank You for your answer. It really makes sense, however, in the mentioned case, I could see an elevation of the number of occurrences just after the migration from nginx 1.6 to nginx 1.8.
This behaviour affects < 0.2 % of our requests, but it means hundreds of requests per hour.
Also, I could see this implementation isn't new, therefore I believe this behaviour maybe another change caused it to start happening.
https://github.com/nginx/nginx/commit/64a9f700929dbc8f0730be4f91cc3bbfde8fc3e6
Regarding your comment about the race condition, most times the object the caused the message appeared only once on the access log, so I don't think it was caused by a concurrency.
Best Regards,
Biazus