It was definitely MinIO related, I probably should have made that clearer. We noticed that with zero fault tolerance, MinIO objects would randomly become corrupted, which MinIO would present as "you're making too many requests, please slow down". We were certainly not making too many requests.