I fixed an almost identical bug at work a few months ago: an internal service was calling some network device, and would only fail for IPv6-only devices. Turns out it was trying IPv4 first, and then when it failed, it would retry the request with IPv6, but the request body was already consumed so the retry failed with a cryptic error