Large uploads through nginx into a Node handler fail in predictable ways when buffering defaults stay on. For multi-GB bodies I set client_max_body_size high enough on that location only, turn proxy_request_buffering off so nginx streams to upstream instead of spooling disk, and raise proxy_read_timeout / proxy_send_timeout well above the slowest legitimate upload. Leaving buffering on looks like random 413/504 under load even when the Node process is healthy.
Ops check after enabling: upload a multi-hundred-MB file through the public path and watch whether nginx workers spike disk I/O (still buffering) versus steady upstream sockets. Keep a separate stricter limit on non-upload API locations so one raised body size does not become an abuse path. If you terminate TLS at the edge, confirm the Node app sees Content-Length or chunked framing unchanged — middleware that re-buffers the whole body in memory will OOM long before nginx complains.