Troubleshooting Intermittent 502 Errors in a Node Backend Behind Nginx Reverse Proxy

Background: The department migrated Java web interfaces to Node, naming the service "HOTDOG". A Node service was set up as a bridge between clients and the web interface, with Nginx as reverse proxy and load balancing.

Problem: After deployment, requests to the new Node API intermittently returned 502 errors. Direct access to the Node server (IP+PORT) worked, indicating the issue lay in the proxy layer.

Investigation steps:

Checked Nginx logs – suggested half‑open TCP connections causing premature closure.

Examined Node logs – no 502 entries, low system load, confirming Nginx was not receiving a response.

Tried long‑lived upstream connections – no improvement.

Verified Node’s default HTTP timeout (2 min) was not the cause.

Captured packets with tcpdump – observed Nginx sending a request, the Node process ACK‑ing then sending FIN, resulting in Nginx returning 502.

Reproduced the issue with a minimal Node server handling a request containing non‑ASCII characters, which triggered the clientError event and caused the socket to hang.

Solution:

Add a clientError handler in the Node server to log the malformed request and respond with HTTP/1.1 400 Bad Request , preventing the socket from hanging.

Configure Nginx (or OpenResty) to treat each request independently and avoid fail‑timeout cascading failures.

Conclusion: The 502 errors were caused by Node rejecting requests with illegal characters at the TCP level, which Nginx interpreted as a server failure. Proper error handling in Node and refined Nginx upstream settings resolve the issue and improve overall reliability.