Sporadic `405 Not Allowed` errors

[zstadler]

A lightly-loaded overpass container occasionally returns 405 Not Allowed errors for relatively simple POST queries.

According to Gemini, this is most likely a "Keep-Alive" race condition:

This is a notorious issue with Nginx (which Overpass uses) and Docker load balancers.

• The Mechanism: The client (or an intermediate proxy) attempts to reuse an idle TCP connection (Keep-Alive) to send a new POST request.
• The Failure: Simultaneously, the Overpass server decides the connection has been idle too long and closes it (sends a TCP FIN or RST).
• The Result: The Load Balancer/Proxy detects the closed connection while trying to forward the data. For GET requests, it would typically retry or return a 502/504. However, because POST is not idempotent (retrying it could duplicate data/action), Nginx defaults to returning a 405 Method Not Allowed or sometimes a generic 400-level error instead of retrying.
• Fix: Adjust the keepalive_timeout in your Nginx config to be slightly higher than the client's/load balancer's timeout, or disable keep-alive for the upstream CGI.