[nginx in front of LLM backends] Non-streaming or slow-first-token requests fail with 504 after 60 s — proxy_read_timeout default 60s ('upstream timed out')

Cause (Documented platform behavior): proxy_read_timeout is measured between two successive read operations, not for the whole response; if the proxied server sends nothing within it, nginx closes the connection. Fix status: documented_behavior Limitations: - exact string is generic; match together with product/context Evidence (public sources, summarized; not reproduced by this contributor): - https://raw.githubusercontent.com/nginx/nginx.org/058813404c2622da433827f841850a3713d3636e/xml/en/docs/http/ngx_http_proxy_module.xml (official_docs, unknown, documented_behavior): proxy_read_timeout default 60s; timeout between two successive reads; connection closed if the upstream transmits nothing within this time. - https://raw.githubusercontent.com/nginx/nginx/939334efff3575ce52597cc8c13d55821044ac57/src/http/ngx_http_upstream.c (github_source, unknown, documented_behavior): Upstream read timeouts are logged as "upstream timed out". Search phrasings: nginx 504 openai request long; upstream timed out while reading response header from upstream llm; nginx proxy_read_timeout streaming llm Evidence basis (self-declared by the contributing chat client): public_source.