Hi,

I am trying to tune our nginx frontends (proxying a pack of mongrels) to avoid this type of message

2008/04/30 11:06:50 [warn] 35585#0: *40967931 an upstream response is buffered to a temporary file /opt/local/var/run/nginx/proxy_temp/ 0/58/0000370580 while reading upstream, client: 202.129.81.166, server: www.redbubble.com, request: "GET /people/basiamc/art/732916-4- fire-fairy? HTTP/1.0", upstream: "http://172.16.0.76:8006/people/basiamc/art/732916-4-fire-fairy ", host: "www.redbubble.com", referrer: "http://www.redbubble.com/people/basiamc/art/732916-4-fire-fairy "

A little bit of curling reveals that our minimum page size is between 11k and 14k, but can be up to 35k for the common pages. So I have set

proxy_buffer_size 32k

In the expectation that this buffer will be allocated for all proxy requests before the overflow into proxy_buffers.

For pages that have a large number of comments the sizes can range up to 150k but even with a setting like this

proxy_buffers 128 4k;

which should yield a maximum buffer of 546k, this request

[dave@crimson ~]$ curl -I http://www.redbubble.com/people/basiamc/art/732916-4-fire-fairy HTTP/1.1 200 OK Server: nginx Date: Wed, 30 Apr 2008 01:13:45 GMT Content-Type: text/html; charset=utf-8 Connection: keep-alive Set-Cookie: _session_id=1263f18e47b978d84cab76592cf240c4; path=/ Cache-Control: private, max-age=0, must-revalidate Content-Length: 282541

Generates a warning.

Should I continue to increase the buffer size? Should I allocate the buffer in larger sizes? Is there an interaction with proxy_buffers and gzip or the output buffer settings ?