Configuration:

worker_processes 2;

upstream app-backend {
server 10.23.1.1:8080;
server 10.23.1.2:8080;
}

nginx version: nginx/0.6.39
built by gcc 4.2.4 (Ubuntu 4.2.4-1ubuntu4)
configure arguments: --prefix=/opt/nginx --group=nogroup --user=nobody
--with-http_gzip_static_module --with-http_ssl_module

I noticed tonight that for a particular run of large requests, the same
upstream IP was being used consecutively instead of round-robin. There were
no entries in the error log for the time-frame, and both upstream servers
were available.

Example:

10.23.1.3 [17/Jan/2010:01:11:39 -0800] http "POST /request HTTP/1.1" 89515
200 10.23.1.1:8080
10.23.1.3 [17/Jan/2010:01:11:40 -0800] http "POST /request HTTP/1.1" 82520
200 10.23.1.1:8080
10.23.1.3 [17/Jan/2010:01:11:41 -0800] http "POST /request HTTP/1.1" 88346
200 10.23.1.1:8080
10.23.1.3 [17/Jan/2010:01:11:42 -0800] http "POST /request HTTP/1.1" 97289
200 10.23.1.2:8080
10.23.1.3 [17/Jan/2010:01:11:44 -0800] http "POST /request HTTP/1.1" 92068
200 10.23.1.2:8080
10.23.1.3 [17/Jan/2010:01:11:45 -0800] http "POST /request HTTP/1.1" 96217
200 10.23.1.1:8080
10.23.1.3 [17/Jan/2010:01:11:45 -0800] http "POST /request HTTP/1.1" 86415
200 10.23.1.1:8080
10.23.1.3 [17/Jan/2010:01:11:47 -0800] http "POST /request HTTP/1.1" 86883
200 10.23.1.2:8080
10.23.1.3 [17/Jan/2010:01:11:48 -0800] http "POST /request HTTP/1.1" 91168
200 10.23.1.2:8080
10.23.1.3 [17/Jan/2010:01:11:50 -0800] http "POST /request HTTP/1.1" 94113
200 10.23.1.2:8080
10.23.1.3 [17/Jan/2010:01:11:52 -0800] http "POST /request HTTP/1.1" 95881
200 10.23.1.2:8080

Most of the responses took ~10-30 seconds to complete. Is there anything
obvious that would cause this behavior? Other than this sequence of
requests, the round-robin behavior logs as expected.

--
Adam
zellster@gmail.com
_______________________________________________
nginx mailing list
nginx@nginx.org
http://nginx.org/mailman/listinfo/nginx