Hi,

with one of my NginX servers I have some timeouts during TCP connection,
I don't know where can be the problem :
It's a dual Xeon E5405 (so, 8 cores) and load average is between 2 and
5. Near all CPU are under 20% usage, except one which is between 60-80%
: this one is doing hardware and software interrupts (I use irqbalance).

I use NginX 0.6.32, under Linux (Debian Lenny Amd64).
There is only one ethernet interface (intel gigabit), and there is about
200Mbps of outgoing trafic.
Netfilter connection tracking is disabled (module is not loaded).

NginX use 3 worker processes, and allow 16384 connections per worker ;
keepalive_timeout is high (65 seconds). Worker priority is set to "-1".

I found no message in NginX error logs, neither in syslog.

So to try to identify the problem I make a small script which
interrogate NginX status every second ; with a 2 seconds timeout on
socket open.
Example of result is :
22:48:17 Active connections: 6139 server accepts handled requests
15950593 15950593 208602816 Reading: 21 Writing: 46 Waiting: 6072 0-
22:48:18 Active connections: 6164 server accepts handled requests
15950684 15950684 208604084 Reading: 15 Writing: 32 Waiting: 6117 0-
22:48:19 Active connections: 6151 server accepts handled requests
15950778 15950778 208605333 Reading: 19 Writing: 42 Waiting: 6090 0-
22:48:20 Active connections: 6172 server accepts handled requests
15950882 15950882 208606576 Reading: 18 Writing: 48 Waiting: 6106 0-


But some times there is :
22:51:52 Active connections: 5914 server accepts handled requests
15969112 15969112 208870244 Reading: 27 Writing: 35 Waiting: 5852 0-
22:51:53
Warning: fsockopen(): unable to connect to 127.0.0.1:80 (Connection
timed out) in /home/daevel/checkHttp on line 8
110-Connection timed out
22:51:56 Active connections: 5857 server accepts handled requests
15969368 15969368 208874875 Reading: 22 Writing: 35 Waiting: 5800 0-


Does someone know where can be the problem ? I repeat, there is nothing
in nginx, system or kernel logs.

Thanks,
Olivier