I do run a web application, which I stripped to the max to avoid a high tcpsndbuf values. Still a failcnt on tcpsndbuf will be reached after a few days and very, very little traffic.
I initially thought it is an application container issue as described in this thread.
The tcpsndbuf values rise and rise. It seems as the buffer value is never released completely. Thus, it continuously increases until it breaches the limit and demands it's tribute of 100% CPU throttle.
However, after ruling out different sources like hibernate configs, mysql driver error, I want to focus on openvz managed by Parallel Plesk/Power Panel and the apache proxy_mod.
The reason why I belive it, is the fact that numeruous process of
/usr/sbin/apache2 -k start are listed in the process list (Parallels Power Panel, Processes)
I use the proxy_mod to route port 80 to my application server which is on a different port but the same host.
How can I analyse the socket states in detail? E.g. what buffer is locked by which application, when has it been assigned? Can I see the proxy_mod connections? Is that type of data available? Any other hints on tcpsndbuf provoking a max CPU load are highly welcome.