cpu spike on HUP

[Igor Sysoev]

On Thu, Oct 28, 2010 at 04:55:40PM -0700, Gary Barrueto wrote:

> Whenever nginx is sent a HUP signal, the cpu jumps to 100% then slowly 
> drops until all the nginx worker processes that are trying to shutting 
> down are done. Those nginx process usually take a while to shutdown 
> since we're serving very large files (200MB-2GB). Any ideas why this is 
> happening?
> 
>  From the strace below it looks like it all in sendfile but I wouldn't 
> think that would so cpu intense.

What kind of CPU is used in these moments: system or user according to top ?

> -gary
> 
> /usr/local/nginx/sbin/nginx -V
> nginx version: nginx/0.8.52
> built by gcc 4.4.3 (Ubuntu 4.4.3-4ubuntu5)
> TLS SNI support enabled
> configure arguments: --with-http_ssl_module --with-http_flv_module
> 
> This is from about 10 seconds of strace of one of the nginx process 
> while shutting down.
> 
> sudo strace -cp 7087
> Process 7087 attached - interrupt to quit
> ^CProcess 7087 detached
> % time     seconds  usecs/call     calls    errors syscall
> ------ ----------- ----------- --------- --------- ----------------
>   99.85    0.071681         139       515           sendfile
>    0.15    0.000110           0       252           epoll_wait
> ------ ----------- ----------- --------- --------- ----------------
> 100.00    0.071791                   767           total


-- 
Igor Sysoev
http://sysoev.ru/en/