Roberto De Ioris wrote:

> Hi Juliusz, check your nginx log, it could be a some form of 'attack',
> where you receive the request but soon the client disconnects.
>
> If you get tons of this error the performance will be hurt for sure as you
> lose precious resources.
>
> If its normal for your clients to have a big rate of prematurely closed
> connections (but this looks strange to me) use this:
>
> --logto /dev/null
>
> to disable all the I/O on your disks

Thank you for a prompt answer.

I tried everything and after some debugging I discovered that all uWSGI 
processes were using 100% CPU. The reason for this was using 'file' 
cache in Beaker. I guess this kind of cache doesn't handle more traffic 
too well (it seems weird though, HDD should be able to cope with sth 
like 1 request per second).

After changing the cache to 'memory' the situation got back to normal - 
less than 1% CPU and no broken pipe error in the log. I guess I will 
have to install memcached because AFAIR 'memory' cache is per process.

BTW Kudos to uWSGI developers. In desperate action of fixing the 
situation I tried using Paster instead and when it hit 100% CPU the 
website was barely usable. uWSGI with 100% CPU usage was slow but served 
all the requests nonetheless.

-- 
Juliusz Gonera
http://juliuszgonera.com/
_______________________________________________
uWSGI mailing list
[email protected]
http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi

Reply via email to