Hi,
Per the below...thats how I have things set up for nginx. Change to
worker_priority -1; in events for at the same level as
worker_rlimit_nofile?
worker_rlimit_nofile 200000;
events {
worker_connections 30000;
multi_accept on;
use epoll;
}
Yes..I have three machines, 4 cores each. The fourth core is used for
another process that gets gets less load. In total, I have 9 cores
across three machines doing 1K qps.
I am looking at htop. 1 core is at around 70% The other three are
around 15%.
On Sat, Oct 20, 2012 at 1:25 AM, Andriy Kornatskyy
<[email protected]> wrote:
>
> May be that is misprint, but you said you have 9 cores and configuration says
> you use 3 process only. It is okay to allow uWSGI use more processes, OS
> scheduler get it managed, especially if your app interacts with database.
>
> With nginx upfront I would recommend add `worker_priority -1;`, so CPU
> intensive things like SSL handshake got priority over uwsgi.
>
> Also I would use unix file socket instead of TCP (since you host everything
> on the same machine).
>
> Thanks.
>
> Andriy
>
>
> ----------------------------------------
>> Date: Sat, 20 Oct 2012 01:13:09 +0800
>> From: [email protected]
>> To: [email protected]
>> Subject: Re: [uWSGI] uwsgi and High load across the servers.
>>
>> Thanks.
>>
>> I need the other core for another process. I just looked at
>> optimize...set the python optimization level (dangerous)
>>
>> Why dangerous? What does that mean? Yeah...I will disable logging in
>> a bit. I just launched in production and want to see what is going on.
>> I have nginx logs off except error.
>>
>> On Sat, Oct 20, 2012 at 1:04 AM, Andriy Kornatskyy
>> <[email protected]> wrote:
>> >
>> > David,
>> >
>> > You can get better throughput by settings `process=<count>` per CPU cores
>> > you have. I usually set `optimize = 2` and `disable-logging = True`.
>> >
>> > Thanks.
>> >
>> > Andriy Kornatskyy
>> >
>> >
>> > ----------------------------------------
>> >> Date: Sat, 20 Oct 2012 00:50:31 +0800
>> >> From: [email protected]
>> >> To: [email protected]
>> >> Subject: [uWSGI] uwsgi and High load across the servers.
>> >>
>> >> Hi,
>> >>
>> >> Below is launch I launch uWSGI. Right now, I have 9 cores doing 1K
>> >> qps in total..each with 4 gigs of ram. For each requests has two
>> >> redis reads and two redis writes and I am using gevent for async
>> >> calls. Agv response time is under 20ms from a post request. In the
>> >> next few days I will get slammed with 3K-5K qps. I have 100 ms to
>> >> respond to a requests hence the --harakiri 1. Would be great if
>> >> harakiri can be in milli seconds in the future. As of now I am stable
>> >> will the requests but feel I dont know enough on how to truly optimize
>> >> what I have.
>> >>
>> >> Its great I am doing 1K qps but i feel someone can do better by
>> >> tweaking. I started with two servers but had to add a third. Am I
>> >> doing good? Bad? Average? I will be adding more servers but want to
>> >> squeeze out the most I cant before a add a new server.
>> >>
>> >>
>> >> /usr/local/bin/uwsgi --loop gevent --carbon
>> >> <%=@carbon_ip_address%>:2003 --harakiri 1 --socket 127.0.0.1:8060
>> >> --processes 3 --pp
>> >> /home/ubuntu/workspace/rtbopsConfig/rtbServers/rtbUwsgiBidderServer/
>> >> --wsgi-file
>> >> /home/ubuntu/workspace/rtbopsConfig/rtbServers/rtbUwsgiBidderServer/uwsgiBidderServer.py
>> >> -b 62768 --master --async 1200 --enable-threads --listen 2048
>> >> --reload-on-rss 800 --memory-report --logto2 /tmp/uwsgiBidder_log.log
>> >>
>> >> Here is a sample log..
>> >>
>> >> {address space usage: 670326784 bytes/639MB} {rss usage: 480604160
>> >> bytes/458MB} [pid: 6823|app: 0|req: 1711887/5151015] 74.125.191.21 ()
>> >> {36 vars in 481 bytes} [Fri Oct 19 17:41:51 2012] POST /bidder/ =>
>> >> generated 2 bytes in 11 msecs (HTTP/1.1 200) 1 headers in 59 bytes (2
>> >> switches on core 1198)
>> >>
>> >> I read a post on using cpu-affinity. Can I use cpu-affinity with the
>> >> gevent loop for the three process? Another uwsgi is taking the last
>> >> core. Any advice on tuning -listen and --async and -b would be
>> >> great. The advice of using --reload-on-rss help a lot! Thanks!
>> >>
>> >> This weekend I will try and use unix sockets. I assume the saving of
>> >> that one context will help?
>> >>
>> >> Yup..I am using nginx.
>> >>
>> >>
>> >>
>> >> Thanks
>> >> _______________________________________________
>> >> uWSGI mailing list
>> >> [email protected]
>> >> http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
>> >
>> > _______________________________________________
>> > uWSGI mailing list
>> > [email protected]
>> > http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
>> _______________________________________________
>> uWSGI mailing list
>> [email protected]
>> http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
>
> _______________________________________________
> uWSGI mailing list
> [email protected]
> http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
_______________________________________________
uWSGI mailing list
[email protected]
http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
