Why not just go with upstream on Nginx?
upstream backend {
server 127.0.0.1:1234;
server unix:/var/run/example.com.sock;
}
uwsgi_pass backend;
got RR by default.
----- Oryginalna wiadomość -----
> Od: "Łukasz Mierzwa" <[email protected]>
> Do: [email protected]
> Wysłane: wtorek, 29 listopad 2011 10:49:45
> Temat: [uWSGI] fastrouter load balancing
>
> Hi,
>
> I'm doing some stress test of uWSGI with 3 servers:
>
> nginx -> uwsgi fastrouter -> 2x uwsgi backend
> (nginx and fastrouter are on same server)
>
> and one backend gets 2x more requests. I've tested it with apache
> benchmark
> using:
>
> ab -n 10000 -c 1 -H "Host: app.domain.com"
> "http://$nginx_address/$some_url";
>
> - c 1 was used so that it's not casued by one backend responding
> faster then
> the other, requests are issued one after one so if fastrouter uses
> round robin
> to load balance, both backend should get equal number of requests but
> it does
> not. So how does fastrouter dispatch request by default?
> http://projects.unbit.it/uwsgi/wiki/FastRouter shows that one can
> write custom
> code for load balancing but all I want right now is simple rr over
> all
> subscribed vassals, how can I achieve that?
>
> results:
> 172.16.200.55: 6667 requests
> 172.16.200.56: 3333 requests
>
> second run, this time I've reloaded vassal on first node just after
> starting
> ab:
> 172.16.200.55: 3303 requests
> 172.16.200.56: 6679 requests
> so now second node got twice as much requests.
> It seems that by default fastrouter will prefer first working node
> (?).
>
> log from fastrouter startup:
> [2011-11-29 09:52:28.789360] [('172.16.200.53', 55923)] received
> message:
> [uwsgi-subscription] new pool: app.domain.com
> [2011-11-29 09:52:28.789494] [('172.16.200.53', 55923)] received
> message:
> [uwsgi-subscription] app.domain.com => new node: 172.16.200.56:3001
> [2011-11-29 09:52:29.721830] [('172.16.200.53', 55923)] received
> message:
> [uwsgi-subscription] app.domain.com => new node: 172.16.200.55:3001
>
> adding stats option to fastrouter instance doesn't give any
> fastrouter
> statistics but looking at the cpu usage on backends during the ab
> run, shows
> that one node constantly gets twice the load, and requests log
> confirms it:
>
> (ip,source port)
> '172.16.200.56',33894
> '172.16.200.56',33894
> '172.16.200.55',55817
> '172.16.200.56',33894
> '172.16.200.56',33894
> '172.16.200.55',55817
> '172.16.200.56',33894
> '172.16.200.56',33894
> '172.16.200.55',55817
> '172.16.200.56',33894
> '172.16.200.56',33894
> '172.16.200.55',55817
>
> Łukasz Mierzwa
> _______________________________________________
> uWSGI mailing list
> [email protected]
> http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
>
--
Łukasz Wróblewski
http://www.nri.pl/ - Nowoczesne Rozwiązania Internetowe
http://www.hostowisko.pl/ - Profesjonalny i tani hosting
http://www.katalog-polskich-firm.pl/ - Najlepszy darmowy katalog firm
_______________________________________________
uWSGI mailing list
[email protected]
http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
