Il giorno 14/giu/2011, alle ore 13.26, Łukasz Wróblewski ha scritto:

> ----- Oryginalna wiadomość -----
>> Od: "Roberto De Ioris" <[email protected]>
>> Do: "uWSGI developers and users list" <[email protected]>
>> Wysłane: wtorek, 14 czerwiec 2011 11:53:28
>> Temat: [uWSGI] The Zerg mode
>> 
>> This mode will allows you to dinamically add workers to already
>> running instances, without taking them offline:
>> 
>> http://projects.unbit.it/uwsgi/wiki/ZergMode
>> 
> 
> Reading about this new option I thought about something like this:
> In Emperor mode define the maximum number of free workers, which can be used 
> automatically by any vasall if necessary.
> 
> If additional workers are not needed could release resources.
> 
> Does this make sense?


The only problem is in how the emperor knows who needs workers/zergs.

Probably the instances can communicate their listen queue size to the emperor 
that will launch
zerg but it will works only for tcp sockets on Linux and (as jean baptise 
reported some days ago) we have the problem that
overloaded instances are indistinguishible after the queue is full.

Probably the webserver can inform the emperor of connection refused/timeout 
errors, and the emperor will trigger zergs
for the requested instance. This would be easy in nginx.
--
Roberto De Ioris
http://unbit.it

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

Reply via email to