2012/9/7 Marcin Deranek <[email protected]>:
> Hi Roberto,
>
> On Thu, 6 Sep 2012 11:42:04 +0200
> Roberto De Ioris <[email protected]> wrote:
>
>> --idle <n> simply destroy ALL of the workers while the whole instance
>> is inactive.
>
> I found that out hard way :-(( Any plans to have something more
> granular eg. on per worker basis ?

--idle option is for stopping ALL workers when the whole app is idle.
If You want uWSGI to autotune the number of workers based on current
load than use --cheaper.

> As far as I can see all these options are used to tune uwsgi to
> quicker react to load spikes and quicker spawn new workers. None of them
> addresses "how long you want to keep idle workers around" just in case.
> Seems like with 1.2.x branch there's no way to keep workers around just
> in case: in my case spawning worker is more expensive than keeping it.

With 1.2.x You can only use backlog and spare cheaper algos, but keep
in mind that they are agressive and they will try very hard to keep
only minimum number of workers running, unless load requires spawning
additional workers.
With 1.3-rc3 (it's as stable as 1.2.x, at least for me) You can use
cheaper busyness algo, which might be better in Your case. See
http://projects.unbit.it/uwsgi/wiki/CheaperBusyness

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

Reply via email to