On Oct 13, 8:43 pm, "Valts Mazurs" <[EMAIL PROTECTED]> wrote:
> On Mon, Oct 13, 2008 at 12:15 PM, xavier dutoit <[EMAIL PROTECTED]> wrote:
> > Thanks for your reply. I do already run php under fcgi (as separated
> > server, as you point out). But the main difference is that this pool
> > of fcgi php processes (or thread, not sure) are available to all the
> > sites I've configured to use php. I don't have to dedicate one
> > specific fcgi server for each different sites, they all talk to the
> > same backend.
>
> I am afraid that single pool of fcgi resources is not currently possible for
> multiple django sites. Maybe it would be possible if someone wrote a
> specific script/wrapper for Django but, I guess, it would create more risks
> (think about thread safety) than benefits.
It is only possible with mod_python/mod_wsgi because they use C API of
Python to create additional sub interpreters within same process.
There is no feature/module available to Python code to do the same
thing. It would be possible to create special C extension module that
allowed it but transferring request from context of sub interpreter
which accepting the request to another sub interpreter would be tricky
and you could run into lots of problems as a result of data sharing
between interpreters.
> with django config, it seems that it needs separate fcgi servers for
>
> > each site, and I was wondering if these separate servers aren't going
> > to use a lot of resources, or in general being harder to manage than a
> > single pool of fcgi processes shared between different sites.
>
> Single django fcgi process on OpenBSD takes 11-12 Mb of RAM when used in
> prefork mode. You will have at least 2 processes (1 manager, 1 worker) for
> each site. I have not tried threaded setup though.
>
> > However, I read on a blog from Graham
> >http://www.technobabble.dk/2008/aug/25/django-mod-wsgi-perfect-match/
>
> > That there is a lot of shared libs between the various fcgi processes/
> > servers, so it doesn't matter much to have unused processes/servers.
>
> > Is my understanding correct and does it apply both for wsgi and
> > fastcgi ?
>
> Sorry, I didn't really catch the idea :) How could libraries be shared
> between processes?
Dynamically loadable shared libraries even though loaded by
independent process instances will be mapped only once by operating
system. In the larger scheme of things this doesn't help, as each
Django process may still have many tens of MBs of local process
private memory due to Python code caching and application data. It is
this latter memory usage which is the real problem.
Graham
> > The issue being that, of course, I don't know in advance which site is
> > going to be inactive and that's likely to change from one day to the
> > other. Not to mention that changing the number of spare servers
> > manually isn't very high on the list of things I like to do on a daily
> > basis ;)
>
> I meant that maxspare=NUMBER sets maximum number of free/idle processes. If
> the load grows new processes are spawned. When the load drops unneeded
> processes are killed.
>
> Valts.
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups
"Django users" group.
To post to this group, send email to django-users@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at
http://groups.google.com/group/django-users?hl=en
-~----------~----~----~----~------~----~------~--~---
