On Tue, Dec 13, 2011 at 01:15:00PM +1100, Graham Dumpleton wrote: > On 13 December 2011 08:09, Rodrigo Campos <rodr...@sdfg.com.ar> wrote: > > Thanks a lot for your quick answer =) > > > > No, static files should be served by nginx. We are using it as a proxy. > > Apache > > should serve only de Python App. Why ? :) > > If there is no static files being served and only dynamic Python > stuff, can continue to use embedded/prefork. You would though just > want to setup Apache MPM settings to control better number of child > worker processes and pre start them and leave them running rather than > allowing Apache to dynamically adjust number of processes.
I'm kind of doing this right now. StartServers and MinSpareServers are the same (50) and MaxSpareServers, MaxClients and ServerLimit are set to 60 and MaxRequestsPerChild to 4000. Is this what you mean ? I was considering to lower it even more, since nginx timeout for proxy mode default value is 50/60 sec (I checked, don't remember right now), to have more a "one on one" apache process per WSGI process. Or perhaps 2 apache process per WSGI process, I was going to try. Any suggestions here ? :) And increasing the process parameter in the WSGIDaemonProcess directive did a great difference (request queueing in newrelic grows a lot when there are many requests at the same time, and it seems they went down even when there are many request with a higher process parameter), so I was planning to increase it up to 16 (on a quad-core) instead of the actual value of 4. I wasn't sure about 16 either, perhaps it is too overloaded, but I was goint to try that first (suggestions are very welcome, of course :)). But if I increase it right now it seems that poorly written queries get stucked on the DB server. So other people is tackling that first :) A value of 16 WSGI process with 32 apache process, in "theory" (my calculator), seems to leave enough RAM free and the perf test I did (kind of quick ones with "ab", nothing really serious) where ok. Is there anything else I should take into account ? > Do this and can use less processes than worker MPM and daemon mode. Really *less* ? Or comparable ? Let's say I have N WSGI process and 2*N apache process. If I use worker I will keep the N WSGI process and use less apache process but with threads. Even if I use N process with prefork, the "worker" equivalente should be less process. Perhaps it is not substancial, since WSGI process use a lot more memory that apache prefork process. What am I missing ? Or you are saying its not a substancial difference ? Thanks a lot, Rodrigo -- You received this message because you are subscribed to the Google Groups "modwsgi" group. To post to this group, send email to modwsgi@googlegroups.com. To unsubscribe from this group, send email to modwsgi+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/modwsgi?hl=en.
