The high load on web2py nodes seems to point to code in web2py. If it were a problem with postgres you would have a high load on postgresql and a lot of wait state and little CPU time resulting in little uptime on web2py nodes but long page rendering times. I suggest to try to convert some logic to use raw resultsets using executesql instead of DAL Row objects. But before doing that try the query on postgres directly: you can use the _select() method to obtain the query generated by the DAL. If postgresql answers slowly try adding indexes on columns as requested by EXPLAIN. If postgresql answers fast try the guilty query with the DAL in a python shell (i.e. python web2py.py -M -S <appname>). If it slow than you have found the cause.
Else keep using top to find if other processes are infesting the CPU maybe it is a simple problem of "ping pong" or swappiness. Simple tuning of uWSGI could suffice. As a rule of thumb you should not have the number of web2py processes be more than twice the number of cores. mic 2012/5/10 Bruce Wade <bruce.w...@gmail.com>: > Web2py is on 3 different servers/nodes, postgresql is on it's own node with > 8GB ram. > > CPU is being used by uwsgi so web2py. The slowness I think is from DB > queries as when you load a page without the DB involved much it loads > quickly > > The serving ads part is not a problem it is the other pages on the website. > At least not the adviewer the banner ads are new. The adviewer has served > over 29 million ads. > > I will try disabling the banner ads for now and set them so they are > querying from a completely different server, maybe using mongodb and node.js > > > On Thu, May 10, 2012 at 11:28 AM, pbreit <pbreitenb...@gmail.com> wrote: >> >> Is your traffic from serving ads or users coming to your web site? Have >> you exhausted caching opportunities? > > > > > -- > -- > Regards, > Bruce Wade > http://ca.linkedin.com/in/brucelwade > http://www.wadecybertech.com > http://www.fittraineronline.com - Fitness Personal Trainers Online > http://www.warplydesigned.com >
