[Zope] Zope woes continue - server going down regularly. (possible solution)

[Tres Seaver]

Michel Pelletier <michel@digicool.com> wrote:
> > -----Original Message-----
> > From: Cayce Ullman [mailto:c_ullman@yahoo.com]
> > Sent: Wednesday, December 08, 1999 11:26 AM
> > To: zope@zope.org; panda@skinnyhippo.com
> > Subject: [Zope] Zope woes continue - server going down regularly.
> > (possible solution)
> >
> >
> > >>> Alternatively, the server just hangs and doesn't
> > respond
> > >>> (but ps -aux shows the process is running).
> > >>
> > >>Is it spinning? (consuming 100% CPU resources?) or
> > hung? (consuming
> > >>none..)
> >
> > I noticed a couple of weeks ago that my Zope,
> > started spinning when I hit a certain page 5 or
> > 6 times.  I noticed that this page was different
> > from the others as it tried to do a Zope.app().
> > This was interesting cause the actual Zope.app()
> > call worked, except it would eventually hose the
> > server.  I figured I was just doing something wrong
> > and figured out another way to do it.  Anyway,
> > here's how to reproduce the problem :
> >
> > Create a external method that looks like this :
> >
> > import Zope
> >
> > def helloworld(self):
> >     x = Zope.app()
> >     return "Hello World"
> 
> This is because the connection gets opened, but never closed.  Zope 2.x
> is hardwired to have only 7 available database connections.  We'll
> probably add a knob in 2.2 that lets you configure the number of db
> connections.  In either case, the problem is not the hardwiring, the
> problem is that your code has a bug in it, because you do not free the
> resource you are consuming.  Do an:
> 
>   x.close()
> 
> at the end of your method and you won't run out of connections.

There is stuff in ZODB.ZApplication.py which looks as though it should not be
necessary to close a ZODB connection manually:


        conn=db.open(version)

        # arrange for the connection to be closed when the request goes away
        cleanup=Cleanup()
        cleanup.__del__=conn.close
        REQUEST._hold(cleanup)

Am I misreading this?  Or do we have a refcount cycle in here somewhere?
-- 
================================================================================
Tres Seaver            tseaver@palladion.com      http://www.palladion.com
Palladion Software     Houston, Texas, USA        713-523-6582