TurboGears Google Group

Tracking DB connection usage

rent.lupin.r...@gmail.com (James) — Sat, 20 Mar 2010 11:48:56 UT

Hi all,
In a TG2 app, I have SQLAlchemy configured with a pool size of 30
(overflow 30), and the Paste HTTP server configured to run with 50
threads. There is a limit of a maximum 50 concurrent connections
coming to the server.
However, I am regularly seeing SQLAlchemy TimeoutErrors with no
connections available in the QueuePool. It seems that there should

Re: When are controllers instantiated in TG2? Also: best practice to create (and pool?) expensive objects?

rent.lupin.r...@gmail.com (James) — Sat, 20 Mar 2010 11:34:26 UT

Great - I've got a threadlocal connection set up and it works
perfectly.
Thanks Diez!

Re: When are controllers instantiated in TG2? Also: best practice to create (and pool?) expensive objects?

john.lora...@gmail.com (John Lorance) — Fri, 19 Mar 2010 17:23:47 UT

Thanks again for the clarification here.. I thought I would losing my
mind for a moment. So, yes, it becomes obvious that controllers that
are simply class attributes and behave as such, and controllers called
as part of the a _lookup method are normally instantiated objects that
are created and destroyed as appropriate to the request (per _lookup

Re: [TurboGears] Re: When are controllers instantiated in TG2? Also: best practice to create (and pool?) expensive objects?

de...@web.de (Diez B. Roggisch) — Fri, 19 Mar 2010 15:27:40 UT

Am 19.03.2010 um 15:51 schrieb John Lorance:
That's a different thing, yes. Maybe I was the dense here - but the OP
asked (IMHO) for the "normal" controller hierarchy & it's objects,
which (with the exception of the root) aren't instantiated after
initial load. Simply because they are *class*-attirbutes.

Re: [TurboGears] Re: When are controllers instantiated in TG2? Also: best practice to create (and pool?) expensive objects?

john.lora...@gmail.com (John Lorance) — Fri, 19 Mar 2010 14:51:42 UT

Thanks Diez. After a re-read of this thread, I am understanding this more
clearlt... just a weird situation to understand. Additionally, if I look at
the sample code:
class BlogController(BaseController) :
@expose()
def _lookup(self, year, month, day, id, *remainder):
dt = date(int(year), int(month), int(day))

Re: [TurboGears] cherrypy.HTTPRedirect vs. cherrypy.InternalRedirect

de...@web.de (Diez B. Roggisch) — Fri, 19 Mar 2010 09:09:54 UT

Please don't hijack other threads & just change the subject. This is
very confusing for those who prefer their mails being threaded.
Well, the disadvantage is pretty much what you say yourself - the
browser doesn't change the url. This might be desirable, but can also
cause problems if the user hits "reload" and is asked if she's sure

Re: [TurboGears] Re: When are controllers instantiated in TG2? Also: best practice to create (and pool?) expensive objects?

de...@web.de (Diez B. Roggisch) — Fri, 19 Mar 2010 09:06:25 UT

Am 18.03.2010 um 18:53 schrieb John Lorance:
No.
No.
The only controller that currently gets instantiated on a per-request-
base is the root-controller. And I don't read anything else from the
_lookup/_default docs either.
HTH,
Diez

Re: [TurboGears] Re: Install TG 2.1 on Centos issues

mjpe...@gmail.com (Michael Pedersen) — Tue, 16 Mar 2010 04:26:03 UT

And, as you can now see at that ticket (
[link] ), I've proffered up a fix that will
(hopefully) make it into the next release. Especially since it was me that
introduced the bug. Sorry about that.

Re: [TurboGears] Where is the TG2.1 document?

t...@alwaysreformed.com (Tim Black) — Thu, 18 Mar 2010 23:34:35 UT

The linked headings here [link] are
no longer underlined when I mouseover them. Of course, my mouse pointer
turns into a hand, which means the headings are a link, but I still bet
other people are going to fail to find the documentation because it's
hard to know those headings are the only links into the documentation.

cherrypy.HTTPRedirect vs. cherrypy.InternalRedirect

pito...@fns.uniba.sk (Michal Pitonak) — Thu, 18 Mar 2010 22:47:14 UT

Hi,
Can I have a naive question? Is there is any hidden catch in using
cherrypy.InternalRedirect instead of cherrypy.HTTPRedirect (in deployed
application behind apache, https. ...)? I'd like to use the
InternalRedirect because it hides (at least in the navigation bar) the
url parameters ...
I'm using TurboGears 1.0.8.

Re: When are controllers instantiated in TG2? Also: best practice to create (and pool?) expensive objects?

john.lora...@gmail.com (John Lorance) — Thu, 18 Mar 2010 17:53:05 UT

Diez,
Just so that I am reading you (and this thread) clearly here about
controller instantiation... all controller classes are instantiated as
objects on a per request basis.. yes? Therefore, if I have a sub
controller with an __init__ that does some work needed, common to all
my controller class methods, it can assign local (self) attributes in

Re: tg/elixir question - select distinct problems..

rotemta...@gmail.com (Rotem Tamir) — Thu, 18 Mar 2010 14:08:46 UT

diez
you have been of great help
my team thanks you very much, you have solved a great mystery for
us :)
rotem

Re: challenging issue with logging issues - need guru help please!

robertnevill...@gmail.com (robneville73) — Thu, 18 Mar 2010 13:51:12 UT

[link]
answered my question

Re: [TurboGears] When are controllers instantiated in TG2? Also: best practice to create (and pool?) expensive objects?

de...@web.de (Diez B. Roggisch) — Thu, 18 Mar 2010 00:33:41 UT

Hi,
I've even been surprised by them being instantiated more than once per
process.
But it's not as bad as it looks.
No. The *root* controller is instantiated as often as that. But
clearly all sub-controllers ar class-attributes, thus they aren't
instantiated more than once per process.
Is that wierd? Certainly. If you insist, just make the root-controller

When are controllers instantiated in TG2? Also: best practice to create (and pool?) expensive objects?

rent.lupin.r...@gmail.com (James) — Wed, 17 Mar 2010 20:32:57 UT

Hi all,
When running a tg2 app with "paster serve", I was expecting controller
objects to be instantiated once per worker thread.
However, when I add logging to BaseController.__init__, it seems that
the full hierarchy of controller objects is instantiated once per
request.
E.g. when I exercised my app with ab -c 100 -n 1000 (100 concurrent