launchpad-dev team mailing list archive

[Maris Fogels <maris.fogels@xxxxxxxxxxxxx>]

On 07/20/2010 11:23 AM, John Arbash Meinel wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Maris Fogels wrote:
> > Hi everyone,
> >
> > During the Epic Robert and I coded up and landed a patch that generates
> > KCacheGrind profiles for launchpad requests.  KCacheGrind is a powerful
> > tool for exploring call graphs, and this change should make it easier to
> > see what pieces of Python are taking the most time to run.
> >
> > To switch on profiling, add this to
> > configs/development/launchpad-lazr.conf:
> >
> >
> >      [profiling]
> >      profile_requests: True
> >
> >
> > Then start the dev app server with 'make run', and use curl or wget to
> > make *one* request to the URL you want:
> >
> >      $ make run
> >
> >      $ curl -k https://launchpad.dev/
> >
> >      $ kcachegrind
> > 2010-07-20_10\:01\:46.680-RootObject\:index.html-OOPS-1662X1-Dummy-2.prof
>
> This is a neat start. Obviously ++profile++ even on some sort of staging
> would be more useful just because of the size of data. Anyway, how does
> it know what directory to dump the profile to? Just wherever 'make run'
> was issued?
>
> John
> =:->

There is a "profile_dir" configuration key that defaults to '.' on development 
systems.  So by default it will just dump the *prof files in the directory that 
'make run' was called from.

For ++profile++ the goal is to add a new key to allow_profiling (already done), 
get Zope to ignore the ++profile++ part of the URL (not done), then turn on 
on-demand profiling on staging (LOSAs).


Maris