[IPython-User] ipcluster, profiles, startup scripts and the web notebook

Matthew Turk matthewturk@gmail....
Fri Jul 27 16:19:05 CDT 2012

Hi Min,

On Fri, Jul 27, 2012 at 4:04 PM, MinRK <benjaminrk@gmail.com> wrote:
> On Fri, Jul 27, 2012 at 10:53 AM, Matthew Turk <matthewturk@gmail.com>
> wrote:
>> Hi there,
>> I'm trying to set up a custom profile for IPython running on my local
>> machine under MPI, but I'm running into trouble.  This is with the
>> released 0.13.  I'd like at startup for a couple commands to be run,
>> to signal to my Python package that it should expect to be running in
>> parallel (with MPI).  Unfortunately, I'm not quite getting the
>> behavior I'd expect, and I think it must be something simple on my end
>> that I'm not understanding.
>> I've created a new profile with:
>> ipython profile create --profile yt --parallel
>> Into the startup directory I've placed the normal startup commands --
>> in this case, the main import for the module I intend to work with --
>> inside a 00-startup.py file.  I'm able to spawn new notebooks that
>> start up as I'd expect if I do "ipython notebook --profile=yt".  So
>> that seems to work properly.  For starting in parallel, I've modified
>> the generated profile in a few places:
>>  * I've set c.IPClusterEngines.profile = u'yt' in ipcluster_config.py
>>  * I've changed c.IPClusterEngines.engine_launcher_class to "MPI" in
>> ipcluster_config.py
> Exactly right.
>>  * In ipengine_config.py I've set c.IPEngineApp.profile = u'yt' (I was
>> not sure if this was either unnecessary or redundant)
> Yes, setting profile in config files is indeed always redundant (both cases,
> here).
>>  * In ipengine_config.py I've added a startup command in
>> c.IPEngineApp.startup_command
> Re-reading the code, this config was broken when the forked Kernel
> implementations were merged for 0.13.
> I think right now it is impossible to trigger startup code on engines in
> 0.13 or master.
> Can you open an Issue for this?
> The reason being that Kernels/Shells are shared, but the Application objects
> are still separate, and that is where this startup config lives.
> Ultimately, IPEngineApp should go away, but we aren't quite there yet.

Ah, okay.  Thanks for digging into this.  I've opened issue #2213:




>> In the web notebook launcher, changes to the number of processors is
>> reflected in what happens when I click on "start".  And looking at the
>> output of top/ps I also see that the engines have been launched with
>> mpiexec as I have requested.  I'm also able to connect to them and
>> verify that they have been launched, but running "dir()" doesn't show
>> the expected imports, from the startup_command.  I'm also unable to
>> get anything to be run if I set c.MPI.init_script to be an existing
>> script.  Am I missing something with how to set up the configuration?
>> Thanks for any ideas,
>> Matt
>> _______________________________________________
>> IPython-User mailing list
>> IPython-User@scipy.org
>> http://mail.scipy.org/mailman/listinfo/ipython-user
> _______________________________________________
> IPython-User mailing list
> IPython-User@scipy.org
> http://mail.scipy.org/mailman/listinfo/ipython-user

More information about the IPython-User mailing list