Hi All,

Currently I encounter one problem when I run the python streaming.

I import a third-party module 'Pandas'.

It's successful when I directly run the python code.
Also successful when run the pig script in local mode.

But has error when run pig script in MapReduce, to debug I comment all of
the code expect printing out one line.
Still does not work.
When I comment the 'import pandas', it works.

no problem at all when 'import numpy'.

So is there anything I missed in the pig script to run the python-pandas?

Thanks,
Dongliang

On Wed, Jan 23, 2013 at 01:58:29PM +0800, Dongliang Sun wrote:
> I import a third-party module 'Pandas'.
>
> It's successful when I directly run the python code.
> Also successful when run the pig script in local mode.
>
> But has error when run pig script in MapReduce, to debug I comment all of
> the code expect printing out one line.
> Still does not work.
> When I comment the 'import pandas', it works.

Is Pandas installed in a virtual environment? Then the problem is
probably that Pig/Hadoop starts your job in a completely fresh
environment: the Python interpreter is invoked from e.g. /usr/bin and
doesn't know anything about the packages installed in the site
packages path of your virtual environment and fails.

How do you invoke your script? Does it start with a shebang?  How is
the script installed? Can you also provide the full trace-back?  You
should find it in the error logs of the job.

Can you boil down the code to a minimal example that fails? Both, the
Python and the Pig code?

Regards,
Thomas Bach.