Create custom IPython kernels for using jupyter notebooks at NERSC.

When you try to connect to jupyter-dev, it will read your JUPYTER_PATH. If for some reasons, there are packages in your path in conflict with the default paths of jupyter-dev, you will not be able to connect. If you encounter such problem, here is a solution:

• Edit your .bashrc.ext (or whatever file where you declare your paths) and reorganise it the following:

export NERSC_HOST=`/usr/common/usg/bin/nersc_host`
if [ ${NERSC_HOST} == "cori19" ]; then
    export NERSC_HOST="toto"
fi

if [ ${NERSC_HOST} == "cori" ]; then
    <all your declarations and module loads>
fi

• By doing so, if you connect to jupyter-dev (cori19), you won't load all your custom packages. However, if you connect to cori, you will load everything as usual.
• Then on cori, run the script makekernel.py and create a kernel with all your custom paths:

python makekernel.py -kernelname mykernel -py_version 2.7

or

python makekernel.py -kernelname mykernel -py_version 3.6

• It will set the $LD_LIBRARY_PATH, $PYTHONPATH, and $PATH. You can choose python2.7 or python3.6 (latest anaconda-4.4).
• Connect to https://jupyter-dev.nersc.gov/hub/login and create a notebook with the kernel mykernel${py_version} you just created (which contains all your favourite paths).

It is easy to modify the script to generate any kind of custom kernels. Feel free to edit it!