remote_ikernel 0.2.8

Running IPython kernels through batch queues

Launch IPython/Jupyter kernels on remote systems so that they can be used with local noteboooks.

Kernels start through interactive jobs in batch queue systems (only SGE and SLURM at the moment) or through SSH connections. Once the kernel is started, SSH tunnels are created for the communication ports are so the notebook can talk to the kernel as if it was local.

Commands for managing the kernels are included.

Install with pip install remote_ikernel. Requires IPython version 3.0 or greater and pexpect. Passwordless ssh to the remote machines is also required.

# Install the module ('python setup.py install' also works)

pip install remote_ikernel

# Set up the kernels you'd like to use

remote_ikernel manage

# Add a new kernel running through GrideEngine

remote_ikernel manage --add \
    --kernel_cmd="ipython kernel -f {connection_file}" \
    --name="Python 2.7" --cpus=2 --pe=smp --interface=sge

# Add an SSH connection to a remote machine running IJulia

remote_ikernel manage --add \
    --kernel_cmd="/home/me/julia-79599ada44/bin/julia -i -F /home/me/.julia/v0.3/IJulia/src/kernel.jl {connection_file}" \
    --name="IJulia 0.3.8" --interface=ssh \
    --host=[e-mail] --workdir='/home/me/Workdir' --language=julia

# Set up kernels for all your local virtual environments that can be run
# from a single notebook server.

remote_ikernel manage --add \
    --kernel_cmd="/home/me/Virtualenvs/dev/bin/ipython kernel -f {connection_file}" \
    --name="Python 2 (venv:dev)" --interface=local

The kernel spec files will be installed so that the new kernel appears in the drop-down list in the notebook.

Warning

IJulia kernels don’t seem to close properly, so you may have julia processes lingering on your systems. To work around this edit the file ~/.julia/v0.3/IJulia/src/handlers.jl so that shutdown_request calls run(`kill $(getpid())`) instaed of exit().

Changes for v0.2

• --remote-launch-args can be used to set qlogin parameters or similar.

• --remote-precmd allows execution of an extra command on the remote host before launching a kernel.

• Better compatibility with Python 3.

• Kernel output on terminals with --verbose option for debugging.

• Connect to a host with ssh, slurm, or local kernels.

• Changed prefix to rik_.

• kernel_cmd now requires the {connection_file} argument.

• remote_ikernel manage --show command to show existing kernels.

• Specify the working directory on the remote machine with --workdir.

• kernel-uuid.json is copied to the working director for systems where there is no access to the frontend filesystem.

• Added compatibility layer to get rid of Jupyter warnings.