Running remote jupyter notebooks in a local browser window.

Free software: MIT license

conda install -c conda-forge jump
jump REMOTENAME

For options, type

jump --help

Starting a new notebook server in remote conda environment myenv on a remote machine mycluster

jump mycluster -e myenv --new

Starting a new jupyter lab server in remote conda environment myenv on a remote machine mycluster

jump mycluster --lab -e myenv --new

Connecting to an existing jupyter server

jump mycluster -e myenv

Starting a new notebook server with remote module cuda/9.2 loaded for GPU support

jump mycluster -e myenv --new -m cuda/9.2

Killing a notebook server

jump mycluster -e myenv --kill

On local (UNIX) machine:

• plumbum and click (are installed by the conda install command)
• ssh

On remote machine:

• anaconda or miniconda
• jupyter notebook (at least installed in one conda environment, jupyter > 5.1 required for the --kill option)
• jupyter lab (to support the --lab option)
• recommended: nb_conda

Windows systems are not supported.

In order to use this script efficiently, it is desirable to have an efficient setup in your ~/.ssh/config file.

Concretely, you should prevent your ssh connection from prompting for a password everytime you rune a local command, like this:

Host *
    ControlMaster auto
    ControlPath /tmp/ssh_mux_%h_%p_%r
    ControlPersist 60m
    ServerAliveInterval 90

Furthermore, you will want to set up shortcuts for the servers that you use most often:

Host YOUR_SHORTCUT
    HostName FULL_REMOTE_NAME
    DynamicForward 8080
    User YOUR_USER_NAME_ON_THE_REMOTE`

If you have remotes that require tunneling through a login node, you may also want to define those remotes explicitly:

Host NAME_OF_REMOTE
    ProxyCommand ssh FULL_LOGIN_NODE_NAME_OR_SHORTCUT -W %h:%p