Skip to content

jupyter_notebook_ssh.md

Latest commit

 

History

History
99 lines (75 loc) · 5 KB

jupyter_notebook_ssh.md

File metadata and controls

99 lines (75 loc) · 5 KB
Raw

How to access a Jupyter notebook on a remote machine locally

This works only if machines are on the same network. The set up required to securely access a remote Jupyter notebook across different networks is significantly more involved.

  1. ssh into the remote machine and initiate a Jupyter notebook but with the no browser option. Set any port number in the option (8889 here is arbitrary)
    • jupyter notebook --no-browser --port=8889
    • Note the access token when the notebook starts. This is a random sequence of characters found in the URL http://localhost:8889/?token=ACCESS_TOKEN_HERE
  2. On a separate terminal window, create an ssh tunnel to your remote machine binding a local port to the port you set in the remote machine
    • ssh -N -L localhost:8887:localhost:8889 $USER@$IPADDRESS
    • Enter localhost:8887 on your browser to access the remote Jupyter notebook
    • If you are asked for an access token, you can find it in the terminal where you activated the remote Jupyter notebook (see 1.)
    • If you get a "bind: Address already in use" error, use a different port number. It may be that you are already using that port number on a local Jupyter notebook.

Summary

  • jupyter notebook --no-browser --port=8889 (remote)
  • ssh -N -L localhost:8887:localhost:8889 $USER@$IPADDRESS (local)
  • localhost:8887 (local browser)

Start a remote Jupyter Lab session and use it locally

Starts a jupyter lab session in remote machine and binds remote port to local port. This allows you to work on a jupyter notebook locally that has been spawned from a remote computer. This function will cd into the project directory, activate the specified conda environment, and initiate a jupyter lab notebook without the browser. Each step requires you to hit control + c to initiate the next step.

Side note

To avoid having to enter passwords, copy public key to host computer by copying LOCAL_HOME/.ssh/id_rsa.pub to a file named REMOTE_HOME/.ssh/authorized_keys

Requirements (on remote pc):

1. miniconda
    - To install miniconde, do a `wget` on one of the installation urls from the [miniconda installation page](https://docs.conda.io/en/latest/miniconda.html#linux-installers) to download the installation script.
2. JupyterLab
    - To install JupyterLab, do `conda install -c conda-forge jupyterlab` following installing of miniconda.
    - Install it outside of the base virtual environment to ensure that it is available for all virtual environments

Copy the code below into your .bashrc or .zshrc file and follow the usage instructions in the doc string:

jupyssh() {
    # Usage:
    # jupyssh <project path in remote machine> <project environment name in remote machine>
    # e.g. jupyssh ~/Dropbox/data_science/fastai_deeplearning/pt1/fastai fastai

    # Information from remote computer
    USER=<USERNAME>
    IPADDRESS=<REMOTE IP ADDRESS>
    ANACONDA_PATH=/home/$USER/miniconda3

    projectpath=$1
    envname=$2

    if [ $1 = "stop" ]; then
        echo "Stopping...";
        ssh $USER@$IPADDRESS \
            "${ANACONDA_PATH}/envs/${envname}/bin/jupyter lab stop 8889;"
        return 0
    fi

    # Find and kill process locking port 8887 in local computer
    # (in case it is being used by a previous connection that was not closed)
    lsof -ti:8887  -sTCP:LISTEN | xargs kill

    # Initializes conda environment and starts a jupyter lab session
    echo "Initialize a Jupyter Lab session"
    echo "To close session, do \"jupyssh stop ENVNAME\""

    ssh -f $USER@$IPADDRESS \
      "cd ${projectpath};
      source ${ANACONDA_PATH}/bin/activate ${envname};
      ${ANACONDA_PATH}/bin/conda env list;
      ${ANACONDA_PATH}/envs/${envname}/bin/jupyter lab stop 8889;
      ${ANACONDA_PATH}/envs/${envname}/bin/jupyter lab --no-browser --port=8889; exit"

    # Binds the remote port 8889 to local port 8887
    echo "Open browser at http://localhost:8887"
    ssh -N -L localhost:8887:localhost:8889 $USER@$IPADDRESS
}

Reference: https://coderwall.com/p/ohk6cg/remote-access-to-ipython-notebooks-via-ssh

Listing and killing servers

  • On the remote pc, you can list all activate servers with jupyter server list

  • To kill a server, get the port number from the command above, then get the port number with (replace PORTNUMBER)

    sudo lsof -iTCP:PORTNUMBER -sTCP:LISTEN

    Then using the PID number, kill it with kill PIDNUMBER

Jupyter Notebooks on GCP Compute Engine

  1. SSH into example-instance and bind remote port 5000 to the local port 2222:
  2. Once you are connected, run jupyter notebook in port 5000
    • jupyter notebook --port=5000 --no-browser
    • You can then send the running process in the background with Control + z, which sends it in a suspended state. To start it, enter bg in command line.
  3. In your local machine's browser, visit http://localhost:2222/ and enter the notebook’s token to access the notebook