Skip to content

Latest commit

 

History

History
54 lines (35 loc) · 1.84 KB

jupyter.md

File metadata and controls

54 lines (35 loc) · 1.84 KB
Raw

Run Jupyter Lab or Notebook on a remote server

Link: https://docs.anaconda.com/anaconda/user-guide/tasks/remote-jupyter-notebook/

1. On the remote server

# start a screen session (to keep it open for a long time)
cd ~; screen -S julab -L -Logfile julab.log

# load the python environment
load_insar

# launch a jupyter instance and port to a specific random number
# NOTE: this port number **MUST BE DIFFERENT** for each user
jupyter lab --no-browser --port=1236

# Ctrl-A D or Ctrl-A Ctrl-D to detach from the screen session

2. On your laptop terminal:

Create an alias in the ~/.bash_profile file.

alias julab_kunlun='ssh -N -f -L 8080:localhost:1236 <REMOTE_USER>@<REMOTE_HOST>'

where -N is for not executing a remote command and just forwarding ports, the -f option requests SSH to go to the background, and 8080 is the local port number. The <REMOTE_USER>@<REMOTE_HOST> could also be the host profile name in your ~/.ssh/config file, e.g. host_kunlun.

Then run julab_kunlun to forward the port connection or to re-connect whenever it's lost.

3. On your laptop web browser:

Go to the web address: http://localhost:8080/lab. On the first time, you may be asked for a token, it's in the output of the jupyter lab command in the screen session. Bookmark this web address for easy re-opening.

Tips:

  1. Add the following in your jupyter notebook to detect and switch between different platforms:
import platform

if platform.system() == 'Linux':
    proj_dir = os.path.expanduser('~/data/2023TurkeyEQ')             # on linux server
else:
    proj_dir = os.path.expanduser('~/data/archives/2023TurkeyEQ')    # on macos/windows laptop
  1. If julab_kunlun returns address already in use error, run the following to kill the process, and then re-connect.
lsof -ti:8080 | xargs kill -9