This repository provides a library and sample notebook which enable access of google colab from a local machine using SSH. An additional, although straightforward step is to configure an IDE to use a remote python interpreter. The guide was presented in a forum post and in accompanying repository that contains a tutorial notebook on how to do the stuff (in polish). Here I did a restructure and simplification of the code using a library and a simple notebook.
A relevant and most complete repository is colab-ssh which is based on SSH tunneling to connect to the colab, however it utilizes cloudflared. Apart from using a public service there are also a self-hosted solutions which I will investigate and integrate later:
Another interesting project is colabcode which runs code-server (VSCode in browser) on Google Colab and Kaggle Notebooks.
The following steps should work for Linux, Windows, macOS. For Windows you need also to have OpenSSH installed as shown in a relevant guide.
First we need to create and configure an account in a port forwarding service like portmap.io. This offers one port mapping for free. This step in required only once.
- Create & login to an account in portmap.io
- Create a new configuration
- The configuration name is fixed as 'first'
- Download the private key. It's name is in the form
username.first.pem
- Create a new mapping
- Leave everything as is and fill a port number to the field "Port on your PC". For simplicity I choose the same port as the above.
- Save the mapping and keep somewhere the rule, which should be in the form:
tcp://<username>-<remote_port>.portmap.host:<remote_port> => <local_port>
- Generate a corresponding public key for the downloaded private key with the following command:
ssh-keygen -y -f <username>.first.pem > public_key.pub
Second we need to keep somewhere the configuration files and the keys. I choose to keep them in google drive rather than as plain text in the notebook. Then the colab notebook will mount and use the files from the google drive.
- Create a desired folder structure in google drive. For example:
<google_drive_root>/colab_data/tunnel_options - Upload the private and public keys in this folder
- Upload a user_config.json file containing the port mapping information. A template can be found in this repo.
A sample notebook which showcases the setup of the SSH tunnel is included in docs. You can also click on the following button to directly test it:
The above notebook outputs the command on how to connect to the VM using SSH.
Instructions on how to setup PyCharm for connecting to the interpreter are shown in the relevant documentation.
During configuring the server you have to put the hostname and the port (same as in user_config.json) and root as user. Instead of giving a password we point to the .pem private key in our local machine. The correct interpreter which we need to select is in /usr/local/bin/python as the /usr/bin/python is for python2.