This JupyterLab Extension allows users to manage Kernelspecs from within JupyterLab.

ksmm is a temporary name, originally standing for Kernelspec Manager and currently ships:

• Kernelspec creation based on parametrized templates.
• Kernelspec Editing: name, attributes.
• Kernel Duplication.
• Kernel Deletion.

On large distributed systems, it is common to wish to parametrize kernels and choose parameters for a remote environment, like number of CPU, Memory limit, presence of GPU. Or even set other parameters in environment variables.

This currently requires to create a new Kernelspec for jupyter using the command line which can be a tedious and complicated task.

Modifying existing Kernelspec also does not always works as they are cached on a per notebook.

This is an attempt to provide a UI based on json-schema and templates, for end users to easily create, duplicate and modify kernelspec, without being exposed to too much of the internal details.

You will need to install some Kernelspec templates.

make install-kernelspecs

This will install the python-template-1 Kernelspec example located in the examples folder into your system environment.

Ensure you have JupyterLab 3.1+, and then run this command the ksmm extension inside your current JupyterLab environment.

pip install --upgrade ksmm

Use the provided environment.yaml to install the conda environment.

conda deactivate && \
  make env-rm && \
  make env
conda activate ksmm

# Install the server and frontend in dev mode.
make install-dev

# In terminal 1, Start the jupyterlab.
# open http://localhost:8234?token=...
make jlab

# In terminal 2, start the extension building in watch mode.
make watch

When making changes to the extension you will need to issue a jupyter labextension build, or, start jlpm run watch in the root of the repository to rebuild on every changes. You do not need to restart or rebuild JupyterLab for changes on the frontend extensions, but do need to restart the server for changes to the Python code.

You system adminstrator can create Kernelspect templates for you. As a user, if you click on the picker icon of a template card, you will be prompted for the Kernelspec parameters.

When you will click on the Create Kernelspec button, a new Kernespec will be created.

This is an example of such a Kernelspec template. The metadata/template/tpl stanza should contain a Json Schema compliant structure. You can browser the react-jsonschema-form for examples.

You can use the metadata/template/mapping stanza to create visual mappings (e.g. Small will be mapped to 102400). The example/python-template-1 contains an example. To install that example template in your environment, you need to run jupyter kernelspec install ./examples/python-template-1 (add --user to install in your user space).

{
  "argv": [
    "python",
    "-m",
    "ipykernel_launcher",
    "-f",
    "{connection_file}",
    "--cache-size={cache_size}",
    "--matplotlib={matplotlib}"
  ],
  "display_name": "Python 3.8 Template 1",
  "language": "python",
  "metadata": {
    "template": {
      "tpl": {
        "argv": [
          "python",
          "-m",
          "ipykernel_launcher",
          "-f",
          "{connection_file}",
          "--cache-size={cache_size_map}",
          "--matplotlib={matplotlib}",
          "--logfile={logfile}",
          "--Kernel._poll_interval={poll_interval}"
        ],
        "display_name": "Python cache_size {cache_size_map} matplotlib {matplotlib}"
      },
      "parameters": {
        "poll_interval": {
          "type": "number",
          "minimum": 0.01,
          "maximum": 1,
          "multipleOf": 0.01,
          "title": "Kernel pool interval in seconds",
          "default": 0.01
        },
        "cache_size": {
          "type": "integer",
          "title": "Set the size of the cache",
          "default": "Medium",
          "enum": [
            "Small",
            "Medium",
            "Big"
          ]
        },
        "matplotlib": {
          "type": "string",
          "title": "Configure matplotlib for interactive use with the default matplotlib backend",
          "default": "widget",
          "enum": [
            "auto", "agg", "gtk", "gtk3", "inline", "ipympl", "nbagg", "notebook", 
            "osx", "pdf", "ps", "qt", "qt4", "qt5", "svg", "tk", "widget", "wx"
          ]
        },
        "logfile": {
          "type": "string",
          "title": "Set the path for the logfile",
          "default": "/tmp/kernel.out"
        }
      },
      "mapping": {
        "cache_size_map": {
          "cache_size": {
            "Small": "102400",
            "Medium": "512000",
            "Big": "1048576000"
          }
        }
      }
    }
  }
}

To publish a release, you need to manually bump the version number of the package.json file, this this diff for example.

 {
   "name": "@deshaw/jupyterlab-ksmm",
-  "version": "0.1.4",
+  "version": "0.1.5",
   "description": "An extension to manage Kernelspecs from JupyterLab",
   "keywords": [
     "jupyter",

Pleas follow Semantic Versioning rules when bumping the version number.

Commmit and push your changes, then run the following comamand which clean, build and push the needed artifact into the PyPi Ksmm project (ensure you have been given the needed authorization for that).

make publish

At some point, it would be interesting to use the https://github.com/jupyter-server/jupyter_releaser tool (tracked in #81).

This was created by the D. E. Shaw group in conjunction with Quansight.

We love contributions! Before you can contribute, please sign and submit this Contributor License Agreement (CLA). This CLA is in place to protect all users of this project.