JupyterHub and Jupyter Notebook – A View Under the Hood

This talk tries to give you an overview of all the little parts that together make up a Jupyter-powered application, and how they fit together. Hint: For clickable links, download the PDF file and view locally.

* The Jupyter Project and its building blocks
* Internal magic: protocols, authenticators, spawners, gateways, kernels, …
* Deploying a local data science platform

- Simple host-based deployments
- Running Jupyter on Kubernetes

* Options for publishing notebooks

⏰ Duration: 30-45min + Q&A

PyData Südwest Meetup · April 2019 in Karlsruhe, Germany
⇒ https://www.meetup.com/PyData-Suedwest/events/258321928/

Transcript

1. Architecture Board
   Architecture Board
   Application Monitoring Service
   Application Monitoring Service
   Phase 1 – Metrics Gateway
   Phase 1 – Metrics Gateway
   Jürgen Hermann
   Karlsruhe · 2018-04-24
   JupyterHub and Jupyter Notebook
   JupyterHub and Jupyter Notebook
   A View Under the Hood
   A View Under the Hood
   jhermann
   jhermann_
   jh@web.de
   

   View Slide

2. Agenda
   Agenda
   ●
   The Jupyter Project and its building blocks
   The Jupyter Project and its building blocks
   ●
   Internal magic: protocols, spawners, gateways, …
   Internal magic: protocols, spawners, gateways, …
   ●
   Deploying a local data science platform
   Deploying a local data science platform
   ●
   Simple host-based deployments
   Simple host-based deployments
   ●
   Running Jupyter on Kubernetes
   Running Jupyter on Kubernetes
   ●
   Options for publishing notebooks
   Options for publishing notebooks
   ●
   The Jupyter Project and its building blocks
   The Jupyter Project and its building blocks
   ●
   Internal magic: protocols, spawners, gateways, …
   Internal magic: protocols, spawners, gateways, …
   ●
   Deploying a local data science platform
   Deploying a local data science platform
   ●
   Simple host-based deployments
   Simple host-based deployments
   ●
   Running Jupyter on Kubernetes
   Running Jupyter on Kubernetes
   ●
   Options for publishing notebooks
   Options for publishing notebooks
   This talk tries to give you an overview of
   This talk tries to give you an overview of
   all the little parts that together make up
   all the little parts that together make up
   a Jupyter-powered application, and how
   a Jupyter-powered application, and how
   they fit together.
   they fit together.
   This talk tries to give you an overview of
   This talk tries to give you an overview of
   all the little parts that together make up
   all the little parts that together make up
   a Jupyter-powered application, and how
   a Jupyter-powered application, and how
   they fit together.
   they fit together.
   

   View Slide

3. It‘s Simple &
   It‘s Simple &
   Shiny
   Shiny
   

   View Slide

4. The ‘Classic’ Notebook User Interface
   The ‘Classic’ Notebook User Interface
   Source:
   Source: https://ipython.org/
   https://ipython.org/
   

   View Slide

5. The Jupyter Promise
   The Jupyter Promise
   ●
   “Data Science IDE” in your browser
   ●
   File system view (notebook dashboard)
   ●
   Notebook editor with markdown + code cells
   ●
   Runtimes for Julia, Python, R, and lots more
   ●
   Wide selection of (interactive) visualizations
   ●
   Customization with widgets and ‘magics’
   ●
   “Data Science IDE” in your browser
   ●
   File system view (notebook dashboard)
   ●
   Notebook editor with markdown + code cells
   ●
   Runtimes for Julia, Python, R, and lots more
   ●
   Wide selection of (interactive) visualizations
   ●
   Customization with widgets and ‘magics’
   

   View Slide

6. The Python
   The Python
   Advantage
   Advantage
   Scientific /
   Scientific /
   Data Science
   Data Science
   Stack
   Stack
   https://www.slideshare.net/icaromedeiros/why-python-is-better-for-data-science
   https://www.slideshare.net/icaromedeiros/why-python-is-better-for-data-science
   

   View Slide

7. PySpark (Local)
   PySpark (Local)
   

   View Slide

8. Python‘s Visualization
   Python‘s Visualization
   Landscape
   Landscape
   https://youtu.be/FytuB8nFHPQ
   https://youtu.be/FytuB8nFHPQ
   

   View Slide

9. Seaborn
   Seaborn
   

   View Slide

10. Bokeh
    Bokeh
    

    View Slide

11. Altair / Vega
    Altair / Vega
    

    View Slide

12. It‘s
    It‘s
    Complicated
    Complicated
    

    View Slide

13. The Jupyter Universe
    The Jupyter Universe
    ●
    Servers (Web UI)
    ●
    Interfaces / Clients
    ●
    Jupyter API
    ●
    IPython Reference Kernel
    ●
    Servers (Web UI)
    ●
    Interfaces / Clients
    ●
    Jupyter API
    ●
    IPython Reference Kernel
    https://jupyter.readthedocs.io/en/latest/architecture/visual_overview.html
    https://jupyter.readthedocs.io/en/latest/architecture/visual_overview.html
    

    View Slide

14. Single-User Notebook Server
    Single-User Notebook Server
    ●
    Web UI for a single notebook file
    ●
    Talks to a kernel via ØMQ
    (for terminal I/O)
    ●
    Web UI for a single notebook file
    ●
    Talks to a kernel via ØMQ
    (for terminal I/O)
    

    View Slide

15. JupyterHub –
    JupyterHub –
    Multi-User
    Multi-User Notebook Web Service
    Notebook Web Service
    ●
    Centrally managed notebook service
    ●
    Web interface: Classic and/or JupyterLab
    ●
    Configurable proxy for dynamic URL routing
    ●
    Authenticators for user login
    ●
    Spawner to start single-user notebooks
    ●
    Kernel gateway for remote control of runtimes
    ●
    Centrally managed notebook service
    ●
    Web interface: Classic and/or JupyterLab
    ●
    Configurable proxy for dynamic URL routing
    ●
    Authenticators for user login
    ●
    Spawner to start single-user notebooks
    ●
    Kernel gateway for remote control of runtimes
    

    View Slide

16. High-Level Architecture
    High-Level Architecture
    

    View Slide

17. Xeus Kernel Framework
    Xeus Kernel Framework
    ●
    Native kernel development is not-so-easy
    ●
    Thus IPython wrapper as a bridge to runtimes
    ●
    Meet Xeus & Xeus-Cling
    ●
    C++ kernel development framework
    ●
    Initially made for native C++ kernel
    ●
    By now almost has feature-parity
    (compared to IPython)
    ●
    C++ easily integrates with any tech
    ●
    Native kernel development is not-so-easy
    ●
    Thus IPython wrapper as a bridge to runtimes
    ●
    Meet Xeus & Xeus-Cling
    ●
    C++ kernel development framework
    ●
    Initially made for native C++ kernel
    ●
    By now almost has feature-parity
    (compared to IPython)
    ●
    C++ easily integrates with any tech
    

    View Slide

18. repo2docker – Create Customized Kernels
    repo2docker – Create Customized Kernels
    ●
    Repeatability of shared notebooks
    ●
    Turn git repos into Jupyter Docker Images
    ●
    Build JupyterHub-ready images (custom kernels)
    ●
    Inspects repo contents for tech stacks
    ●
    environment.yml, requirements.txt, …
    https://github.com/jupyter/repo2docker
    ●
    Repeatability of shared notebooks
    ●
    Turn git repos into Jupyter Docker Images
    ●
    Build JupyterHub-ready images (custom kernels)
    ●
    Inspects repo contents for tech stacks
    ●
    environment.yml, requirements.txt, …
    https://github.com/jupyter/repo2docker
    

    View Slide

19. Binder – Advanced Technology, Almost Like Magic
    Binder – Advanced Technology, Almost Like Magic
    Combine Docker & JupyterHub, for…
    ●
    Repeatable notebook execution from git
    ●
    Ephemeral custom Jupyter runtimes
    ●
    Builds & spawns ad-hoc kernels (repo2docker)
    ●
    Free service at https://mybinder.org/
    Combine Docker & JupyterHub, for…
    ●
    Repeatable notebook execution from git
    ●
    Ephemeral custom Jupyter runtimes
    ●
    Builds & spawns ad-hoc kernels (repo2docker)
    ●
    Free service at https://mybinder.org/
    

    View Slide

20. Deploying
    Deploying
    a Local
    a Local
    Data Science
    Data Science
    Platform
    Platform
    NASA's Juno Mission: Infrared Tour of Jupiter's North Pole
    NASA's Juno Mission: Infrared Tour of Jupiter's North Pole
    

    View Slide

21. JupyterHub Deployment Options
    JupyterHub Deployment Options
    ●
    Debian Package (1and1/debianized-jupyterhub)
    ●
    The Littlest JupyterHub (TLJH · official)
    ●
    Zero to JupyterHub with Kubernetes (official)
    ●
    And lots of 3rd party projects
    ●
    Debian Package (1and1/debianized-jupyterhub)
    ●
    The Littlest JupyterHub (TLJH · official)
    ●
    Zero to JupyterHub with Kubernetes (official)
    ●
    And lots of 3rd party projects
    

    View Slide

22. Showcase: DevOps Intelligence
    Showcase: DevOps Intelligence
    ●
    Optimize dev + ops processes
    ●
    Generate actionable insight
    ●
    Support risk analysis & decisions
    ●
    Typical use-cases:
    – Migration processes of all kinds
    (current state, progress tracking, achievement of objectives)
    – Inventory reporting for increased transparency
    – Automate internal reporting, liberating scarce human expertise
    ●
    Optimize dev + ops processes
    ●
    Generate actionable insight
    ●
    Support risk analysis & decisions
    ●
    Typical use-cases:
    – Migration processes of all kinds
    (current state, progress tracking, achievement of objectives)
    – Inventory reporting for increased transparency
    – Automate internal reporting, liberating scarce human expertise
    https://blog.jupyter.org/devops-intelligence-3ff48a76b525
    https://blog.jupyter.org/devops-intelligence-3ff48a76b525
    

    View Slide

23. DevOps Intelligence Platform:
    DevOps Intelligence Platform:
    Simple Single-Host JupyterHub Deployment
    Simple Single-Host JupyterHub Deployment
    

    View Slide

24. JupyterHub Debian Package
    JupyterHub Debian Package
    ●
    JupyterHub‘s “Installation Guide” as software
    ●
    Turn-key setup (just add Python, NodeJS, and Chromium headless)
    ●
    Debian packaging of all core components and standard dependencies
    ●
    JupyterHub, notebook server, configurable HTTP proxy (CHP)
    ●
    JupyterLab, PySpark, NumPy, SciPy, Pandas, Matplotlib, Seaborn, HoloViews, …
    ●
    Best used on Debian Stretch / Ubuntu Bionic
    ●
    Configured for PAM authorization and sudo spawner by default
    ●
    Systemd process control / NginX SSL off-loader
    https://github.com/1and1/debianized-jupyterhub
    ●
    JupyterHub‘s “Installation Guide” as software
    ●
    Turn-key setup (just add Python, NodeJS, and Chromium headless)
    ●
    Debian packaging of all core components and standard dependencies
    ●
    JupyterHub, notebook server, configurable HTTP proxy (CHP)
    ●
    JupyterLab, PySpark, NumPy, SciPy, Pandas, Matplotlib, Seaborn, HoloViews, …
    ●
    Best used on Debian Stretch / Ubuntu Bionic
    ●
    Configured for PAM authorization and sudo spawner by default
    ●
    Systemd process control / NginX SSL off-loader
    https://github.com/1and1/debianized-jupyterhub
    

    View Slide

25. Just 617 MB ☺
    Just 617 MB ☺
    

    View Slide

26. K8s Setup from 10 Miles High
    K8s Setup from 10 Miles High
    

    View Slide

27. Bloomberg‘s Architecture (Kerberos)
    Bloomberg‘s Architecture (Kerberos)
    

    View Slide

28. Bloomberg‘s Architecture (Docker)
    Bloomberg‘s Architecture (Docker)
    

    View Slide

29. View Slide

30. The “nbconvert” Tool
    The “nbconvert” Tool
    ●
    CLI tool: jupyter nbconvert …
    ●
    https://github.com/jupyter/nbconvert
    ●
    Convert into…
    ●
    HTML · PDF · Markdown · ReST · Script · and more
    ●
    Basis for other tools & most simple workflow:
    ●
    Convert notebook to HTML page
    ●
    Upload HTML file to webserver (e.g. Artifactory)
    ●
    CLI tool: jupyter nbconvert …
    ●
    https://github.com/jupyter/nbconvert
    ●
    Convert into…
    ●
    HTML · PDF · Markdown · ReST · Script · and more
    ●
    Basis for other tools & most simple workflow:
    ●
    Convert notebook to HTML page
    ●
    Upload HTML file to webserver (e.g. Artifactory)
    

    View Slide

31. nbreport –
    nbreport – Clean
    Clean Notebook HTML Rendering
    Notebook HTML Rendering
    ●
    CLI tool or notebook extension
    ●
    Download cleaned-up single HTML page
    ●
    Remove technical ornaments
    ●
    Empty or explicitly hidden cells, hidden code
    ●
    Input / output counters, and stderr
    ●
    Add header information (author & title)
    ●
    CLI tool or notebook extension
    ●
    Download cleaned-up single HTML page
    ●
    Remove technical ornaments
    ●
    Empty or explicitly hidden cells, hidden code
    ●
    Input / output counters, and stderr
    ●
    Add header information (author & title)
    

    View Slide

32. Pandoc – Swiss Army Knife for Publishing
    Pandoc – Swiss Army Knife for Publishing
    ●
    Convert from / to many document formats
    ●
    Part of many rendering pipelines
    ●
    Needed to unlock all nbconvert features
    (together with TeX)
    ●
    Convert from / to many document formats
    ●
    Part of many rendering pipelines
    ●
    Needed to unlock all nbconvert features
    (together with TeX)
    https://pandoc.org/
    https://pandoc.org/
    

    View Slide

33. Publishing Notebooks to Atlassian Confluence
    Publishing Notebooks to Atlassian Confluence
    https://github.com/Valassis-Digital-Media/nbconflux
    https://github.com/Valassis-Digital-Media/nbconflux
    

    View Slide

34. nbviewer – A simple way to share notebooks
    nbviewer – A simple way to share notebooks
    ●
    Render notebooks from git repositories
    ●
    Link back to repo, and to Binder (live notebook)
    ●
    Does not execute the notebook
    commit pre-rendered output cells
    ⇒
    e.g. https://nbviewer.jupyter.org/github/jhermann/jupyter-by-example/tree/master/how-tos/
    ●
    Render notebooks from git repositories
    ●
    Link back to repo, and to Binder (live notebook)
    ●
    Does not execute the notebook
    commit pre-rendered output cells
    ⇒
    e.g. https://nbviewer.jupyter.org/github/jhermann/jupyter-by-example/tree/master/how-tos/
    

    View Slide

35. nbgallery – Enterprise Sharing / Collaboration Platform
    nbgallery – Enterprise Sharing / Collaboration Platform
    ●
    https://github.com/nbgallery/nbgallery
    ●
    Using Jupyter to Empower
    Enterprise Analysts
    https://youtu.be/9qS1U-ySwzE
    ●
    RoR web application
    ●
    MySQL / MariaDB
    ●
    Apache Solr indexing
    ●
    https://github.com/nbgallery/nbgallery
    ●
    Using Jupyter to Empower
    Enterprise Analysts
    https://youtu.be/9qS1U-ySwzE
    ●
    RoR web application
    ●
    MySQL / MariaDB
    ●
    Apache Solr indexing
    

    View Slide

36. Automation:
    Automation: Papermill by
    Papermill by nteract
    nteract
    Parameterize, execute
    & analyze notebooks
    Parameterize, execute
    & analyze notebooks
    https://medium.com/netflix-techblog/notebook-innovation-591ee3221233
    https://medium.com/netflix-techblog/notebook-innovation-591ee3221233
    

    View Slide

37. Automation:
    Automation:
    Paperboy by Tim Paine
    Paperboy by Tim Paine
    Web UI for scheduling
    notebook reports
    ●
    Runtime: Papermill
    ●
    Scheduling: Airflow
    ●
    Persistence: SQLAlchemy
    Web UI for scheduling
    notebook reports
    ●
    Runtime: Papermill
    ●
    Scheduling: Airflow
    ●
    Persistence: SQLAlchemy
    

    View Slide

38. Even More Tools…
    Even More Tools…
    ●
    RISE – ‘Live’ reveal.js Jupyter / IPython Slideshow Extension
    https://github.com/damianavila/RISE
    ●
    QuantStack ‘Voila’ – Interactive renderer for Jupyter notebooks
    https://github.com/QuantStack/voila
    ●
    Anaconda‘s “Exploring Data using Python Visualization”
    https://anaconda.org/jbednar/exploring_data/notebook
    ●
    PyViz – Make data visualization easier to use & learn, and more powerful
    https://pyviz.org/
    ●
    Knitty – Pandoc filter and Atom-friendly reports via Jupyter
    https://github.com/kiwi0fruit/knitty
    ●
    nbdime · nbstripout · jupytext · …
    ●
    RISE – ‘Live’ reveal.js Jupyter / IPython Slideshow Extension
    https://github.com/damianavila/RISE
    ●
    QuantStack ‘Voila’ – Interactive renderer for Jupyter notebooks
    https://github.com/QuantStack/voila
    ●
    Anaconda‘s “Exploring Data using Python Visualization”
    https://anaconda.org/jbednar/exploring_data/notebook
    ●
    PyViz – Make data visualization easier to use & learn, and more powerful
    https://pyviz.org/
    ●
    Knitty – Pandoc filter and Atom-friendly reports via Jupyter
    https://github.com/kiwi0fruit/knitty
    ●
    nbdime · nbstripout · jupytext · …
    

    View Slide

39. View Slide

40. References
    References
    ●
    Project Jupyter Homepage
    https://jupyter.org/
    ●
    JupyterHub Homepage
    https://jupyterhub.readthedocs.io/en/stable/
    ●
    Jupyter Community Channels
    https://jupyter.rtfd.io/en/latest/community/content-community.html#jupyter-communications
    ●
    Jupyter learning resources and practical tips
    https://github.com/jhermann/jupyter-by-example
    ●
    Project Jupyter Homepage
    https://jupyter.org/
    ●
    JupyterHub Homepage
    https://jupyterhub.readthedocs.io/en/stable/
    ●
    Jupyter Community Channels
    https://jupyter.rtfd.io/en/latest/community/content-community.html#jupyter-communications
    ●
    Jupyter learning resources and practical tips
    https://github.com/jhermann/jupyter-by-example
    

    View Slide

41. Acknowledgements
    Acknowledgements
    ●
    https://jupyter.readthedocs.io/
    ●
    https://dailyhealthpoints.com/2016/11/23/the-importance-of-setting-a-training-goal/
    ●
    https://commons.wikimedia.org/wiki/File:Porsche_911_(997)_GT3_RS_3.6_-_capot_arri%C3%A8re_ouvert_2.jpg
    ●
    https://commons.wikimedia.org/wiki/File:Porsche_911_(997)_GT3_RS_3.6_-_d%C3%A9tail_capot_avant.jpg
    ●
    https://www.jpl.nasa.gov/news/news.php?feature=7096
    ●
    https://www.slideshare.net/SparkSummit/secured-kerberosbased-spark-notebook-for-data-science-spark-summit-east-talk-by-j
    oy-chakraborty
    ●
    http://www.picpedia.org/highway-signs/p/publish.html
    ●
    https://commons.wikimedia.org/wiki/File:Emojione_1F44D.svg
    ●
    https://jupyter.readthedocs.io/
    ●
    https://dailyhealthpoints.com/2016/11/23/the-importance-of-setting-a-training-goal/
    ●
    https://commons.wikimedia.org/wiki/File:Porsche_911_(997)_GT3_RS_3.6_-_capot_arri%C3%A8re_ouvert_2.jpg
    ●
    https://commons.wikimedia.org/wiki/File:Porsche_911_(997)_GT3_RS_3.6_-_d%C3%A9tail_capot_avant.jpg
    ●
    https://www.jpl.nasa.gov/news/news.php?feature=7096
    ●
    https://www.slideshare.net/SparkSummit/secured-kerberosbased-spark-notebook-for-data-science-spark-summit-east-talk-by-j
    oy-chakraborty
    ●
    http://www.picpedia.org/highway-signs/p/publish.html
    ●
    https://commons.wikimedia.org/wiki/File:Emojione_1F44D.svg
    

    View Slide

42. Questions?
    Questions?
    Thank you!
    Thank you!
    

    View Slide