JupyterHub and BinderHub Team Meeting - December¶
Date: 20 December 2018 at 6pm Zurich time¶
Welcome to the Team Meeting¶
If you are joining the team video meeting, sign in below so we know who was here. Roll call:
name / institution / GitHub handle
Tim / Wild Tree Tech / @betatim
Min / Simula / @minrk
Erik / Sandvik CODE / @consideRatio
Zach / Cal Poly / @Zsailer
Carol / Project Jupyter / @willingc
James / UW-Madison / @jrbourbeau
Joe / NCAR / @jhamman
Brian / Cal Poly / @ellisonbg
Kacper / Whole Tale / @xarthisius
Yuvi / Berkeley? / @yuvipanda
Lindsey / Berkeley / @lheagy
Craig / Whole Tale / @craig-willis
Jessica / @jzf2101
Murali / Capital One/
Chris H / BIDS / @choldgraf
name / affiliation / github
Should Binder seek out a next partner to grow adoption?¶
(silent writing exercise while people arrive)
If you think yes, who should we start reaching out to? * if there is interest we can create an agenda point around this at next months meeting
a conference to which a large fraction of submissions is “code”, use mybinder.org as part of the submission
teaching programme? Help teachers build classes or education that is otherwise hard to do because people don’t have access to computing
Erik connected to students in Gothenburg University, will have a talk about Jupyter in February for them, 20-60 people. My brother is an “Docent” there and utilizes Jupyter somewhat already.
(Zach) At University of Oregon, a few professors use Binder repos to teach chemistry/biochemistry concepts using interactive widgets (example). Advertising this use case at universities might generate some traffic.
(Yuvi) Working on getting a grant from Space Telescope Institute to pay for my time at Berkeley, involving unspecified Binder work
Conferences that use binder’s underlying tech (eg Kubernetes or dev ops people)
Let’s collect all potential agenda items here. We will then sort them at the start of the meeting. If there are similar items try and group them already.
What should we do with github issues that are support requests? (e.g., https://github.com/jupyterhub/zero-to-jupyterhub-k8s/issues/1006) [CH]
people have questions about technical issues that can be out-of-scope of the project directly
currently the issues remain open forever
What should we do with them?
GitHub for bug reports and feature requests only
questions and how-do-i-do… are directed to discourse/Stack Overflow
have a bot that replies for us directing people at discourse
TODO: Open an issue to discuss providing either 1. instructions in the issue template for what kinds of issues should go here, 2. a bot that jumps in on new issues and posts a friendly message about what kinds of issues are appropriate.
general agreement that discourse is a good place for more general tech support so we can start people there.
Any news about the OVH deployment?
OVH: cloud deployment in Europe
Google has agreed to pay our cloud bill by giving us a lump sum to spend
repo2docker extensibility. how can we make it possible for users to define “patterns” that they can share with others without forking repo2docker. e.g., for NeurIPS we set up a CUDA-based cluster and it’d be great to make it easy for people to “enable CUDA” without requiring a Dockerfile. Another example are more complex but common setups like Latex etc.
We can already do this with traitlets, if there is an easy driving use case we can do a sprint to make it happen (Yuvi)
can a repo request a extension to repo2docker be enabled/used?
Q: What if we used entrypoints as a potential extension mechanism? A: Yuvi thinks this would be a good option.
combination of entrypoints and traitlets sound like a good avenue to explore
TODO: Find a use-case to motivate extensibility on repo2docker and put together a prototype to test out different approaches.
rocker related question: does repo2docker want to support running images outside of BinderHub?
RStudio not through the route of Jupyter is more familiar to RStudio users
already possible, however needs more exposure/documentation so people know how to do this
TODO: let’s find some use-cases that we can use to get a good idea what are requirements and trade-offs
TODO: Write up the current design / structure of the proxying that we use for RStudio support. Maybe a little blog post + feed that into the docs afterward.
mybinder.org at NeurIPS2018
success! Thanks to Jess for getting us there. Thanks to everyone who helped make it happen!
MILA is very interested @jzf2101 and Felix-Antoine will follow up since I think they may be interested in putting up a Binderhub or JHub
report from the demo day:
generally people seemed interested and encouraging
some groups now want to setup their own bhub and jhub
surprisingly large number of industry people attendingngngngngng
blog post: https://hackmd.io/ZEVbmiWHQiOuRUQzRmjnNg#
publish date: XXX
Tim would like to discuss how we can make these kinds of events less stressfull
when we use the Jupyter and/or Binder name we have a big reputation to keep up with and are representing everyone who puts their name and effort to that name.
having to finish things last minute leads to mistakes and burnout
for example “any custom deployment has to be demo’ed in ‘ready to go setup’ five days prior to the event”??
TODO: create team-compass issue to talk about this and have a more organised discussion at the next team meeting
binder.pangeo.io at AGU2018 (Joe)
Just sharing slides: https://doi.org/10.6084/m9.figshare.7492661.v1
And 4hr tutorial: https://github.com/pangeo-data/pangeo-tutorial-agu-2018
Zach has been exploring implementations
Chris brought up question of looking at design requirements and protocol again
What is the best way for that to move forward
Zach has some time allocation to work on this in Dec
Zach has been working on a document to describe usage cases and design requirements
Maybe look at https://github.com/nteract/bookstore for some ideas. It’s early stage but may solve some of your issues.
Could we use nbviewer to create a gallery of reproducible papers w/ Binder?
what should we discuss here?
Outreachy 🎉 !!!
traefik-proxy project going great! It already works with jupyterhub in local testing. https://github.com/jupyterhub/traefik-proxy
jupyterhub-nativeauthenticator project going good. Outreachy intern Leticia is moving continents this week, so off for this and next weeks
TODO: We should invite them to this meeting!
TODO: we should help them speak at (non-jupyter-specific) conferences about their work
releasing z2jh 0.8 https://github.com/jupyterhub/zero-to-jupyterhub-k8s/issues/1054
should be only docs and testing (and related fixes) remaining before the release
releasing jupyterhub 1.0
need to finish auth expiry https://github.com/jupyterhub/jupyterhub/pull/2342
probably pushing oauth scopes to 2.0
repo2docker roadmap, only items that need to be discussed
repo2docker.version issue (Craig/Kacper)
FYI: Erik is thinking about pushing for a national JH for Swedens educational system. Step 1: get universities starting to use a Z2JH.
I am seeing/hearing interest in JHub variants that are backed by a kernel service and have a persistent UI (not neccesarily containerized). We may want to brainstorm on JHub architecture changes needed for that (Brian).
Moore Foundation roadmap / wishlist document
Meeting with Chris from Moore foundation about new Binder grant
Asked to write up a document with ideas on what we might do, so we can go from there
https://docs.google.com/document/d/1cyliT2G1e1j81fQEKZUQ2IKb1g_J91g9zF0QowN0hcA/edit#heading=h.gg6hz2kdjmoc is the document, so please add your crazy & non-crazy ideas
Extended binderhub user API for resource requests and federation (Joe)
need (want) a way to specify binder cpu/memory requirements and/or other resources (gpu)
need a way to specify where binders will run (GCP, AWS, etc.), so we’re thinking about how to federate resources
planning work on this for early-2019
related but differnt: limit resources by user launching the binder instead of the creator of the binder
Erik had a idea, that wasn’t good: we can always let them define whatever, and then restrict it using kubernetes mutating webhooks or something like this… Thinking further, it may be better… yes… to do it before it goes that far into the deployment…
TODO: create a use-case and flesh it out to find out where this should be at home
read the previous discussion on the original Binder repo: https://github.com/jupyter/enhancement-proposals/pull/5
Things people should know about
Outreachy interns have started (Welcome interns!)
repo2docker 0.7.0 is out!
repo2docker blog post is out!