JupyterHub HPC Meeting - April 2021#
Time: 8:30 AM PST
Calendar for future meetings: https://jupyterhub-team-compass.readthedocs.io/en/latest/meetings.html
Welcome to the Meeting#
Hello! If you are joining the team video meeting, sign in below so we know who was here. Roll call:
name / institution / GitHub handle
Rollin / NERSC / @rcthomas
Zach Price / ORNL
Michael Milligan / UMN MSI
Jens Henrik Goebbert / FZJ
60 second updates on things you have been up to, questions you have, or developments you think people should know about. This is also a chance to suggest a future presentation if you’ve got work currently in progress you might want to share. Please add yourself, and if you do not have an update to share, you can pass.
NAME: What you’d like to update on
Rollin Jupyter Community Workshop series is coming back to life, “plans for plans” for security-focused workshop will be coming together soon.
Reports and celebrations#
This is a place to make announcements (without a need for discussion). This is also a great place to give shout-outs to contributors! We’ll read through these at the beginning of the meeting.
Let’s collect all potential agenda items here before the start of the meeting. We will then attempt to create a coherent agenda that fits in the 60m meeting slot. If there are similar items try and group them together.
Batchspawner 1.1 released
On PyPI, open for business for new PRs!
Acceptance/acquiescence on reformatting
Now outdated needs to be brought up to date
Ask Erik to update the PR then MM will merge
If there was a push to upstream something:
Remote port stuff would really be nice to have upstreamed because multiple projects need it
Port range implementation is also a feature people have been asking for
Certain issues with traitlets
Code needs a pass to catch up to modern
ProfileSpawner prevents using internal SSL
Release in current state pretty soon
But after that, want to deep dive into some issues
Kubespawner - cross cluster? (Zach Price):
ARM (ORNL group) has access to 2 HPC clusters, one at OLCF, one not
JupyterHub that allows launching on HPC or not (2 profiles)
OLCF also has a new hub deployment they’d like to use but trying to avoid users having to know the differences between the 2
So one hub in one cluster wants to start notebooks in another cluster, is there a way to do that?
Single hub at a big instution, main problem is a common identity management system
Jens asked how people handle MFA:
Basically everyone does their own thing
rkdarst: help editing to get it in?
Startup time when launching a new JupyterLab (Jens Henrik)
They just switched from home-grown Docker to k8s
Performance testing (why 15-20 s to launch?)
Partially parallel file system
Any nice solutions?
Jens: Turn off checking build update! 7 seconds!
Richard observed some things with Jupyter Docker stacks with chown’ing, but Jens is doing bare metal GPFS
Login node setup is over GPFS
Compute node setup is over Cray DVS, there the filesystem is mounted read-only w/client-side caching on
Where is the time going and what JupyterLab is this?
Lots of requests to NFS
Median start-up time at MSI looks like 20-40s
Richard observed 22s at Aalto during call
What kind of overhead does a non-Jupyter pod get? Is this Jupyter specific? Start a pod and run mdtest :)