wildtreetech / ohjh Goto Github PK

We currently assign 10Gig per home directory per user. Do they need this much? More? Less?

There are two things that can be updated:

• kubernetes itself; and
• JupyterHub

To upgrade JupyterHub edit ohjh/requirements.yaml replacing the version of the jupyerhub chart with the one you want to use from https://jupyterhub.github.io/helm-chart/ Probably the latest development build is a good idea.

Sometimes this happens when people try to login. Seems to be hard to reproduce. Some kind of race condition or trying to login twice at the same time maybe.

Min says that the cookie is set before the redirect and then deleted after successful login. This means if two logins are happening at the same time you could end up with the second one failing.

Move the crypt key for auth state to the dedicated place in config.yaml

https://github.com/jupyterhub/zero-to-jupyterhub-k8s/blob/master/jupyterhub/values.yaml#L112

After this we might have a clear separation between secret config values and public config values.

Follow up jupyterhub/zero-to-jupyterhub-k8s#440 about moving client_id and friends to the secrets part of k8s instead of a ConfigMap.

Add OpenHumans to this list: https://zero-to-jupyterhub.readthedocs.io/en/latest/users-list.html

Install textblob and emoji from PyPI in the single user image

When launching a Terminal from the Notebook interface

https://notebooks.openhumans.org/user/{user}/terminals/1

groups: cannot find name for group ID 1000
jovyan@jupyter-{user}:~$ ls
lost+found
jovyan@jupyter-{user}:/$ groups
users groups: cannot find name for group ID 1000
1000

Likely just needs to be defined in your docker's /etc/group

For mybinder.org auto-scaling the cluster up works well. More machines get spawned without any of the "admins" noticing. What still requires manual intervention is scaling the cluster back down. See this comment for the proposal on how to make it easier for the AS to scale back down (use strategy "pack").

Manual scale down isn't difficult but yet another thing to think off.

Todo: watch how switching to "pack" works out for mybinder.org. Start with a small cluster known to a few OH users to get a feeling for the usage pattern before investing a lot of time in any particular setup (AS vs fully manual vs X).

This setup suffers from the same "excessive" logging as mybinder.org (see jupyterhub/mybinder.org-deploy/#309). Apply same exclusions to the stackdriver ingestion.

Reduce logging from CHP by reducing the log level to "ERROR" as done in jupyterhub/mybinder.org-deploy#356

So that we can restart the hub from the web! 👍

Currently the only way to renew your token is to:

1. stop server
2. logout
3. login
4. start server

Pretty tedious. We should install a server extension that refreshes the token periodically so that it is always valid.

1. Setup Jupyterhub
2. ...
3. World domination!

We want to have some custom libraries and tools installed in the single user image.

Things we need for sure:

• server extension (#3)
• OH API library
• R kernel (and R studio via nbrsessionproxy?)
• httpimport
• myvariant

More ideas?