Zero-to-jupyterhub-k8s: Cannot use named servers feature

[ablekh]

Just upgraded our pilot cluster to the latest dev. version of the Helm chart (to test named servers feature), however I don't see relevant UI appearing as expected. The redacted config.yaml follows. Thank you!

hub:
  cookieSecret: <SECRET>
  extraConfig: |
    c.KubeSpawner.singleuser_image_pull_secrets = "sbdh-acr-secret"
    c.JupyterHub.allow_named_servers = True

proxy:
  secretToken: <SECRET>
  type: ClusterIP
  https:
    enabled: true
    type: letsencrypt
    letsencrypt:
      contactEmail: <EMAIL>
    hosts:
      - <FQDN>

prePuller:
  continuous:
    enabled: true

imagePullSecrets:
  - name: sbdh-acr-secret-contrib

rbac:
  enabled: true

auth:
  admin:
    users:
      - ablekh
  type: github
  github:
    clientId: <CLIENT_ID>
    clientSecret: <CLIENT_SECRET>
    callbackUrl: "https://<FQDN>/hub/oauth_callback"

singleuser:
  profileList:
    - display_name: "Minimal Jupyter environment"
      description: "Provides access to Python kernel only."
      default: true
    - display_name: "Data science environment"
      description: "Provides access to Python, R and Julia kernels."
      kubespawner_override:
        image: jupyter/datascience-notebook:6fb3eca57bd3
    - display_name: "Custom SBDH data science environment - SBDH-DS-O"
      description: "Provides access to Python, R, Julia and Octave kernels."
      kubespawner_override:
        image: <PRIVATE_CONTAINER_REPO>/datascience-octave-notebook:v3

[consideRatio]

I lack too much knowledge and understanding about named servers sadly =/

Off topic details:

• I have noticed there is a configuration option, hub.allowNamedServers that you can set to true in order to accomplish the same thing as doing it from the extraConfig and c.JupyterHub.allow_named_servers = True.
• c.KubeSpawner.singleuser_image_pull_secrets should be written without the singleuser_ in order to not use a deprecated solution.
• No matter what, you could also use the Helm chart's automatic generation of a k8s Secret that is then automatically referenced by the spawned user pods as a image pull secret. This would be done using the following configuration option: http://z2jh.jupyter.org/en/latest/reference.html#singleuser-imagepullsecret
• The current imagePullSecrets entry, in the helm chart values you provide, will not be interpreted by the Helm chart.

[ablekh]

@consideRatio No problem - I hope others will chime in soon. Thank you for your helpful insights. :-)

[minrk]

named server UI is new in jupyterhub 1.0, which wasn't in the chart until yesterday. Using chart 0.9-8ed2f81, you can enable named servers with:

hub:
  allowNamedServers: true

[clkao]

There should be some docs about this, particularly it won't work if user home pvc is enabled and it is not RWX. and even if it's RWX, multiple jupyterlab instances potentially opening same notebook is going to be funny.

[ablekh]

@minrk Thank you for sharing that info.

@clkao Thank you for letting me know. Should you run across relevant docs, please share with me. Regarding your multi-user access concerns: my setup is that I mount a shared volume (Azure Files) as a directory in users' home directories (/home/jovyan); the shared volume is supposed to be for read access for the most part, even though it's configured as RWX. Any thoughts?

[Edward-liang]

Excuse me, I noticed the named-server feature in jupyterhub's RestAPI doc, in which it says :

With the named-server functionality, it’s now possible to launch more than one specifically named servers against a given user. This could be used, for instance, to launch each server based on a different image.

And I noticed that if I have several profiles in my config.yaml,
a post request like following would lead to booting of the first profile image.

curl -X POST -H "Authorization: token <token>" http://ip:port/hub/api/users/jupyter2/server

And I am wondering is there anyway to start the other images provided via profiles, please?

[ablekh]

@Edward-liang Sorry, I don't (yet) have an experience launching single-user servers (named or not) via JupyterHub's REST API. Do you really need to manually do that (perhaps, for some automation)?

In any case, please see @minrk's comment above and verify that your software components' versions match relevant dependencies.

[Edward-liang]

@ablekh Yes, I want to take launching progress over because of some automation requirement. Thank you for your advice.

[ablekh]

@Edward-liang You are welcome. Please keep us posted on your progress.

[consideRatio]

This issue seems resolved by https://github.com/jupyterhub/zero-to-jupyterhub-k8s/issues/1186#issuecomment-477648626, closing!