You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@jlewi sort of. Looks like we were dually supporting py3 and py2 by installing and activating py2 in the jovyan homedir, which of course gets wiped in prep for the PVC check. Does that track?
Which means I need to preserve that as well. 😁
/assign @pdmack
@jlewi The theory presented by @pdmack sounds right. If the user directory is being wiped out after installing the ipykernel, then it will not be available.
It appears we don't need to install to jovyan at all, but instead the new global env location /opt/conda. But I'm going to review the older images and see what (if any) differences there may be.
I just tried gcr.io/kubeflow-images-public/tensorflow-1.7.0-notebook-cpu:v20180601-2464851a
And I didn't see an option for a python2 runtime which is needed for TFX/Beam support.
My guess is this was a regression introduced when refactored the notebook image to support mounting all of /home/jovyan on a PVC.
/cc @pdmack
The text was updated successfully, but these errors were encountered: