6

I created a deep learning instance inside the AI platform of google cloud . I use the built in jupyterlab notebooks running on the browser (I use chrome). Recently I have a problem with saving the code. autosave as well as saving the notebook files does not work. I keep see the message "saving started" when I try to save, but nothing other than that happens, and the code is not saved. I tried restart the kernel as well as restart the instance but the problem keeps returning. Anyone here encountered the same thing? have a solution?

thanks

Rachel Shalom
  • 399
  • 2
  • 10
  • I just got a file save error message 504. any idea why that is happening? – Rachel Shalom Jun 15 '20 at 11:01
  • What version (Base image) of AI Platform Notebooks are you using (TF 1.15, 2, Pytorch? I can try to replicate it – gogasca Jun 15 '20 at 19:29
  • Sounds like the kernel isn't being found. Do you have a live or dead kernel? If you have a missing / unfound kernel, then you probably have a "version" issue - meaning different versions of the kernel used when the document was created, compared to the version of the kernel that is attempting to open the file. – Gray Jun 16 '20 at 23:16
  • @gogasca I use an image with pytorch 1.4 – Rachel Shalom Jun 17 '20 at 13:34
  • @Gray. I see the kernel is idle or connected when it happens. what do you think should help when there is versioning problem? restarting didn't work – Rachel Shalom Jun 17 '20 at 13:40
  • Assuming that Anaconda is not involved and causing untraceable issues, I can share a similar problem. I hate Anaconda, but that's my issue. I had a similar problem with Jupyter that was caused by a small change I manually made in the kernel's config file. I changed the display name to R3.5 from the default display, R. This change worked well and did not cause a problem until I upgraded the version of R to 3.6. The new 3.6 R kernel, could display existing Jupyter Notebook files, but nothing else. A dead kernel. FYI, a kernel's name gets registered, and you can re-register names. – Gray Jun 18 '20 at 06:25

2 Answers2

3

What worked for me eventually was instead of using https connection I connected to the instance via ssh and then to accessed jupyterlab in local host. I followed this link: https://cloud.google.com/ai-platform/deep-learning-vm/docs/jupyter

export PROJECT_ID="my-project-id"
export ZONE="my-zone"
export INSTANCE_NAME="my-instance"
gcloud compute ssh --project $PROJECT_ID --zone $ZONE \
$INSTANCE_NAME -- -L 8080:localhost:8080
Rachel Shalom
  • 399
  • 2
  • 10
0

I had a similar problem, I opened a notebook and GCP did not save it, but after I started the notebook from a folder inside Jupyterlab - it saved it.

AvivLevi815
  • 37
  • 11