Are notebook saves not Atomic in jupyterhub?

JupyterHub Zero to JupyterHub on Kubernetes
1

Hello,
I have been using jupyterhub for some time now and recently i faced an error saying
File Load Error for notebook.ipynb

Unreadable Notebook: /path/notebook.ipynb NotJSONError('Notebook does not appear to be JSON: \'{\\n "cells": [\\n {\\n "cell_type": "m...')

I fixed this once by updating the notebook json by removing one of the metadata section (since there should only be one metadata per notebook but there were two)
“metadata”: {“kernelspec”: {…}}
along with updating some braces. This fixed the issue but i am encountering this again, this time null was appended at the very end of the file and was wondering it is due to some failure during saving? also shouldnt the save be atomic to being with ? Can someone please help me with this.
Few other things :

  1. I have auto-save enabled for every 10 seconds
  2. The notebook size is around 100mb since it have output of some analysis.

Thanks

2

One low likelihood possibility is aging storage, as I found with one of my jupyter hub installations. See this thread Intermittent problems rendering saved widget states

1 Like