Jupyter Notebook not saving: '_xsrf' argument missing from post

Question

I've been running a script on jupyter notebooks for about 26 hour; I haven't really been using my computer for anything else, but it needs to run this program that will take ~30 hours to complete. At about 21 hours in, it stopped saving and my terminal had this:

403 PUT /api/contents/[file.ipynb] (::1): '_xsrf' argument missing from POST

where [file.ipynb] is the location of my jupyter notebook. It also says:

'_xsrf' argument missing from post

in the top right part of the notebook again. The program is still running and I don't want to restart jupyter notebook and have to run the program again, as I have a deadline, is there anything else I can do?

I'm using google chrome, but I don't have the LastPass extension or any '%' characters in my code, as another post suggested.

Thanks for any help!

• AlexK: There are discussions online on this topic if you do a search. One suggestion that I've seen made is to open the same notebook in a new browser window, which fixes the problem. Then you can close that new window and continue working in the previous one.
• RandomWalker: Thanks @AlexK, that solves the same issue on my MBP.
• tbrugere: Just a tip, in this kind of case, store your important variables using %store magic. That way even if your notebook was closed and not saved, your variables still are, so you can just display them again.
• Bill: @AlexK What about the code though? Will any changes to the code be preserved if I open in a new window?
• Bill: Update on my previous comment: @AlexK's suggestion works and afterwards you can save the original notebook and any code changes made. Thanks.

Answer 1

The easiest way I found is this:

https://github.com/nteract/hydrogen/issues/922#issuecomment-405456346

Just open another (non-running, existing) notebook on the same kernel, and the issue is magically gone; you can again save the notebooks that were previously showing the _xsrf error.

If you have already closed the Jupyter home page, you can find a link to it on the terminal from which Jupyter was started.

• Hari Dattada: This saved my day. 4 days worth of work would have been lost, if I hadnt come across this solution. Thank you!
• user650654: I did not need to open a new notebook. Instead, I reopened the tree, and reconnected the kernel. At some point I also restarted the kernel.
• Pavindu: I wasn't even allowed to open a new notebook until I refresh the notebooks list. After that I opened a new notebook and the error was disappeared..😄
• mirekphd: Notice that the workaround requires that the notebook was created earlier, so open another existing non-running notebook, do not create a new one or open a running one, both of which would raise the same error.
• Albert Chen: This is why I love stackoverflow

Answer 2

The solution I came across seems too simple but it worked. Go to the /tree aka Jupyter home page and refresh the browser. Worked.

• DISC-O: This worked for me whereas the solution to open another notebook failed in my case.
• eric: Thank goodness I can see which posts I've upvoted so I know which solution to use when I forget and come back in a year.
• Verena Praher: The accepted answer did not work for me, but this did (although I am using Jupyter Lab in case this matters). I didn't dare to refresh my Jupyter Lab (running at mydomain.sth/lab?), because it might have lost my changes, so I opened mydomain.sth/tree as suggested here, refreshed and it worked. Thanks!
• yellowjacket05: The accepted answer did not work for me, but this did. I am not using Jupyter Lab but Jupyter Notebook with a conda-based kernel.
• doraemon: The accepted answer did not work for me, but this did

Answer 3

The only solution worked for me was:

1. I opened a new tab in chrome
2. I pasted : http://localhost:8888/?token=......
3. then I went to my original notebook and I was able to save it

• avriiil: This worked for me, none of the others did.
• Fabian: Worked for me too (also none of the others did) and just localhost:8888 without the token was sufficient for me :)
• Dustin Butler: This worked for me too. I did right click, duplicate tab. Closed the duplicate tab and was able to save on working tab.

Answer 4

You can disable the XSRF checks by launching e.g. JupyterLab with:

jupyter lab --ServerApp.disable_check_xsrf=True

This probably means that your server is not token-authenticated. For instance, you may have launched JupyterLab with:

jupyter lab --NotebookApp.token='' --NotebookApp.password=''

And it also probably means that you are running JupyterLab on you local machine and your server is not accessible remotely, so this is fine.

But in general beware that if your server doesn't have authentication, it is vulnerable to XSRF.

• Max Barrass: This will also help if you do not have token auth set and trying to shutdown jupyter from command line.

Answer 5

I use jupyter notebooks daily and had never experienced this issue before... until today. I had the notebook open all day but it wasn't running anything and then for no apparent reason stopped auto-saving with the '_xsrf' argument missing from POST error message in the top right. FYI - this is a python3 notebook.

I don't know the cause of this problem but I have recently upgraded my python3 version to 3.7.2 and upgraded all of my site-packages to their latest version as of a few days ago which could possibly be the cause.

As for a solution, as suggested in the comment by @AlexK, I opened the same notebook in a new window (different browser in fact), using

jupyter notebook list

in the terminal to get the URL with login token.

This resulted in me having the notebook open and savable again but the information I had entered since the last successful auto-save was missing. Thankfully, my broken instance was still open and working apart from saving so I was able to simply copy and paste the information across then hit save. So, keep the broken instance open if you try this!

• DataOrc: After you open the second tab, you should be able to go back to the first and save, so you don't even have to copy and paste to the new one. At least, that's how it worked out for me.

Answer 6

Simply refreshing the tree tab worked for me.

• Imtinan Azhar: This really helped out, Thanks a lot! Just a slight addition, if your tree tab is not currently in the directory where your notebook is then it won't do anything, so just redirect to the correct directory first, then refresh the page, and you should be able to save the notebook now

Answer 7

When I click 'save' button, it has this error. Based on the answers in this post and other websites, I just found the solution. My jupyter notebook is installed from pip. So I access it by typing 'jupyter notebook' in the windows command line.

(1) open a new command window, then open a new jupyter notebook. try to save again in the old notebook, this time ,the error is 'fail: forbidden'

(2) Then in the old notebook, click 'download as', it will pop out a new windows ask you the token.

enter image description here

(3) open another command window, then open another jupyter notebook, type 'jupyter notebook list' copy the code after 'token=' and before :: to the box you just saw. You can save this time. If it fails, you can try another token in the list

• Jorge Aurelio Menendez: This did it for me, thanks! Just wanted to clarify the last step - it took me a minute to decipher what you meant. The command jupyter notebook list will spit out a long url, part of which is a long string of characters appearing in between token= and ::. Copy and paste this string of characters into the "password or token" box from step (2) and the "forbidden" error in the original jupyter notebook will go away.

Answer 8

In my case, this problem was solved by clicking 'Kernel' (shown on the top of notebooks) and then 'Reconnect'.

Note Added: In some versions of Jupyter, there is not 'Reconnect'.

• Bill: This didn't work for me. It reconnected the Kernel but the '_xsrf' argument missing... error remained.

Answer 9

I also came across the same error. I just opened another non-running Juputer notebook and an error is automatically gone.

Answer 10

1 workaround to solve this problem is:

1. Download the notebook which you are not able to save by going to: File --> Download as --> Notebook (ipynb).

2. Open the downloaded notebook by clicking on the downloaded file on the jupyter browser UI

Now, you should be able to save (or rename) this notebook from jupyter UI

• Alex: Opening another notebook didn't solve it for me, but this did (i.e. downloading the nb)

Answer 11

I believe there are enough solutions posted to solve the problem. But just in case someone wants to know why this error arises:

The "XSRF argument missing from POST" error is typically encountered when you are trying to make a POST request to a Jupyter notebook server, and the server is unable to verify that the request is legitimate.

This error can occur when the request is missing the required XSRF (Cross-Site Request Forgery) token. This token is a security measure that helps to protect against malicious attacks by ensuring that requests to the server are legitimate.

Answer 12

I was able to solve it by clicking on the "Kernel" drop down menu and choosing "Interrupt."

Answer 13

The error was gone just after refreshing the tab of my browser.

• Sherman: This is a bit risky because you could lose your work. I would at least File > Download as > ipynb or File> Make a Copy first.

Answer 14

What worked for me was just changing the kernel.
To change the kernel just go to the top-page bar of the notebook, select Kernel, then Change kernel, and then select another Kernel among the ones you have (you need to have at least two preset kernels, obviously). After changing it the message '_xsrf' argument missing from post disappeared and the notebook could be saved again.

• Community: As it’s currently written, your answer is unclear. Please edit to add additional details that will help others understand how this addresses the question asked. You can find more information on how to write good answers in the help center.

Answer 15

I got the same problem (impossible to save either notebooks and .py modules) using an image in the nvidia docker. The solution was just opening a terminal inside jupyter without typing anything but exit once the files were saved. It was done in the same browser/jupyter instance.

Machine OS: Ubuntu 18.04

Answer 16

Open the developer setting and click console and type the following

JSON.parse(document.getElementById('jupyter-config-data').textContent).token

Then try saving the Notebook. The notebook that was not saving previously will save now.

Answer 17

The most voted answer doesn't seem to work when using Jupyter Lab. This one does, however. Just copy the url into a new tab, replace 'lab' with 'tree' and hit enter to load the page. It will generate a new csrf token for your session and you're good to go!

I would suggest enabling Settings > Autosave Documents by default to avoid worrying about losing work in future. It saves very regularly so everything should be up to date before any timeouts happen anyway.

I did not need to open a new notebook. Instead, I reopened the tree, and reconnected the kernel. At some point I also restarted the kernel. – user650654 Oct 9 '19 at 0:17

Answer 18

In My Case, I have a close tab of Home Page. After Re-opening the Jupyter.The Error was automatically gone and We can save the file.

Answer 19

For anyone using jupyterlab on AWS Sagemaker - duplicating the tab is enough to solve the issue.

Answer 20

The rest of the answers didn't work for me. I went to localhost:8888 and clicked on running, then shut down the notebook from there. Then I swapped back to the original notebook and clicked "don't restart". After that save the notebook and restart it.

Answer 21

Today, the solution is this: duplicate the tab / open the same URL in a new browser window. Whatever it does behind the scenes on a full page load seems to update the browser state enough to make the issue go away.

Answer 22

This is the easiest way.

I did not need to open a new notebook. Instead, I reopened the tree, and reconnected the kernel. At some point I also restarted the kernel. – user650654 Oct 9 '19 at 0:17