Currently being generated cell output disappears once you close/reload the page

[deKeijzer]

Maybe not so much a bug, more a quality of life request.
Would it be possible to 'refresh' the cell output, after closing/reloading the webpage?

Currently while running e.g.

import time

for i in range(100):
     print(i)
     time.sleep(1)

When you reload the page when i=5, the cell output will stay at 5 and not update. However, the kernel does stay running and after this loop is done, one can just continue using the kernel of said notebook (which is convenient).

When one is using loops to process data, it occasionally can take some hours for a cell to complete, while the output of the cell is printing statistics. It would be nice to be able to close the page, open it a could hours later and get the 'updated' or at least new cell output.

Terminals already have this ability, which is awesome. If notebooks could get this ability, that would be great.

[jasongrout]

This is an issue with the larger Jupyter architecture and the 'right' fix requires significant changes to how Jupyter kernel interaction works. In particular, kernel messages and the notebook state should be stored in the server, not in browser. Good news, though: there is active work, notably in the nteract project, to fix this architectural issue with Jupyter.

CC @rgbkrk @captainsafia

[rgbkrk]

Yes! We're exploring this now. However, there's no firm deadline on this, as it's still formative.

We're going to lean heavily on the implementation of sync within sagemathinc's cocalc which supports realtime interactive jupyter notebooks. That work is going to be extracted into a BSD-3-Clause sync library.

The intention is that we'll create a well documented and specified protocol that all modern jupyter frontends can use (jupyterlab, nteract, papermill, nbconvert).

Another part of this effort has been to firm up some of the kernel status and messaging with a relay with message recall built in (contrary to our current message replay mechanism): https://github.com/nteract/kernel-relay

Note: I include papermill and nbconvert as we'd ❤️ to be able to write to a notebook as part of a job in automation while seeing a live view in a separate frontend.

---

Additional interested participants: @willingc, @williamstein, @mpacer, @MSeal

Related issues:

• Restoring computation output after disconnect in Notebook jupyter/notebook#641
• keep notebook running after the browser tab closed jupyter/notebook#1647

[vidartf]

Note, that once RTC is in place, we could add a "super node" running in Node on the server to incorporate the jupyter messages into the notebook document.

[ellisonbg]

To add to what @vidartf has said, this has come up a number of times and the tentative idea is to have the kernel service take kernel messages and covert them to patches to the CRDT store, essentially acting as another "collaborator". Non stateful kernel messages (widgets, maybe others) can still be let through a propagate to the frontend.

[ellisonbg]

Moved this last comment over to #5382 as it was related to the bigger picture of real time collaboration.