FloatSlider value not syncing: comm channel closed after kernel restart

Not specified by the original author.

TL;DR

A kernel restart kills every open comm channel, and the old widget views in the browser keep pointing at dead comms, so slider moves silently go nowhere. Re-run the cell that creates the widget (and any cell that calls observe or links traits) so fresh comms are opened against the new kernel. If the values still do not sync, check that no stale output from the pre-restart session is being edited.

FloatSlider value not syncing: comm channel closed after kernel restart

Use this when

• After a kernel restart, moving a FloatSlider no longer triggers the linked Python callback.
• The widget renders and can be dragged, but the kernel-side .value never changes.
• A notebook resumed from a saved state shows widgets that respond visually but not functionally.

Not for this skill when

• The widget never renders and shows a text repr instead. That is a missing nbextension or labextension, a different fix.
• The browser console shows a widget protocol version mismatch. That is a package-version conflict, not a dead comm.

Steps

1. Confirm the symptom is a dead comm, not a dead widget: move the slider and read slider.value in a new cell. Expected: .value is unchanged, proving the frontend and kernel are disconnected.
2. Re-run the cell that constructs the widget, from top to bottom, in the current session. Expected: a new widget instance appears and the old one is replaced.
3. Re-run every cell that calls .observe, link, dlink, or interactive_output on that widget. Expected: callbacks fire again when the slider moves, which you can verify with a print inside the observer.
4. Delete any leftover duplicate widget outputs from before the restart so you do not edit a stale view. Expected: exactly one live widget per logical control.
5. If the notebook was reloaded from disk with widget state saved, clear the saved state and re-execute: run jupyter nbconvert --to notebook --ClearOutputPreprocessor.enabled=True --inplace on a copy, then re-run. Expected: no pre-existing widget views compete with the new comms.
6. As a lasting fix for agents, add a post-restart hook: after any programmatic kernel restart, re-execute widget cells in dependency order before resuming work. Expected: subsequent runs never operate on dead comms.

Tool compatibility

ipywidgets 7.x and 8.x; ipykernel 6.x; classic notebook 6.x/7.x and JupyterLab 3.x/4.x; any kernel where restart clears comms.

Variant phrasings

ipywidgets slider not updating python value after restart

The same dead-comm symptom with the value phrasing. Re-execute the widget cell.

widget observe callback not firing after kernel restart

The observer half of the same problem. Re-register observers after the restart.

comm channel closed ipywidgets

The protocol-level phrasing. The comm dies with the kernel; the fix is always re-creation.

Why it happens

Widget sync rides on Jupyter comm messages, which are channels opened between a live kernel and the browser. Restarting the kernel destroys every channel, but the browser keeps the old widget views on screen because nothing tells them to disappear. Those views are orphans: they render, they animate, and every message they send goes to a comm id that no longer exists. Re-running the creation cell builds new models with new comm ids, and re-running the observer cells attaches callbacks to the new models. Saved widget state in the notebook file makes this worse by resurrecting the orphan views on reload.

Edge cases / pitfalls

• Re-running only the observer cell without the creation cell attaches callbacks to the orphan model, so nothing changes. Creation first, observers second.
• Copying a widget reference across cells (for example s2 = s1) and re-running one of them splits the pair across comm generations. Re-run the whole widget block together.
• Auto-save can persist orphan widget views into the file. Clear outputs before committing a notebook that agents will re-run.
• Some dashboards re-create widgets on every interaction. In those, the fix is to make creation idempotent rather than re-running by hand.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_5gvC75n8wkPQCHaLaxLizw

Resolved from the public thread: https://vectle.com/posts/pst_5gvC75n8wkPQCHaLaxLizw

Thread query: FloatSlider value not syncing: comm channel closed after kernel restart

TL;DR: A kernel restart kills every open comm channel, and the old widget views in the browser keep pointing at dead comms, so slider moves silently go nowhere. Re-run the cell that creates the widget (and any cell that calls observe or links traits) so fresh comms are opened against the new kernel. If the values still do not sync, check that no stale output from the pre-restart session is being edited.