How are dashboards served and what would happen to them if the cluster attached to the notebook terminates?

[fa]

I have two dashboards in presentation mode both from notebooks being run on the same compute cluster. Last night the cluster terminated due to idle time and in the morning one of my dashboards was fine but the other one was set to the default stab diffuser demo! I wonder how dashboards are served. Is each running on an isolated compute instance and how my other dashboard managed to keep up on its presentation link while the other needed a rerun?

The only difference between the two notebooks is that one of them is only using python models from standard libraries and the other is using a model I have trained but again both model have been registered in production mode using MLFlow. If I change the models that a running dashboard is using (e.g., change the version of the model for production) without changing the notebook code or restarting it, what would be the expected behaviour observed in the presentation link?

[Ajay-Pandey]

If your query were scheduled, it's automatically started the cluster at the scheduled time.

[fa]

I don't have a query and the notebook is not currently scheduled to restart. So I was supposing the app link would be valid for 72 hours (using a gradio interface and databricks presentation mode). The question is if the cluster terminates for some reason, do the apps running off it stop working and would need a manual rerun or not. In each case, one of my apps was showing a weird behaviour, as one was still running and the other was set to the default/demo gradio app.

[akdm]

If the cluster terminates there's nothing to serve the app/dashboard. Is one of the two static or "more static"?

[User16515675729]

When the cluster gets terminated, we can still see the UI, but it will be loading forever and the app is not up and running anymore in my understanding.

[akdm]

It's possible that the portion that is still visible doesn't need to be generated so it looks like it's working but it is just left over from the prior run.

[Manoj12421]

​If your query were scheduled, it's automatically started the cluster at the scheduled time Or might be possible that the portion that is still visible doesn't need to be generated so it looks like it's working but it is just left over from the prior run.