We occasionally see a dashboard crash under load. Thus far, it’s been challenging to reproduce, so we’re looking to add monitoring/resilience to the dashboard as we test out the job library.
Is there a way to programmatically restart the dashboard (either via shell/Python) or set it to auto restart?
Hi, unfortunately there is no way to automatically restart the dashboard as of now.
We’re currently working on several fixes to the stability / scalability of the dashboard. For example, we have greatly reduced its memory usage.
As the ray state observability APIs become stable, we hope to completely deprecate some of this older dashboard api code since the state observability APIs are more scalable.