search

rapidsai / jupyterlab-nvdashboard

A JupyterLab extension for displaying dashboards of GPU usage.
BSD 3-Clause "New" or "Revised" License
598 stars 78 forks source link

add dynamic updateFrequency via settingsRegistry #182

Closed AjayThorve closed 7 months ago

AjayThorve commented 8 months ago

Implements #179. Also added the ability to set a custom refresh speed via jupyterlab settings. The settings once set will be global across the jupyterlab session, even when you restart the jupyterlab server. The setting also has type checking along with a minimum value of 1ms.

cc @jacobtomlinson

Recording 2024-02-07 at 14 36 28

AjayThorve commented 8 months ago

I definitely appreciate making this configurable.

Have we done any testing to see what the performance impact is when setting it to smaller numbers?

The previous bokeh implementation updated many times per second with no noticeable impact. I think we should endeavour to set the default poll rate as fast as we can without it taking noticeable amount of resources.

@jacobtomlinson I have updated the default to 100ms, there is no performance impact at this value. Below 50ms is when I start to see some rendering backlog, but even then, you eventually see the entire timeline in a couple of seconds.