Description
Thank you for this great extension! It is one of the basic extensions in our system-wide JupyterLab installation on multiple clusters and we love it :) Because of that I have very high interest that the following issue is known.
jupyter-resource-usage sets enable_prometheus_metrics = True by default.
But this has a huge impact on the whole JupyterLab installation:
At the moment jupyter-resource-usage is overloading tornado's task-pool for the prometheus- (and the cpuload- feature (?) ) without taking into account that this will result in less (or even worst) performance for the critical webserver.
We had a big issue with our server-proxies started via jupyter-server-proxy as we saw a significant performance degradation which was extremely time consuming to track down in a complex cloud environment. The remote desktop integration Xpra was showing stuttering effects ( jupyterhub/jupyter-server-proxy#494 ) and the robustness of web socket connections in general were affected at multiple places.
We were looking in all directions (NGINX, JupyterHub, Kubernetes, OpenStack, Network, JupyterLab, jupyter-server-proxy, etc) but unfortunately the last we had in mind was jupyter-resource-usage.
We are fine now as we have changed our jupyter-resource-usage's config of course. But other users will stumble over the issue in the same way.
This problem has already been discussed in the past here: #123 and an opt-out functionality was implemented here dleen@8c5d477 .
=> I would suggest to change that to an opt-in.