Pyroscope is an interesting tool that can be very useful when reporting/troubleshooting a low level performance issue.
Deploying and configuring it as a 2nd day-op once the cluster is ready would be great.
For the moment, I'd configure pyroscope to scrape these components, which already expose the /pprof endpoints by default:
- kube-apiserver (Bearer token authentication)
- kube-controller-manager (Bearer token authentication)
- etcd (certificate authentication)
We can also consider scraping other core components like (optionally ? ):
- ovnkube-master
- ovnkube-node
- kubelet
- cri-o
- kube-scheduler
- operator-lifecycle-manager
- openshift-apiserver
- openshift-kube-controller-manager
- prometheus
Prior start developing this RFE, we should investigate how much space requires pyroscope, and determin the overhead (if relvant) it adds to these components
Pyroscope is an interesting tool that can be very useful when reporting/troubleshooting a low level performance issue.
Deploying and configuring it as a 2nd day-op once the cluster is ready would be great.
For the moment, I'd configure pyroscope to scrape these components, which already expose the /pprof endpoints by default:
We can also consider scraping other core components like (optionally ? ):
Prior start developing this RFE, we should investigate how much space requires pyroscope, and determin the overhead (if relvant) it adds to these components