Compute resource metrics roadmap

[mvdan]

As a means to keep track of what I am doing so that it is clear to others.

• Write proposal - Add compute resource metrics API proposal #11951
• Add apiserver experimental nodeMetrics resource endpoint - apiserver nodemetrics endpoint #12952
• Write nodeMetrics resource endpoint client interface - also apiserver nodemetrics endpoint #12952
• Add util.Duration type required to index time windows by duration - Move util.Time to api/unversioned and add Duration #14096
• Add Heapster experimental nodeMetrics endpoint (to be used as data source) - Add nodeMetrics expapi/v2 endpoint kubernetes-retired/heapster#549
• Write program to periodically pipe metrics from a heapster to apiserver - Add nodeMetrics expapi/v2 endpoint kubernetes-retired/heapster#549
• Write rc config to run apiserver, etcd, heapster and the program above - Add nodeMetrics expapi/v2 endpoint kubernetes-retired/heapster#549
• Make kubectl show resource metrics on nodes for cpu and memory

Extra steps to improve efficiency and move some of the workload to the kubelets:

• Add instantaneous cpu metrics to cadvisor - Add instantaneous cpu metrics google/cadvisor#861
• Improve kubelet metrics APIs (Improve kubelet metrics APIs #12483)

CC @vishh

[vishh]

cc @bgrant0607

[resouer]

/sub Great work!

[bgrant0607]

If we add metrics to existing kubectl commands, such as describe pods, we need them to fail gracefully/silently in the case that the experimental metrics API is not enabled/present.

[mvdan]

Updated progress since the bridge between heapster and apiserver is now done, as well as the rc config. I'm pushing all of those to kubernetes-retired/heapster#549 too.

[vishh]

@mvdan: Can you publish your draft PRs that resolves some of the pending work items? I'm referring to the metrics service with a secondary api server, kubectl changes, etc.

[vishh]

Closing this issue since @piosz has picked up this effort.