Metrics Server: Difference between revisions
DrEdWilliams (talk | contribs) (Created page with "Kubernetes provides for metrics collection on performance/capacity of nodes, pods, etc. Previously, this was done by the Heapster application, but that is now deprecated in f...") |
DrEdWilliams (talk | contribs) mNo edit summary |
||
| Line 8: | Line 8: | ||
<pre> containers: | <pre> containers: | ||
- name: metrics-server | - name: metrics-server | ||
image: k8s.gcr.io/metrics-server-amd64:v0. | image: k8s.gcr.io/metrics-server-amd64:v0.x.x | ||
imagePullPolicy: Always | imagePullPolicy: Always | ||
command: | command: | ||
Revision as of 12:28, 25 December 2019
Kubernetes provides for metrics collection on performance/capacity of nodes, pods, etc. Previously, this was done by the Heapster application, but that is now deprecated in favor of an internal metrics server. This capability is still in kubernetes incubator status, but it provides a significant capability at this time.
Installation
There is a helm chart (untested), but the deployment is very straightforward using the GitHub repository:
git clone https://github.com/kubernetes-incubator/metrics-server.git
Due to the fact that the default kubeadm installation uses self-signed certificates for the kubelets (not certificated signed by the main CA for the cluster), one minor change needs to be made to the deployment if this is being applied to an existing cluster. Modify the file metrics-server/deploy/1.8+/metrics-server-deployment.yaml, adding the command: section to the containers spec so that it looks like this<ref>https://github.com/kubernetes-incubator/metrics-server/issues/146#issuecomment-430483619</ref>:
containers:
- name: metrics-server
image: k8s.gcr.io/metrics-server-amd64:v0.x.x
imagePullPolicy: Always
command:
- /metrics-server
- --kubelet-insecure-tls
volumeMounts:
- name: tmp-dir
mountPath: /tmp
This tells the metrics server to ignore the fact that the certificates are self-signed -- in general, not the best solution, but as it is totally within the cluster, the exposure is minimal.
Deploy the metrics server:
kubectl apply -f metrics-server/deploy/1.8+/
<references/>