Add option to enable/disable heapster integration from dashboard

[mccormd]

Environment

Dashboard version: 1.6.3
Kubernetes version: 1.7.3
Operating system: CoreOS

Steps to reproduce

When I start the dashboard with the workaround for heapster (it does not work without it): -

      - name: kubernetes-dashboard
        image: gcr.io/google_containers/kubernetes-dashboard-amd64:v1.6.3
        args:
        - --heapster-host=http://heapster.kube-system

If heapster service comes up after the dashboard pods then the dashboard does not display any graphs.

Observed result

Kubernetes dashboard does not retry for heapster availability and so no graphs are displayed within the dashboard.

Forcing a redeployment/killing the dashboard pods once heapster is up brings back the graphs.

Expected result

I don't want to have to start the dashboard after heapster as this is hard to orchestrate.
The dashboard should continue to check for heapster availability and start displaying graphs once it is available rather than checking once.

[floreks]

We have disabled this option to avoid unnecessary timeouts and checks if heapster is not present during the start. Some people were reporting that if heapster could not be reached it caused major delays in loading dashboard page. It will stay this way until we add settings page with option to enable/disable integrations on-the-fly.

[floreks]

As for the --heapster-host=http://heapster.kube-system it should not be needed if cluster is working properly. Heapster is automatically discovered by dashboard on start . Only requirement is that heapster service is named heapster and is located in kube-system namespace. We are using then in-cluster communication through kubernetes service proxy.

[floreks]

I have changed title so we can keep it as a tracking issue.

[mccormd]

Hi, I may have missed the point, sorry, but I don't want a single check for heapster upon start up or a settings page within the dashboard. We bootstrap a number of different kubernetes clusters and so I want the dashboard to come up with heapster integrated graphs reliably, even if the heapster pod has not yet started when the dashboard is loading.

Maybe there could be a command line flag to enable/disable heapster integration or not but when enabled the dashboard should periodically check for heapster availability so that it starts serving the graphs once heapster becomes available. The same mechanism might also be used to prevent attempts to get the graphs if heapster has been tested to be unavailable.

[floreks]

Ok, I get what you mean. We'll add such feature together with option to enable/disable integrations but currently we have more important issues to work on. Previously on every request we've tried to get heapster metrics without checking if it is present. Due to that some people reported that dashboard loading time increased. For now we've chosen to resolve this issue this way and work on improvements in the future.