Is there any plan to provide metrics of clusters joined in karmada or other similar functionality?

[feiyudev]

Please provide an in-depth description of the question you have:

How to get metrics of all clusters after several k8s cluster joined in karmada? It seems that I need to set up metrics for each cluster now, and get metrics from each cluster NOT karmada.

What do you think about this question?:

Karmada can manage multiple clusters, make multi-cluster into one-cluster. So here is the question, can we get metrics of multi-cluster just like one-cluster?

At present, maybe I can use the aggregator apiserver to proxy the service of metrics pod in joined clusters, and get the metrics by a long url. But I think it is not an official support.

So has the functionality above been implemented already? Or is there any plan to support the functionality above?

Thanks in advance.

Environment: (not important)

• Karmada version: v1.2.1
• Kubernetes version: v1.23.3
• Others: None

[zirain]

I think you can get it by running kubectl get cluster --kubecofig <karmada-apiserver> --context <karmada-apiserver>

BTW, it's a good idea to add some basic metrics for karmada components, WDYT? cc @RainbowMango

[RainbowMango]

Yes, I agree, but for now just a draft idea, no clear plan for it.

[GitHubxsy]

We've been thinking about the observability of multiclusters. What do you suggest on this topic? Or describe what you think of as a multi-cluster metric.

[feiyudev]

I think you can get it by running kubectl get cluster --kubecofig <karmada-apiserver> --context <karmada-apiserver>

BTW, it's a good idea to add some basic metrics for karmada components, WDYT? cc @RainbowMango

Sorry for my bad expression, the metrics I mentioned can be scraped by prometheus or other TSDB, provided by node_exporter or something else. The cluster info from karmada is not what I mean.

[feiyudev]

We've been thinking about the observability of multiclusters. What do you suggest on this topic? Or describe what you think of as a multi-cluster metric.

Yeah, "observability of multilusters" is more accurate. Below is the situation for single cluster and multi cluster.

• For single cluster, create a node_exporter workload & svc, configure the prometheus in the cluster or outside to scrape;
• For multi clusters, create a node_exporter workload & svc in every cluster, configure the prometheus for each cluster, and maybe the communication between prometheus and metrics scraped was direct, karmada was bypassed.

So can karmada provide a method for prometheus to scrape all cluster's metrics. In other words, can karmada provide a method to reach cluster's metric instead of reaching cluster's metrics directly.

I don't know whether this is a good idea for karmada or not, so asked in this issue.

[XiShanYongYe-Chang]

/cc @zirain @RainbowMango @GitHubxsy Ask related guys to help take a look again.