I am trying to understand how the exposure and collection of pod-level metrics on Windows nodes/containers works.
I have an EKS mixed cluster, with both Windows and Linux worker nodes. The version of the control plane and nodes is 1.15.10. Metrics are aggregated in Prometheus and visualised from there. I have been reasonably successful in getting Prometheus to scrape the Windows node metrics (via wmi_exporter) as well as the application metrics exposed on the pods themselves. However, I can’t seem to get the pod-level metrics to work.
To my understanding, they should be provided by kubelet through the winstats package. In my Prometheus install, I can see the scrape target on the Windows nodes, with the URL https://<node_IP>:10250/metrics/cadvisor, and Prometheus indicates that it is able to scrape that target - no errors are shown. However, when I browse the metrics, none of the windows container pod metrics show up.
Is there something else I have to do to actually make these metrics available?
Any assistance is appreciated.