Many cloud providers expose platform-specific metrics that can be used by the admin to understand complex problems. One example is throttling in Azure. Today if an OCP cluster is being throttled the only thing observable is the latency through disk or network. But this is not enough because the user is left to assume the root cause for the latency.

For this reason, it is important that the cluster has observability of such metrics so that alerts can be crafted around these events. Imagine the reduced support calls for Azure performance if an Alert was constantly firing stating that the node is being throttled and that they should review steps to vertically scale.

cloud-credential-operator is the logical place for this code as they have the credentials already for the API requests.

ref:

• https://github.com/RobustPerception/azure_metrics_exporter
• https://azure.microsoft.com/en-us/updates/new-disk-io-and-bandwidth-metrics-to-identify-bottlenecks/
• https://coreos.slack.com/archives/C0VMT03S5/p1604424088162000
• https://github.com/openshift/cloud-credential-operator