We have three sources of metrics so far:
- Usage metrics of containers coming from the caadvisor see full list of caadvisor metrics and node usage metrics is coming from node-exporter
- Istio nework metrics coming from the istio exproter
- Seldon specific metrics coming from the Seldon core exporter
Here is the table of the metrics we used in this project and are of interest to us:
TODO make a table of metrics like full list of caadvisor metrics
Check the following links for a better understanding of the Prometheus and PromQL language:
The metrics metadata are accessible from the HTTP API of the prometheus accessible at /api/v1
. The api can also gives us the same information as the dashboard. However, the metadata of the metrics is only avaialable from the http API. E.g.
$PROM_SERVER=http://192.5.86.160:30090
curl -G $PROM_server/api/v1/targets/metadata --data-urlencode 'metric=container_cpu_cfs_periods_total'
returns:
{"status":"success","data":[{"target":{"endpoint":"https-metrics","instance":"10.140.81.236:10250","job":"kubelet","metrics_path":"/metrics/cadvisor","namespace":"kube-system","node":"k8s-cluster","service":"kubelet"},"type":"counter","help":"Number of elapsed enforcement period intervals.","unit":""}]}%