metric-proxy

metric-proxy

The metric-proxy acts as a translation layer between the Kubernetes Metrics Server and CAPI for use in cf-for-k8s. When a request comes into the metric-proxy the proxy will make a request to the Kube API and format the results to match the existing log-cache API. This ensures developers can get metrics from existing versions of the cf cli. when running commands cf push and cf app.

Usage

Push an application and see its metrics with cf push <app_name>

$ cf push diego-docker-test -o cloudfoundry/diego-docker-app

Pushing from manifest to org system / space system as admin...
Getting app info...
Updating app with these attributes...
  name:                diego-docker-test
  docker image:        cloudfoundry/diego-docker-app
  disk quota:          1G
  health check type:   port
  instances:           1
  memory:              1G
  stack:               cflinuxfs3
  routes:
    diego-docker-test.example.com

Updating app diego-docker-test...
Mapping routes...
timeout connecting to log server, no log will be shown


Stopping app...

Waiting for app to start...

name:                diego-docker-test
requested state:     started
isolation segment:   placeholder
routes:              diego-docker-test.example.com
last uploaded:       Fri 06 Mar 16:03:06 MST 2020
stack:
docker image:        cloudfoundry/diego-docker-app

type:           web
instances:      1/1
memory usage:   1024M
     state     since                  cpu    memory        disk      details
#0   running   2020-03-09T15:33:37Z   0.3%   27.2M of 1G   0 of 1G

Get application metrics using cf app <app_name>

$ cf app diego-docker-test

Showing health and status for app diego-docker-test in org system / space system as admin...

name:                diego-docker-test
requested state:     started
isolation segment:   placeholder
routes:              diego-docker-test.example.com
last uploaded:       Fri 06 Mar 16:03:06 MST 2020
stack:
docker image:        cloudfoundry/diego-docker-app

type:           web
instances:      1/1
memory usage:   1024M
     state     since                  cpu    memory        disk      details
#0   running   2020-03-06T23:03:08Z   0.3%   27.1M of 1G   0 of 1G

Metric Details

The kubelet reports CPU usage in millicores (thousandths of a core). See Kubernetes docs for more details. This is reported as a percentage in the Cloud Foundry CLI.

For backwards compatibility, disk usage is still presented in the output but will always return 0 as there is no equivalent pod metric in Kubernetes.

How to Contribute/Develop metric-proxy

There are several scripts to help automate building and deploying new versions of metric-proxy.

./hack/build-dev-image.sh — runs tests and builds and pushes a new docker image to cloudfoundry/metric-proxy:dev

./hack/bump-cf-for-k8s.sh - updates the vendired metric-proxy directory in the cf-for-k8s repo with any changes made to metric-proxy

./hack/cf_for_k8s_env_deploy.sh - helper script to create a GCP K8s cluster, deploy cf-for-k8s, and update DNS records. Update this script with your GCP Service Account key and DNS information.

./hack/cf_for_k8s_login.sh - helper to cf login to your cf-for-k8s deployment using CF CLI v7

Confirm metric-proxy works by following steps in the Usage section.

See Concourse CI: cf-k8s-metric-proxy-validation

Have a question or feedback, reach out to us

Reach out to us in the Cloud Foundry Slack channel #cf-for-k8s