Prometheus Operator

by  CoreOS

The Prometheus operator provisions and configures the popular Prometheus monitoring system and time series database on top of Kubernetes. It allows setting both Prometheus configuration like versions, persistence, retention policies as well as monitoring targets through Kubernetes custom resources and manages the lifecycle of the Prometheus instances.

You can use the operator to provision Prometheus instances to monitor any workload that exposes metrics in the Prometheus format. Additionally the operator can also provision AlertManager for alerting.

TL;DR:

  • Installing, updating or removing follows Kubestack's GitOps flow.
  • Instructions assume the default repository layout.
  • Bases can be consumed as-is or customized.
  • Step-by-step instructions are framework specific but bases can be used independently.

Install

  1. Vendor the base

    # Run these commands from the root of your Kubestack infra repository
    wget https://storage.googleapis.com/catalog.kubestack.com/prometheus-v0.43.0-kbst.0.zip
    unzip -d manifests/bases/ prometheus-v0.43.0-kbst.0.zip
    rm prometheus-v0.43.0-kbst.0.zip
  2. Include resource in apps overlay

    cd manifests/overlays/apps
    kustomize edit add resource ../../bases/prometheus/clusterwide
  3. Commit and push

    cd -
    git checkout -b add-prometheus
    git add manifests/bases/prometheus manifests/overlays/apps/kustomization.yaml
    git commit -m "Add prometheus v0.43.0-kbst.0 base"
    git push origin add-prometheus
  4. Review PR and merge

    Finally, review and merge the PR into master. Once it's been successfully applied against the Ops-Cluster set a prod-deploy tag to also apply the change against the Apps-Cluster.

Update

To update the operator delete the previously vendored base and then vendor the new version.

  1. Delete the previous vendored version

    # Run these commands from the root of your Kubestack infra repository
    rm -r manifests/bases/prometheus
  2. Vendor the new version

    # Run these commands from the root of your Kubestack infra repository
    wget https://storage.googleapis.com/catalog.kubestack.com/prometheus-v0.43.0-kbst.0.zip
    unzip -d manifests/bases/ prometheus-v0.43.0-kbst.0.zip
    rm prometheus-v0.43.0-kbst.0.zip
  3. Commit and push

    git checkout -b update-prometheus
    git add manifests/bases/prometheus
    git commit -m "Update prometheus base to v0.43.0-kbst.0"
    git push origin update-prometheus

Remove

Operators often create resources based on custom objects. When removing an operator, follow a two-step process to ensure operator provisioned resources are purged properly.

  1. Remove all the operator's custom objects.
  2. Once the operator had time to de-provision all resources it created, follow the instructions below to remove the operator itself.
  1. Remove resource from apps overlay

    cd manifests/overlays/apps
    kustomize edit remove resource ../../bases/prometheus/clusterwide
  2. Delete the vendored base from your repository

    cd -
    # Run these commands from the root of your Kubestack infra repository
    rm -r manifests/bases/prometheus
  3. Commit and push

    git checkout -b remove-prometheus
    git add manifests/bases/prometheus
    git commit -m "Remove prometheus base"
    git push origin remove-prometheus

Usage

Using the Prometheus operator to provision a Prometheus instance and start monitoring an application requires three steps:

  1. Deploy the operator
  2. Provision a Prometheus instance
  3. Create ServiceMonitors for each service exposing metrics

The first step was to deploy the Prometheus operator following the instructions on the install tab.

Prometheus Instance Manifests

Next, we can continue with step number two and use the operator to provision our Prometheus instance.

Each Prometheus instance needs read-only access to the Kubernetes api in order to keep its monitoring targets up to date. To make this easier, the clusterwide base includes two ClusterRoles. One for the operator, called prometheus-operator, and one for the instances, aptly named prometheus-instance.

So, in order to provision a Prometheus instance we need to first create a ServiceAccount and a RoleBinding linking that service account to the pormetheus-instance cluster role. Finally, we can create a Prometheus resource that instructs the operator to provision a Prometheus instance that uses our service account. Below example does just that.

---
apiVersion: monitoring.coreos.com/v1
kind: Prometheus
metadata:
name: example-instance
namespace: default
labels:
prometheus: example-instance
spec:
serviceAccountName: prometheus-example-instance
serviceMonitorSelector:
matchLabels:
prometheus-instance: example-instance
resources:
requests:
# by default the operator requests 2Gi of memory
# adapt the line below if required to schedule pods
memory: 2Gi
---
apiVersion: rbac.authorization.k8s.io/v1beta1
kind: RoleBinding
metadata:
name: prometheus-example-instance
namespace: default
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: prometheus-instance
subjects:
- kind: ServiceAccount
name: prometheus-example-instance
namespace: default
---
apiVersion: v1
kind: ServiceAccount
metadata:
name: prometheus-example-instance
namespace: default

ServiceMonitors Manifest

Finally, we need to create a ServiceMonitor to instruct the operator, to configure our Prometheus instance to scrape metrics from each of the replicas of an application that exposes its metrics in the Prometheus format.

Since this step is depending on the application and where it exposes the metrics, you will have to adapt the example below.

But there are basically three things to note. First, the metadata.labels need to match the spec.serviceMonitorSelector.matchLabels from the Prometheus resource. Next, the spec.selector.matchLabels below needs to match the metadata.labels set for the service of the application you are trying to monitor. Finally, the spec.endpoints need to match the port name, configured in the application's service.

apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
name: application-example
labels:
# this label instructs Prometheus to include this ServiceMonitor
# based on the `spec.serviceMonitorSelector.matchLabels` above
prometheus-instance: example-instance
spec:
selector:
matchLabels:
# this selector is how the `ServiceMonitor` finds the
# application's service
app: application-example
endpoints:
# this tells Prometheus on what port the app exposes its metrics
- port: web

Prometheus UI

To check if Promethues has started scraping your applicaiton's metrics, you can access the Prometheus UI using kubectl port-forward and use it to run a query.

kubectl port-forward prometheus-example-instance-0 9090

Advanced Configuration

Above instructions only cover the basics. The Prometheus operator has many more features and options. For more extended documentation about Prometheus operator configuration, please refer to the official documentation.