Skip to content
Branch: master
Find file Copy path
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
187 lines (128 sloc) 7.42 KB

Monitoring and Visualization with Prometheus and Grafana

Illuminating Your Environment with Prometheus

In this tutorial, we will install Prometheus on our cluster to gather metrics, and Grafana to enhance the metrics. Grafana is another OSS solution often paired with Prometheus and other monitoring tools to model the data collected into beautiful and useful dashboards.

Getting Started

For this solution we will use Prometheus and Grafana to monitor a containerized environment managed by Kubernetes. We will use Oracle Container Engine for Kubernetes (OKE) for our Kubernetes cluster. OKE is a fully-managed, scalable, and highly available service that you can use to deploy your containerized applications to the cloud. To start an OKE cluster follow this friendly guide.

We can get started with Helm as soon as our cluster is ready. You can find the proper binary to download and install Helm on your system in the GitHub repository of the project. After the download is complete, open a terminal window and type in helm You will see an output similar to this:

helm image

If you've gotten this far, you're halfway there! Now we install the Prometheus operator to our cluster. To do this, run:

helm repo add coreos

Installing Tiller

OKE can come preconfigured with Tiller, a component that sits on your Kubernetes cluster to help manage your Helm deployments. If you clicked the box that said "Tiller (Helm) Enabled" simply run to upgrade it to match the client.

helm init --upgrade

There is no need to worry if you chose not to check that box. If so, follow these steps to configure Tiller.

First, create a service account on your cluster that's configured with the proper permissions.

kubectl -n kube-system create sa tiller

Next, we create a cluster role binding for our service account, to give it the proper permissions to manage releases.

kubectl create clusterrolebinding tiller --clusterrole cluster-admin --serviceaccount=kube-system:tiller

Lastly, we configure Helm to use Tiller by running:

helm init --service-account tiller

Now you are ready to install the operator on your system in a separate namespace.

helm install --namespace monitoring coreos/prometheus-operator

Installing Prometheus

Next, let's install kube-prometheus which gives us some Kubernetes manifests for Grafana dashboards, and Prometheus rules that allow us to operate Kubernetes. But first we need to download some values to use for our deployment:


And install it with the following command:

helm install coreos/kube-prometheus --name kube-prometheus --namespace monitoring --values values.yaml

You can view all of your newly started pods by running:

kubectl get po --namespace monitoring

The Magic Behind the Charts

Let's take a look at what we deployed by reviewing our values.yaml file, which contains the default values for our chart. As an example of how this works, let's zoom in on a specific section describing our storage.


        storageClassName: oci
        accessModes: ["ReadWriteOnce"]
            storage: 50Gi

This specifies the use of an OCI Block Volume. This allows for our metrics to persist in the event of a pod restart. If you find you need additional space, simply change the configuration.

The Prometheus endpoint is exposed by default as a ClusterIP, which means that it will not be reachable outside of the cluster. To make it reachable, edit the service and replace 'ClusterIP' with 'NodePort':

kubectl edit svc kube-prometheus -n monitoring

Prometheus image

Run the following command to discover the port used to access Prometheus:

kubectl get svc --all-namespaces

Run the following command to get the IP addresses of your nodes:

kubectl get nodes

Add the port number to the IP address of your node to get access to the Prometheus portal.

Prometheus image2

Try querying for basic pod information. For example, you can search 'kube_pod_status_phase'

Prometheus image3

The next thing to do is to add custom dashboards to Grafana to make our cluster monitoring easier. We are going to once again leverage our Helm chart's values.yaml file to accomplish this. Let's edit the values serverDashboardFiles under the Grafana object in this file. The dashboards tend to be a bit long, so we'll go over this at a very high level.

In order to include alternative ways of looking at cluster health you can create a new file under the serverDashboardFiles that holds the dashboard information within it. If you want to add more, simply write a new file with a dashboard in it!

    kube-custom-dashboard.json: |-
      "dashboard": {
        "timezone": "browser",
        "title": "Kubernetes Cluster (Prometheus)",
        "uid": "4XuMd2Iiz",
        "version": 1

If you have existing Grafana dashboards that you would like to define, use this script to transform them into a format that the watcher can understand.

# !/bin/bash
# 2. ./ new_dashboard_xxx
# - This will create new_dashboard_xxx-dashboard.json

DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"

if [ "$#" -ne 1 ]; then
  echo "Usage: ./ <name_of_dashboard>"

# Prepare json for Grafana Watcher
pbpaste | \
  jq '.id = null | del(.__requires) | del(.uid) | { dashboard: . , inputs: .__inputs, overwrite: true  }' | \
  jq '.inputs[0].value=.inputs[0].pluginId' | \
tee "$DIR/$1-dashboard.json"

Finally, we need to access Grafana to view our metrics. To accomplish this, change the Grafana service type to be NodePort as we did with Prometheus. People typically will use ingress or the "LoadBalancer" service type with Grafana because of how often the dashboards are viewed.

Run the following command to discover the port used to access Grafana:

kubectl get svc --all-namespaces

Run the following command to get the IP addresses of your nodes:

kubectl get nodes

Add the port number to the IP address of your node to get access to the Grafana portal.

It should look something like this:

Grafana image

If you are prompted for login information, use the default username admin and password admin specified in the Helm chart.

Grafana image2

Explore the Grafana documentation for information about how to create more advanced charts to take full advantage of the service.

Grafana image3

And with that you are done! Now that you have an OKE cluster running a highly available monitoring setup complete with Grafana and Prometheus you can spend more time writing your applications rather than instrumenting and maintaining monitoring infrastructure. Happy hacking!

You can’t perform that action at this time.