Self-hosted Grafana

This guide shows you how to deploy and configure a self-hosted Grafana instance inside a CoreWeave Kubernetes Service (CKS) cluster. If you need more customization and control than CoreWeave Grafana provides, you can deploy a self-hosted Grafana instance. With self-hosting, you can create new dashboards, export or edit CoreWeave Grafana dashboards, and freely explore all available metrics in your environment. CoreWeave Charts deploys self-hosted Grafana instances. This handles most of the process of configuring data sources, but authentication requires some manual configuration. By the end of this guide, you have a running self-hosted Grafana instance connected to CoreWeave Metrics and CoreWeave Logs data sources.

​

Prerequisites

To set up a self-hosted Grafana instance, you need the following:

• A CoreWeave API Access Token for an active CKS account.
• A CKS cluster where you intend to self-host (deploy and manage) Grafana.
• kubectl installed.
• helm installed.

​

Install Grafana with CoreWeave Charts

This section installs the Grafana Helm chart into your cluster so the application is running and ready to configure.

1. Add the CoreWeave Charts repository:
   helm repo add coreweave https://charts.core-services.ingress.coreweave.com
   
2. Install the grafana chart:
   helm install -n "[MY-NAMESPACE]" grafana coreweave/grafana
   

​

Log in to Grafana with admin credentials

With Grafana installed, the next step is to reach the application in your browser and sign in as the administrator so you can finish configuration. Before you log in to Grafana, you must port forward your Grafana service.

1. Port forward your Grafana service:
   kubectl port-forward svc/grafana -n "[MY-NAMESPACE]" 8900:80
   
2. Navigate to localhost:8900. You see a login page similar to the following:
3. Obtain the admin password, which is required to log in to Grafana for the first time. This password is stored as a Kubernetes Secret. Retrieve the Secret value, replacing [MY-NAMESPACE] with your namespace ID:
   kubectl get secret grafana -n "[MY-NAMESPACE]" -o=jsonpath='{.data.admin-password}' | base64 --decode
   
   Note: If you use a Mac, you might need to remove the trailing %.
4. Use the returned value as your password when you log in to Grafana with the following credentials:

   Field	Value
   Username	admin
   Password	[MY-SECRET]

​

Configure CoreWeave Metrics data source credentials

After you log in to Grafana with your admin credentials, you must configure credentials for the CoreWeave-provided data sources so Grafana can authenticate to the CoreWeave observability backend. First, configure the CoreWeave Metrics data source.

​

Configure the data source HTTP headers

The following HTTP header values are required to connect to CoreWeave’s data sources, where [MY-TOKEN] is your CoreWeave API Access Token. To configure these headers, navigate to Connections from Grafana’s left-hand menu to display the dropdown options. Then, click Data sources to proceed to the Data sources page. Select CoreWeave Metrics from the options listed to proceed to the Settings menu for the new data source. Next, navigate to the Authentication section. This is the only section that requires editing. Then, click the dropdown arrow on the right-hand side of the HTTP headers subsection to display the + Add header button. In the Header field, enter Authorization. In the Value field, provide the value Bearer [MY-TOKEN], where [MY-TOKEN] is your CoreWeave API Access Token. Configure the HTTP method to use GET by changing the HTTP dropdown method in the Other subsection to GET:

​

Configure the CoreWeave Logs data source

After the configuration for the CoreWeave Metrics data source is complete, repeat the preceding steps for the CoreWeave Logs data source. With both data sources configured, your self-hosted Grafana instance can query CoreWeave metrics and logs.

​

Learn more

For more information, explore the following resources:

• Custom GPU utilization alerts to create Grafana-managed alert rules that monitor GPU utilization across your cluster.
• Grafana documentation for guidance on tasks such as building dashboards, working with data sources, and exploring metrics.
• Prometheus Query Language (PromQL) documentation to learn how to query the CoreWeave Metrics data in both ad-hoc Explore queries and advanced dashboard panels.