Skip to main content

Get Started with Managed Grafana

Accessing CoreWeave's managed Grafana instance

CKS provides a managed Grafana instance with multiple pre-configured dashboards optimized to display the metrics, events, and logs most relevant to your workload.

CoreWeave's managed Grafana instance does not support the creation of new dashboards or the modification of existing ones. If your workload requires a more customized solution, consider implementing a self-hosted Grafana instance in your CKS environment.

Accessing Grafana​

To access Grafana from within the Console UI, click the bar graph icon on the far left of the screen. This button is only available to users in the admin group.

Available dashboards​

CoreWeave provides a suite of custom dashboards, available by default in a managed Grafana instance. To access these dashboards in a self-hosted Grafana instance, install the Helm Chart in the CoreWeave Charts repository.

Pod Inspector​

The Pod Inspector dashboard provides a combined view of common pod metrics, Kubernetes events, and pod logs, filtered to a specific pod or container in any namespace.

SUNK dashboards​

SUNK dashboards provide insight into the state of your Slurm cluster, including job metrics, node health, and resource consumption.

  • Slurm is a fault-tolerant, highly scalable cluster resource management and job scheduling system for large HPC jobs such as LLM model training in supercomputing centers, government laboratories, universities, and companies worldwide.
  • SUNK (Slurm on Kubernetes) is CoreWeave's implementation of Slurm deployed on Kubernetes with a Helm chart.

SLURM / Job Metrics​

The SLURM / Job Metrics dashboard displays metrics used for debugging and analyzing the performance of specific jobs within your cluster. It provides detailed information about the nodes running a given job, including alerts and states.

SLURM / Namespace​

The SLURM / Namespace dashboard provides an overview of the state of SUNK jobs within your clusters, with less focus on the specifics of a given job. It aggregates metrics by relevant properties, such as job state, resource type, and user, to illustrate trends in jobs.

Knative dashboards​

Knative dashboards provide insight into the state of your Knative workloads, including revision metrics, pod health, and resource consumption.

  • Knative is a Kubernetes-based platform to build, deploy, and manage modern serverless workloads.
  • Revisions contain point-in-time snapshots of the application code and configuration for each change made to a Knative Service.

Kourier Gateway​

The Kourier Gateway dashboard tracks the state of Kourier Gateways for Knative, with panels for traffic metrics. It provides useful insight for debugging connection and networking issues specific to Knative deployments and revisions from an ingress perspective.

Envoy Circuit Breaker​

The Envoy Circuit Breaker dashboard displays specific metrics that monitor upstream connections and overflows in the Envoy data plane. It assists with observing and troubleshooting Knative connections from a data plane perspective.

Revision dashboards​

The Knative Revision dashboards all provide similar types of Knative metrics, but at different levels of aggregation:

  • Global Revision Overview offers a high-level overview across all namespaces within a specific cluster.
  • Namespace Revision Overview displays statistics per namespace.
  • Revision Overview shows detailed statistics per Knative configuration.

Fleet Management dashboards​

Fleet Management dashboards provide insight into trends in resource consumption, state, and events for the nodes and pods in your clusters.

Node Details​

The Node Details dashboard provides a detailed breakdown of node and GPU health, both at the physical hardware level and in Kubernetes. This dashboard contains additional log panels to allow simultaneous viewing of metrics along with the log and event types commonly needed to debug or analyze a node.

Node Wrangler​

The Node Wrangler dashboard provides visibility into the overall health and state of all nodes in your clusters, along with their specs and hardware architectures.

Last updated on