View a markdown version of this page

Ray Grafana dashboards - Amazon SageMaker AI
Filtering a dashboard

Ray Grafana dashboards

The SageMaker HyperPod Observability add-on provisions Grafana dashboards for your Ray workloads. You get them without importing dashboard JSON yourself.

The add-on provisions the following dashboards:

  • Ray Core — cluster resources, tasks, actors, and node state.

  • Ray Data — dataset throughput and operator progress.

  • Ray Train — training run progress and worker state.

  • Ray Serve — request rate, latency, and replica state.

Filtering a dashboard

Each dashboard exposes filters so you can scope a view to one workload:

  • Cluster name — the HyperPod cluster, backed by the cluster_name label

  • Cluster ID — the HyperPod cluster, backed by cluster_id

  • Namespace — backed by namespace

  • Ray cluster — the Ray cluster, backed by ray_io_cluster

Set the filters to a running cluster to see its metrics. For more information about opening Grafana and access roles, see Observability for Amazon SageMaker HyperPod cluster orchestrated by Amazon EKS.