Workload dashboards
This feature is in a beta release and is in active development. Reach out with any feedback or suggestions.
Workload dashboards are workload-specific dashboards that aid with debugging Ray workloads. They are exclusive to Anyscale. Access them by clicking the Ray Workloads tab in the Jobs or Workspaces page.
The Ray Workloads tab has multiple sub-tabs, each of which is dedicated to a specific type of workload. To see more information about these dashboards, view the other docs in this section.
Dashboards
- Data Dashboard: For debugging Ray Data workloads.
- Train Dashboard: For debugging Ray Train workloads.
Ray sessions
Anyscale persists workload dashboards so that you can debug the workload even after the cluster has terminated. The data is organized by Ray Session, which is a Ray concept for an instance of a Ray cluster. Ray creates new session whenever it launches. The ID of the session is based on the timestamp of when the cluster was launched.
Select the session of the workload dashboard in the dropdown in the top right corner. Anyscale supports viewing data from the 10 most recent sessions.
Stale data
If a Ray cluster terminates unexpectedly, the workloads dashboard may show stale data. This behavior occurs because the dashboard persists events received by the Ray cluster and the cluster may not emit a termination event if it terminates unexpectedly. In this case, the dashboard shows the last known state of the cluster.