Monitor and debug

After launching your Ray application, you can rely on Anyscale's observability tools to monitor performance and debug issues. This section provides an introduction to the available dashboards and debugging tools.

Monitoring

In the top navigation menu, you find tabs for a few different dashboards. You may already be familiar with the Ray Dashboard, which you can use to view the cluster's state and the progress of submitted Ray Jobs. Anyscale's dashboards expand upon the Ray Dashboard to provide more details on your clusters and workloads.

The usual starting point for monitoring is the Metrics tab. Here you find hardware metrics like the number of nodes and the available memory. If you want to further dive into the data, click View in Grafana to open the metrics directly in Grafana. From there, you can also build custom data visualizations to monitor your applications. To learn more about the metrics dashboard, read Metrics.

Aside from the general Metrics dashboard, Anyscale also provides dashboards for specific Ray Workloads like Ray Data and Ray Train. To learn more about these dashboards and how to use them, read Workload dashboards.

Debugging

To debug issues with your application, you can rely on the Logs view. Anyscale splits logs into two categories: Application logs and Workspace events.

Application logs show logs from the code that you execute on your cluster. Logs are categorized by component, and you can use filters or the search box to find specific logs. The logs exist in the cluster's storage, so they disappear when the cluster is terminated. To retain the logs, you can either download them while the cluster is running, or set up log ingestion.

The Workspace events pertain to the hardware of the cluster. The logs here show data like when the cluster comes online, goes offline, scaling events, and spot preemptions. Anyscale retains the workspace events even after the cluster terminates.

To learn more about Anyscale's logs, read Accessing logs.