Azure Machine Learning

Azure Machine Learning is a cloud service for accelerating and managing the machine learning project lifecycle which includes training, deploying, and monitoring models. This integration helps in monitoring the training runs, model deployments, and resource utilization of your nodes present in your workspace.

Log and metric types

For Azure Machine Learning, you can collect the following logs and metrics:

• Resource logs. These logs contains information about clusters, nodes, jobs, deployments, models, and pipeline events. To learn more about the different log category types and schemas collected for Azure Machine Learning, refer to the documentation.
• Platform Metrics for Azure Machine Learning. These metrics are available in Microsoft.MachineLearningServices/workspaces namespace. For more information on supported metrics and dimensions, refer to the Azure documentation.

Setup

Azure service sends monitoring data to Azure Monitor, which can then stream data to Eventhub. Sumo Logic supports:

• Logs collection from Azure Monitor using our Azure Event Hubs source.
• Metrics collection using our HTTP Logs and Metrics source via Azure Functions deployed using the ARM template.

You must explicitly enable diagnostic settings for each machine learning workspace you want to monitor. You can forward logs to the same event hub provided they satisfy the limitations and permissions as described here.

When you configure the event hubs source or HTTP source, plan your source category to ease the querying process. A hierarchical approach allows you to make use of wildcards. For example: Azure/MachineLearning/Logs, Azure/MachineLearning/Metrics.

Configure metrics collection

In this section, you will configure a pipeline for shipping metrics from Azure Monitor to an Event Hub, on to an Azure Function, and finally to an HTTP Source on a hosted collector in Sumo Logic.

1. Configure an HTTP Source.
2. Configure and deploy the ARM Template.
3. Export metrics to Event Hub. Perform below steps for each machine learning workspace that you want to monitor.
   • Choose Stream to an event hub as destination.
   • Select AllMetrics.
   • Use the Event hub namespace created by the ARM template in Step 2 above. You can create a new Event hub or use the one created by ARM template. You can use the default policy RootManageSharedAccessKey as the policy name.

Configure logs collection

In this section, you will configure a pipeline for shipping diagnostic logs from Azure Monitor to an Event Hub.

1. To set up the Azure Event Hubs cloud-to-cloud source in Sumo Logic portal, refer to our Azure Event Hubs source documentation.
2. To create the Diagnostic settings in Azure portal, refer to the Azure documentation. Perform below steps for each machine learning workspace that you want to monitor.
   • Choose Stream to an event hub as the destination.
   • Select allLogs or individual log categories.
   • Use the Event hub namespace and Event hub name configured in previous step in destination details section. You can use the default policy RootManageSharedAccessKey as the policy name.

Troubleshooting

Azure Event Hubs Source

Common error types are described here.

You can try restarting the source for ThirdPartyConfig errors.

HTTP Logs and Metrics Source used by Azure Functions

To troubleshoot metrics collection, follow the instructions in Collect Metrics from Azure Monitor > Troubleshooting metrics collection.