Introduction

Azure AI (Azure AI Foundry / Azure Machine Learning) is a managed platform for building, deploying, and operating machine learning models and generative AI applications. OpsRamp discovers and monitors Azure AI resources, including model deployments, online endpoints, online endpoint deployments, and workspace connections. It also collects Azure Monitor metrics from the Microsoft.MachineLearningServices namespace, providing visibility into resource health, performance, and usage.

Use the Azure Public Cloud integration to discover Azure AI resources and collect associated performance and operational metrics.

Setup

Before discovering Azure AI resources, ensure that the OpsRamp Microsoft Azure Integration is configured.

To enable Azure AI resource discovery, select Azure AI as a resource type in the Azure Integration Discovery Profile.

OpsRamp collects Azure Monitor metrics for the following resource types:

  • Microsoft.MachineLearningServices/workspaces/onlineEndpoints
  • Microsoft.MachineLearningServices/workspaces/onlineEndpoints/deployments

Ensure that the integration application is assigned the Monitoring Reader role on the target Azure subscription(s).

Supported resource types

Resource Type (Native Type)Resource TagDescription
Azure AI Model DeploymentAZ_AI_DEPLOYMENTModel deployments within an Azure AI and ML workspace.
Azure AI Online EndpointAZ_AI_ONLINE_ENDPOINTOnline (real-time) inference endpoints.
Azure AI Online Endpoint DeploymentAZ_AI_ONLINE_DEPLOYMENTDeployments serving models behind an online endpoint.
Azure AI Workspace ConnectionAZ_AI_CONNECTIONConnections from a workspace to external resources and services.

Supported metrics

Online Endpoint

OpsRamp MetricAzure MetricMetric Display NameUnitAggregation TypeDescription
azure_machinelearningservices_workspaces_onlineendpoints_ConnectionsActiveConnectionsActiveConnections ActivecountAverageNumber of active connections.
azure_machinelearningservices_workspaces_onlineendpoints_DataCollectionErrorsPerMinuteDataCollectionErrorsPerMinuteData Collection Errors Per MinutecountAverageNumber of data collection errors per minute.
azure_machinelearningservices_workspaces_onlineendpoints_DataCollectionEventsPerMinuteDataCollectionEventsPerMinuteData Collection Events Per MinutecountAverageNumber of data collection events per minute.
azure_machinelearningservices_workspaces_onlineendpoints_NetworkBytesNetworkBytesNetwork BytesBpsAverageNetwork throughput in bytes per second.
azure_machinelearningservices_workspaces_onlineendpoints_NewConnectionsPerSecondNewConnectionsPerSecondNew Connections Per Secondcounts/sAverageNew connections established per second.
azure_machinelearningservices_workspaces_onlineendpoints_RequestLatencyRequestLatencyRequest LatencymsAverageRequest latency.
azure_machinelearningservices_workspaces_onlineendpoints_RequestLatency_P50RequestLatency_P50Request Latency P50msAverage50th percentile request latency.
azure_machinelearningservices_workspaces_onlineendpoints_RequestLatency_P90RequestLatency_P90Request Latency P90msAverage90th percentile request latency.
azure_machinelearningservices_workspaces_onlineendpoints_RequestLatency_P95RequestLatency_P95Request Latency P95msAverage95th percentile request latency.
azure_machinelearningservices_workspaces_onlineendpoints_RequestLatency_P99RequestLatency_P99Request Latency P99msAverage99th percentile request latency.
azure_machinelearningservices_workspaces_onlineendpoints_RequestsPerMinuteRequestsPerMinuteRequests Per MinutecountAverageNumber of requests per minute.

Online Endpoint Deployment

OpsRamp MetricAzure MetricMetric Display NameUnitAggregation TypeDescription
azure_machinelearningservices_workspaces_onlineendpoints_deployments_CpuMemoryUtilizationPercentageCpuMemoryUtilizationPercentageCPU Memory Utilization Percentage%AveragePercentage of memory used on an instance.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_CpuUtilizationPercentageCpuUtilizationPercentageCPU Utilization Percentage%AveragePercentage of CPU utilization on an instance.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_DataCollectionErrorsPerMinuteDataCollectionErrorsPerMinuteData Collection Errors Per MinutecountAverageThe number of data collection events dropped per minute.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_DataCollectionEventsPerMinuteDataCollectionEventsPerMinuteData Collection Events Per MinutecountAverageThe number of data collection events processed per minute.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_DeploymentCapacityDeploymentCapacityDeployment CapacitycountAverageThe number of instances in the deployment.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_DiskUtilizationDiskUtilizationDisk Utilization%AveragePercentage of disk utilization on an instance.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_GpuEnergyJoulesGpuEnergyJoulesGPU Energy in JoulescountAverageEnergy consumed on a GPU node during the interval, in joules.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_GpuMemoryUtilizationPercentageGpuMemoryUtilizationPercentageGPU Memory Utilization Percentage%AveragePercentage of GPU memory utilization on an instance.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_GpuUtilizationPercentageGpuUtilizationPercentageGPU Utilization Percentage%AveragePercentage of GPU utilization on an instance.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_RequestLatency_P50RequestLatency_P50Request Latency P50msAverageThe average P50 request latency.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_RequestLatency_P90RequestLatency_P90Request Latency P90msAverageThe average P90 request latency.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_RequestLatency_P95RequestLatency_P95Request Latency P95msAverageThe average P95 request latency.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_RequestLatency_P99RequestLatency_P99Request Latency P99msAverageThe average P99 request latency.
azure_machinelearningservices_workspaces_onlineendpoints_deployments_RequestsPerMinuteRequestsPerMinuteRequests Per MinutecountAverageNumber of requests sent to the online deployment within one minute.

External reference