Import PAI-EAS inference service and resource group metrics into Managed Service for Prometheus through the CloudMonitor 2.0 Integration Center. Query metrics with PromQL and connect to external monitoring systems such as Grafana.
Billing
Enabling PAI-EAS Inference Service Metrics and PAI-EAS Resource Group Metrics in the Integration Center of CloudMonitor 2.0 incurs the following charges:
Step 1: Enable EAS monitoring metrics
Log on to the CloudMonitor 2.0 console, select the target workspace, and then click Integration Center in the left-side navigation pane.
On the Integration Center page, search for "PAI-EAS" in the search box and click the result card.
Select PAI-EAS Inference Service Metrics and PAI-EAS Resource Group Metrics as the data sources.
ImportantAdvanced monitoring metrics include inference framework metrics (vLLM or SGLang), GPU compute metrics, PAI-EAS gateway metrics, service-level and tenant-level statistics, and inference application custom metrics (prefixed with
custom_). For details, see EAS service and resource group metrics. Supported regions include Beijing, Shanghai, Hangzhou, Ulanqab, Singapore, and Heyuan. To enable additional regions, submit a ticket.Enable advanced monitoring metrics to build Grafana dashboards or configure custom alerts.
The integration runs automatically in the background and takes about 1–2 minutes. Click Integration Management to view the integration status and confirm that the integration is complete.
Step 2: Get Prometheus data source information
CloudMonitor 2.0 creates a Prometheus instance to store EAS monitoring metrics. The default storage location is RegionShare:{{workspaceName}}:{{regionId}}, where {{workspaceName}} is the workspace name and {{regionId}} is the region ID.
To retrieve monitoring data through the Prometheus API, get the region and access URL of this Prometheus instance.
Switch to the Integration Management tab. In the integration configuration list, click the expand icon next to the integration that you want to view.
In the expanded items, find PAI-EAS Inference Service Metrics or PAI-EAS Resource Group Metrics to view the integration component, status, and region.
In the Actions column, click Data Source to view the region and access URL of the Prometheus service.
Step 3: Query and explore metrics
Explore EAS monitoring metrics through the Metrics Explorer or PromQL queries in the Prometheus service.
CloudMonitor metric names use the
AliyunLearn_easprefix. They correspond to the EAS metric definitions in CloudMonitor but include richer tag information.For advanced monitoring metrics, see Advanced EAS monitoring metrics.
On the Data Source page from Step 2, click the instance ID to go to the Prometheus Service details page.
On the Metrics Explorer tab, view the metric details for EAS services.
Method
Description
Filter metrics to view details
On the Metrics Explorer tab, click Metric Explorer in the query builder.
Enter a metric keyword such as
AliyunLearnin the search box, and then select the target metric from the drop-down list, such as CPU core usage, CPU utilization, or GPU memory utilization.In the Actions column, click the explore icon to set filter conditions, and then click the add icon to add the metric to the query box.
Click Run Query to view the metric trend chart.
Query metrics with PromQL expressions.
For example, to query the total QPS across all services:
Enter
sum(AliyunLearn_eas_qps_total). (For PromQL syntax, see Metric data query and analysis syntax)Click Run Query to view the trend of the total QPS across all EAS services in the current region.
Step 4: Use Grafana dashboards
View the default Grafana dashboard
CloudMonitor 2.0 provides a default Grafana dashboard:
In the left-side navigation pane of the Prometheus service details page, click Dashboard List.
Click a dashboard name to view the built-in Grafana dashboard.
You can also switch to the Dashboard tab from the integration policy in Provisioning.
Add a panel
Add a global QPS panel to the default Grafana dashboard. For more information about Grafana, see Managed Service for Grafana.
On the dashboard details page, click the Add panel button in the upper-right corner
, and then click Add a new panel in the dialog that appears.On the right side of the Edit Panel page, switch the chart type to Stat.
In the lower-left corner of the page, switch Data source to ${datasource}.
In the Query area, switch to code on the right side. In the Metrics browser text box, enter the PromQL query
sum(AliyunLearn_eas_eas_qps_total)and then click Run queries.Adjust the thresholds to configure different display colors for different value ranges. After the chart preview appears, click Apply to save the settings.
Step 5: Configure alert rules
On the Prometheus instance details page, you can view the built-in alert rules. All built-in alert rules are P2-level and initially Stopped.
In the left-side navigation pane of the Prometheus service details page, click Alert Rules.
On the Alert Rules page, view the preset alert rules. To enable a rule, modify its Enabled/Disabled status. To modify the notification configuration, click Edit.
If the default templates don't meet your needs, click Create Alert Rule to create a custom alert rule. For parameter descriptions, see Alert rules.
Appendix: Advanced EAS monitoring metrics
The following metrics are available only when advanced monitoring metrics are enabled in Step 1.