Platform For AI:View EAS events in CloudMonitor

Use the console

1. Log on to the CloudMonitor console.

2. In the left navigation pane, choose Event Center > System Event.

3. On the Event Monitoring tab, select PAI from the product drop-down list, and click Search.

   

4. In the Actions column of the target event, click Details to view the event details. Example:

   

Event parameters

{
  "serviceName": "my-inference-service",
  "serviceId": "eas-cn-shanghai-1234567",
  "serviceGroup": "default",
  "resourceType": "DedicatedResource",
  "instanceType": "ecs.gn6i-c4g1.xlarge",
  "cpu": 4,
  "memory": 16384,
  "gpu": 1,
  "gpuMemory": 16,
  "nvidiaName": "Tesla T4",
  "role": "Standard",
  "isBurst": false,
  "isSpot": false,
  "callerUid": "123456789012345678",
  "timestamp": 1711929600000,
  "restartCount": 2,
  "exitCode": "1",
  "status": "CrashLoopBackOff",
  "reason": "OOMKilled",
  "message": "Container killed due to out-of-memory condition"
}

Use the API

Call DescribeSystemEventAttribute to retrieve EAS events programmatically.

Use the console

1. Create a system event-triggered alert rule. Configure the following parameters:

   • Product type: Select PAI.

   • Event type: Select ServiceInstance or Service.

   • Event level: Select one or more event levels.

   • Event name: Select one or more event names from the Name column in Supported EAS events.

   • Keyword filtering: Enter keywords to match against event content and filter which events trigger the alert.

   

2. (Optional) Use alert callbacks for system events (legacy) to send event data to an external endpoint, such as a webhook or notification service.

Use the API

Create an event-triggered alert rule and enable it.

Does a service instance refer to an inference service or a pod?

A service instance refers to a pod instance, not the inference service itself. The Service event type covers service-level changes (such as replica count changes or service unavailability), while the ServiceInstance event type covers pod-level events (such as pod restarts, image pull failures, or preemption).