Help Center/ ModelArts/ Troubleshooting/ Inference Deployment (New Version)/ Service Deployment/ Abnormal Real-Time Service Deployment Instances
Updated on 2026-09-02 GMT+08:00
View PDF
Share

Abnormal Real-Time Service Deployment Instances

Symptom

The deployment status of the real-time service changes to CONCERNING or ERROR due to abnormal instances.

  1. Log in to the ModelArts console. In the navigation pane, choose Model Inference > Real-Time Inference. In the service list, click the target service name to go to the service details page.
  2. View events in the Events tab.

    An abnormal event occurred in service events: "Deployment [XXX] status change from RUNNING to CONCERNING" or "Deployment [XXX] status change from RUNNING to ERROR."

    • The error information in the event contains "Available instances decrease to N."
      Figure 1 CONCERNING

      Figure 2 ERROR

    • The error information in the event contains "The resource pool has been unmanaged."
      Figure 3 The resource pool has been unmanaged

    • The error information in the event contains "no instance available."
      Figure 4 No instance available

Cause Analysis and Solution

Table 1 Cause analysis and solution of abnormal deployment instances

Message

Possible Cause

Solution

Available instances decrease to N

Faulty resource pool nodes may reduce the number of deployed instances.

Check the resource pool node status. For details, see Viewing Resource Pool Nodes.

If you delete the instance manually, the pod might not schedule properly and could prevent restarting the instance.

The pod Events tab appears on the right side of the Service Events page. Click it to check pod event details and identify why the pod failed to start.

The health check may fail.

The resource pool has been unmanaged

The Model Deploy option has been removed from the resource pool's job types. This change makes the resource pool unmanaged and causes all deployments within it to show abnormal statuses.

Modify the job types supported by the resource pool. Add Model Deploy to the resource pool's job types. Re-manage the resource pool. For details, see Modifying the Job Types Supported by a Dedicated Resource Pool.

Then, stop and start the service.

No instance available

When you power off the resource pool node, its operator stops working, and all pods running on that node become abnormal.

Power on the resource pool node. After that, its operator and the pods deployed on the node will automatically recover.
Parent topic: Service Deployment