What Should I Do If a Workload Remains in the Creating State?

Symptom

The workload remains in the creating state.

Troubleshooting

Possible causes are listed in order of likelihood.

Check these causes one by one until you find the cause of the fault.

• Check Item 1: Whether the cce-pause Image Is Deleted by Mistake
• Check Item 2: Modifying Node Specifications After the CPU Management Policy Is Enabled in the Cluster

Check Item 1: Whether the cce-pause Image Is Deleted by Mistake

Symptom

When creating a workload, you may see an error indicating that the sandbox cannot be created. This is because the cce-pause:3.1 image fails to be pulled.

Failed to create pod sandbox: rpc error: code = Unknown desc = failed to get sandbox image "cce-pause:3.1": failed to pull image "cce-pause:3.1": failed to pull and unpack image "docker.io/library/cce-pause:3.1": failed to resolve reference "docker.io/library/cce-pause:3.1": pulling from host **** failed with status code [manifests 3.1]: 400 Bad Request

Possible Cause

The image is a system image added during node creation. If the image is deleted by mistake, the workload cannot be created.

Solution

1. Log in to the faulty node.
2. Decompress the cce-pause image installation package.

   tar -xzvf /opt/cloud/cce/package/node-package/pause-*.tgz

3. Import the image.
   • Docker nodes:

     docker load -i ./pause/package/image/cce-pause-*.tar

   • containerd nodes:

     ctr -n k8s.io images import --all-platforms ./pause/package/image/cce-pause-*.tar

4. Lock the image to prevent accidental deletion.
   • Docker nodes do not support image deletion protection.
   • containerd nodes:

     ctr -n k8s.io images label docker.io/library/cce-pause:{xxx} io.cri-containerd.pinned=pinned

     

     

     Run crictl images | grep pause to obtain the cce-pause image tag. As shown above, the image tag is 3.1.

     This operation requires containerd version 1.7.0 or later on the node. To check the version, run containerd version.

     

5. Create a workload.

Check Item 2: Modifying Node Specifications After the CPU Management Policy Is Enabled in the Cluster

The kubelet option cpu-manager-policy defaults to static. This allows granting enhanced CPU affinity and exclusivity to pods with certain resource characteristics on the node. If you modify CCE node specifications on the ECS console, the original CPU information does not match the new CPU information. As a result, workloads on the node cannot be restarted or created.

1. Log in to the CCE node (ECS) and delete the cpu_manager_state file.

   Example command for deleting the file:

   rm -rf /mnt/paas/kubernetes/kubelet/cpu_manager_state

2. Restart the node or kubelet. The following is the kubelet restart command:

   systemctl restart kubelet

   Verify that workloads on the node can be successfully restarted or created.

   For details, see What Should I Do If I Fail to Restart or Create a Workload on a Node After Modifying the Node Specifications?