Help Center> Cloud Container Engine> FAQ> Workload> Workload Abnormalities> Why Do Pods Start Slowly on a CCE Turbo BMS Node?
View PDF

Why Do Pods Start Slowly on a CCE Turbo BMS Node?

Symptom

Pods start slowly on a CCE Turbo BMS node.

Fault Locating

Check the time required for binding an ENI of a pod. Replace PODName with the name of your pod.

# grep `crictl pods | grep $PODName | awk '{print $1}'`  /var/paas/sys/log/yangtse/yangtse-agent.log | grep "finished add network for container" 
I0701 15:27:21.795199 3761852 service.go:480] finished add network for container ae97a9784b88d4142068448b66b45b5b299775d7c14b17a9f82013f8bebaa587 took 25 seconds

If the time exceeds 20 seconds, the ENI used by the pod is not pre-bound. Generally, it takes about 20 seconds to bind an ENI to a BMS node in a CCE Turbo cluster. You can pre-bind ENIs to accelerate networking provisioning.

You can run the following command to check whether an ENI is pre-bound to the node:

# kubectl get pni -A | grep $NodeName | grep -w PreBound | wc -l
0

If the command output is 0, no ENIs are pre-bound. In this case, configure the preheating ratio of the ENIs by following the instructions provided in Configuring NIC Multi-Queue for BMS Nodes in the Same CCE Turbo Resource Pool.

Parent topic: Workload Abnormalities