kubernetes-deploy-qna.md

What will pai do on your host when deploying or adding new k8s nodes?

• If there is no docker on the target host, pai will install the latest docker-ce.
  • The Script Link
• Change the host file in the path /etc/hosts
  • The Script Link
  • Because hadoop service will be deployed on hostnetworking environment. And some configuration of hosts file will have a bad influence on hadoop’s name resolve behavior.
• Prepare kubelet environment and start kubelet through docker
  • Prepare kubelet environment accoding to node role
  • The Python Module To Deploy PAI
  • The Environment Preparation Script Link
  • The Starting Script Link

Kubernetes Disk Pressure Status

• Why does this issue happen on your cluster?

  • Because the usage of the disk has reach the limit which we set through kubelet.
  • To change the threshold according to your environment. You should change the field the line --eviction-hard="memory.available<5%,nodefs.available<5%,imagefs.available<5%,nodefs.inodesFree<5%,imagefs.inodesFree<5%" in the kubelet.sh.template

• Some useful references:

  • Configuring kubelet Garbage Collection
  • Configure Out Of Resource Handling

Unable to start kubelet

1. Please ensure whether you could access to gcr.io or not. If you couldn't access to gcr.io, you will failed to pull kubernetes image.

   • In the kubernetes-configuration.yaml, you could find a field docker-registry. This field is set the docker registry used in the k8s deployment. To use the official k8s Docker images, set this field to gcr.io/google_containers, the deployment process will pull Kubernetes component's image from gcr.io/google_containers/hyperkube. But if you can't access to it, you can also set the docker registry to docker.io/openpai, which is maintained by pai.

2. Known Issue 813