Upgrade from 1.5.2 or 1.5.3 to 1.5.4 (ECS)
You can upgrade your existing CDP Private Cloud Data Services version 1.5.2 or 1.5.3 to 1.5.4 without performing uninstalling the previous version.
- Review the Software Support Matrix for ECS.
- The Docker registry that is configured with the cluster must remain the same during the upgrade process. If CDP Private Cloud Data Services 1.5.2 or 1.5.3 was installed using the public Docker registry, CDP Private Cloud Data Services 1.5.4 should also use the public Docker registry, and not be configured to use the embedded Docker registry. To use a different configuration for the Docker registry, you must perform a new installation of CDP Private Cloud Data Services.
- If the upgrade stalls, do the following:
- Check the status of all pods by running the following command on the ECS server
node:
export PATH=$PATH:/opt/cloudera/parcels/ECS/installer/install/bin/linux/:/opt/cloudera/parcels/ECS/docker export KUBECONFIG=~/kubeconfig kubectl get pods --all-namespaces
- If there are any pods stuck in "Terminating" state, then force terminate the pod
using the following
command:
kubectl delete pods <NAME OF THE POD> -n <NAMESPACE> --grace-period=0 —force
If the upgrade still does not resume, continue with the remaining steps.
- If there are any pods in the "Pending" state, then schedule the pods in the "Pending
state" by running the following commands:
kubectl get pods -n yunikorn kubectl get deploy -n yunikorn kubectl scale --replicas=0 -n yunikorn deployment/yunikorn-scheduler kubectl get deploy -n yunikorn kubectl scale --replicas=1 -n yunikorn deployment/yunikorn-scheduler kubectl get deploy -n yunikorn
- In the Cloudera Manager Admin Console, go to the ECS service and click
The Longhorn dashboard opens.
. -
Check the "In Progress" section of the dashboard to see whether there are any volumes stuck in the attaching/detaching state in. If a volume is that state, reboot its host.
- In the LongHorn UI, go to the Volume tab and check if any of the volumes are in the "Detached" state. If any are in the "Detached" state, then restart the associated pods or reattach them to the host manually.
- Check the status of all pods by running the following command on the ECS server
node:
- You may see the following error message during the Upgrade Cluster >
Reapplying all settings > kubectl-patch
:
If you see this error, do the following:kubectl rollout status deployment/rke2-ingress-nginx-controller -n kube-system --timeout=5m error: timed out waiting for the condition
- Check whether all the Kubernetes nodes are ready for scheduling. Run the following
command from the ECS Server
node:
You will see output similar to the following:kubectl get nodes
NAME STATUS ROLES AGE VERSION <node1> Ready,SchedulingDisabled control-plane,etcd,master 103m v1.21.11+rke2r1 <node2> Ready <none> 101m v1.21.11+rke2r1 <node3> Ready <none> 101m v1.21.11+rke2r1 <node4> Ready <none> 101m v1.21.11+rke2r1
-
Run the following command from the ECS Server node for the node showing a status of
SchedulingDisabled
:
You will see output similar to the following:kubectl uncordon
<node1>node/<node1> uncordoned
- Scale down and scale up the rke2-ingress-nginx-controller pod
by running the following command on the ECS Server
node:
kubectl delete pod rke2-ingress-nginx-controller-<pod number> -n kube-system
- Resume the upgrade.
- Check whether all the Kubernetes nodes are ready for scheduling. Run the following
command from the ECS Server
node:
- After upgrading, the Cloudera Manager admin role may be missing the Host Administrators
privilege in an upgraded cluster. The cluster administrator should run the following
command to manually add this privilege to the
role.
ipa role-add-privilege <cmadminrole> --privileges="Host Administrators"
- If you specified a custom certificate, select the ECS cluster in Cloudera Manager, then
select
cert.pem
and key.pem files from the Cloudera Manager server host to the ECS Management Console host.
. This command copies the - After upgrading, you can enable the unified time zone feature to synchronize the ECS cluster time zone with the Cloudera Manager Base time zone. When upgrading from earlier versions of CDP Private Cloud Data Services to 1.5.2 and higher, unified time zone is disabled by default to avoid affecting timestamp-sensitive logic. For more information, see ECS unified time zone.