Related issue
https://github.com/longhorn/longhorn/issues/2207
Test step
- Deploy a cluster that each node has different CPUs.
- Launch Longhorn v1.1.0.
- Deploy some workloads using Longhorn volumes.
- Upgrade to the latest Longhorn version. Validate:
- all workloads work fine and no instance manager pod crash during the upgrade.
- The fields
node.Spec.EngineManagerCPURequest
andnode.Spec.ReplicaManagerCPURequest
of each node are the same as the settingGuaranteed Engine CPU
value in the old version * 1000. - The old setting
Guaranteed Engine CPU
is deprecated with an empty value.
- Modify new settings
Guaranteed Engine Manager CPU
andGuaranteed Replica Manager CPU
. Validate all workloads work fine and no instance manager pod restart. - Scale down all workloads and wait for the volume detachment.
- Set
node.Spec.EngineManagerCPURequest
andnode.Spec.ReplicaManagerCPURequest
to 0 for some node. Verify the new settings will be applied to those node and the related instance manager pods will be recreated with the CPU requests matching the new settings. - Scale up all workloads and verify the data as well as the volume r/w.
- Do cleanup.