-
Bug
-
Resolution: Unresolved
-
Major
-
None
-
4.19.0
-
Quality / Stability / Reliability
-
False
-
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Description of problem:
For VM with non-live migration the eviction strategy must be set to external. This setting is essential to ensure that, during a node drain on the underlying infrastructure cluster (e.g. during a cluster upgrade), the eviction request for the KubeVirt VM is visible to the Cluster API Provider for KubeVirt (CAPK) controller.Only when evictionStrategy=External, CAPK can detect the eviction and initiate a coordinated drain of the Kubernetes node inside the guest cluster before the VM is terminated.If the VM supports live migration, the CAPK controller will skip draining the guest node, allowing the VM to be moved without disruption.
Version-Release number of selected component (if applicable):
4.19.0
How reproducible:
100%
Steps to Reproduce:
1. Deploy HCP on Ovirt with third party CNI say, calico 2. Check if the VMs are created with non-live migration. 3. Shutdown the nodepool VM manually and observe the HCP pods are failing.
Actual results:
HCP functionality disrupts and pod goes to failed state
Expected results:
Node draining should happen to safely move the pods from the VM being shutdown.
Additional info:
- links to