You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Short answer: Safe to ignore — it's a false positive
What's actually happening
The MAC 0a:58:0a:80:00:02 decodes like this:
0a:58 = OVN-Kubernetes's hardcoded prefix for its management port MAC addresses
0a:80:00:02 = hex for 10.128.0.2 (OpenShift's default pod network starts at 10.128.0.0/14)
OVN-Kubernetes deterministically derives the ovn-k8s-mp0 MAC from the interface's IP address using the formula: 0a:58 + IP-in-hex. So any OCP node whose ovn-k8s-mp0 gets assigned 10.128.0.2 will always have this exact MAC.
Both ocp1-cp-1 and ocp11-cp-0 are running nested OCP clusters with the same default pod CIDR (10.128.0.0/14). Inside each VM, OVN-Kubernetes independently assigns 10.128.0.2 to its first ovn-k8s-mp0 interface — producing the same MAC in two separate VMs.
Why KubeMacPool is wrong to flag this
KubeMacPool's job is to prevent collisions on the host L2 network (ISAR's pod/VM network). But ovn-k8s-mp0 is an in-guest virtual interface — it lives inside the VM's network namespace, never touches ISAR's network fabric, and is completely invisible to anything outside the VM. The actual KubeVirt NIC MACs your inventory assigns are 0E:C0:EF:* — those are what KubeMacPool should care about.
KubeMacPool is scanning VMI status and picking up guest-internal OVN-K8s interfaces it has no business managing. This is a known false positive in nested OCP-on-OCP-Virt setups.
Seems there is already a very releated bug existing: https://redhat.atlassian.net/browse/CNV-95971
I added a comment and will go ahead and silence this alert for our infra-projects.
Activity
DanielFroehlich commentedon Sep 1, 2026
Following runbook procedure: https://github.com/openshift/runbooks/blob/master/alerts/openshift-virtualization-operator/KubemacpoolMACCollisionDetected.md
mac-Adresses:
Lets take a look at one:
Looking at ocp1-cp-1, thats the ovn-k8s-mp0 bridge with that MAC address.
Same on ocp11-cp-0.
Strange. Asking claude for help...
DanielFroehlich commentedon Sep 1, 2026
Claude analysis:
Short answer: Safe to ignore — it's a false positive
What's actually happening
The MAC 0a:58:0a:80:00:02 decodes like this:
OVN-Kubernetes deterministically derives the ovn-k8s-mp0 MAC from the interface's IP address using the formula: 0a:58 + IP-in-hex. So any OCP node whose ovn-k8s-mp0 gets assigned 10.128.0.2 will always have this exact MAC.
Both ocp1-cp-1 and ocp11-cp-0 are running nested OCP clusters with the same default pod CIDR (10.128.0.0/14). Inside each VM, OVN-Kubernetes independently assigns 10.128.0.2 to its first ovn-k8s-mp0 interface — producing the same MAC in two separate VMs.
Why KubeMacPool is wrong to flag this
KubeMacPool's job is to prevent collisions on the host L2 network (ISAR's pod/VM network). But ovn-k8s-mp0 is an in-guest virtual interface — it lives inside the VM's network namespace, never touches ISAR's network fabric, and is completely invisible to anything outside the VM. The actual KubeVirt NIC MACs your inventory assigns are 0E:C0:EF:* — those are what KubeMacPool should care about.
KubeMacPool is scanning VMI status and picking up guest-internal OVN-K8s interfaces it has no business managing. This is a known false positive in nested OCP-on-OCP-Virt setups.
DanielFroehlich commentedon Sep 1, 2026
@rbo do you agree? I would silence this alert for the stormshift-*infra namespaces
rbo commentedon Sep 7, 2026
@DanielFroehlich Agree!
IMHO it's also an kind of a bug, because with openshift on openshift we are not the only who hits the alert.
DanielFroehlich commentedon Sep 7, 2026
Seems there is already a very releated bug existing:
https://redhat.atlassian.net/browse/CNV-95971
I added a comment and will go ahead and silence this alert for our infra-projects.
DanielFroehlich commentedon Sep 7, 2026
Alert is silenced now:
https://console-openshift-console.apps.isar.coe.muc.redhat.com/monitoring/silences/f771e774-57f1-4d30-9d95-438c51f2f4a9
Downside: cant be silence per project regexp - is global now. We need to remember to remove this once the fix is delivered.