-
Bug
-
Resolution: Unresolved
-
Undefined
-
None
-
4.14.z, 4.15.z, 4.17.z, 4.16.z, 4.18.0
-
Moderate
-
None
-
False
-
-
Document that SR-IOV VFs included in SriovNetworkNodePolicy CRs are fully managed by the SR-IOV Network Operator, with the exception when `externallyManaged: true` is set.
As a consequence, users must not alter any VF configuration when VFs are in the root network namespace, i.e. not attached to Pods. Should users change anything, the expectation is that the SR-IOV Network Operator may reconcile (i.e. undo/revert) to the defined configuration in the SriovNetworkNodePolicy.
For example, a SriovNetworkNodePolicy is created and defines a VF partitioning with `pfNames: "netpf0#2-7"` and with `mtu: 9800`. A user changes the MTU value of any of VFs 2 to 7 in the root network namespace to a different MTU value. The SR-IOV Network Operator will detect an undesired MTU value and set MTU 9800 per the policy. As of today (Oct 22, 2024) the reconciliation logic in the SR-IOV Network Operator will trigger a node drain (reboot) which can impact workloads (SR-IOV and non-SRIOV workloads).
In order to inhibit the MTU reconciling process, the field `SriovNetworkNodePolicy.Spec.Mtu` field must be left blank.
Version-Release number of selected component (if applicable):
4.14+ (4.12 and 4.13 will soon reach end of life)
- relates to
-
OCPBUGS-42552 Pod deletion triggering a node drain
- New