-
Bug
-
Resolution: Unresolved
-
Major
-
None
-
4.16
-
None
-
Quality / Stability / Reliability
-
False
-
-
8
-
Critical
-
None
-
None
-
None
-
CORENET Sprint 279
-
1
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Description of problem:
OVN-Kubernetes tries to assign more EgressIPs to a node than available capacity.
Version-Release number of selected component (if applicable):
4.16.46
How reproducible:
How reproducible:
Intermittent, but consistently reproducible when:
- Cluster has multiple nodes labeled with k8s.ovn.org/egress-assignable.
- Workload requires more egress IPs than a single node can support (e.g., > X secondary IPs on AWS).
- CNCC continues to assign new CloudPrivateIPConfig to the saturated node.
See OCPBUGS-60806
Actual results:
- CNCC keeps assigning new IPs to a saturated node.
- CloudPrivateIPConfig objects remain stuck in CloudResponseError.
- No automatic redistribution to other egress-assignable nodes.
Expected results:
- CNCC should detect that a node has reached its IP/ENI limit.
- Scheduler logic should redistribute new or failing CloudPrivateIPConfig objects to other available `egress-assignable` nodes automatically.
Additional info:
OpenShift version: 4.16.46
Cloud provider: AWS
Example error from object status:
status:
conditions:
- lastTransitionTime: "2025-08-21T09:41:59Z"
message: cloud API failed to assign IP: exceeded interface address quota
reason: CloudResponseError
status: "False"
type: Assigned
- split from
-
OCPBUGS-60806 [AWS, EgressIP] CNCC and OVN-Kubernetes are not handling 0 capacity in cloud env's correctly (0 and unset is not differentiated - 0 capacity is mistaken for Unlimited capacity by OVNK
-
- Verified
-
- links to