-
Bug
-
Resolution: Duplicate
-
Critical
-
None
-
4.12.z, 4.14.z, 4.16.z
-
None
-
Critical
-
None
-
False
-
Description of problem:
Currently IngressOperator support nodePlacement, but with just "nodeSelectors" and "tolerations" options. In a cloud scenario there is the need for a customer to leverage on affinity options to better spread workload among all the region/zones. If we have a 3 worker nodes cluster with a replica of 3 router pods, and a node (zoneA) suddenly goes offline there will be an unbalanced scenario with more router pods per node (es. 2 on zoneB and 1 on ZoneC), despite the node (zoneA) that previously had gone offline come back on line again, so leaving a zone without any router pod and having so a less high availability behaviour in place.
Version-Release number of selected component (if applicable):
How reproducible:
In a cluster with: 3 worker nodes (each for zone/region) 3 router replicas (each for zone/region) if one node goes down there will be an unbalanced scenario where two pods will remain on one node despite the node that suddenly was disappeared will get back online again.
Steps to Reproduce:
1. 2. 3.
Actual results:
Expected results:
Additional info: