-
Bug
-
Resolution: Unresolved
-
Major
-
None
-
4.21
-
None
Description of problem:
nodes in 3 node pools don't scale up evenly with balancingIgnoredLabels set
Version-Release number of selected component (if applicable):
How reproducible:
Using the following autoscaling set up for the hostedcluster to test the autoscaling balancing feature:
{
"autoscaling": {
"scaling": "ScaleUpOnly",
"expanders": ["Random"],
"balancingIgnoredLabels": ["common-label-in-3-nodepools"],
"maxFreeDifferenceRatioPercent": 5
}
}
Please see the following comment on the test case link.
Actual results:
The workload leads to scaling up without evenly distributed.
Expected results:
The difference between the scaling up nodes count in the nodepools must not exceed 1
Additional info:
This label of +hypershift.openshift.io/nodepool-globalps-enabled+ is added to worker node by globalps controller to indicate if it is eligible to run the global-pull-secret-syncer daemonset, which are Replace nodes only. autoscaler does not care about Replace or InPlace.
Also, it takes time to add the label to the worker nodes after they are ready, in some cases, the autoscaler gets triggered before this label gets added.
So we need to ignore it for cluster-autoscaling balancing by default.