chore: update sigs.k8s.io/karpenter to v1.8.2#1361
Conversation
…provider-azure into tallaxes/go1.25
|
Manual run of upstream StaticCapacity passed 10/12, the failures are likely due to metric-server etc. sticky overprovisioning, but have not investigated yet |
matthchr
left a comment
There was a problem hiding this comment.
Approved - feel free to ping me if you push updates and need me to reapprove
|
Update on testing: There are three (upstream) tests failing. Two are issues with test code (in one case it loses our NodePool labels/startupTaints causing overprovisioning, another needs a bit of timing adjustment / Eventually to work for Azure provider. The third one however appears to be a real problem: The static provisioner controller does not check if NodePool is being deleted, and can keep provisioning NodeClaims. For some reason this does not surface in AWS testing (I presume). All of the above should be fixed by kubernetes-sigs/karpenter#2840; we should hold this until fixes land. |
|
Temp point to fork for validation. All three previously-failing StaticCapacity tests pass with the tallaxes/karpenter-core fork (branch tallaxes/staticcapacity-fix-v1.8.2). The fork's fixes (DeletionTimestamp check in controller.go, preserving labels/startupTaints from provider NodePool, and Eventually() wrapper in the test) are validated end-to-end against the AKS cluster. |
Description
Update
sigs.k8s.io/karpenterto v1.8.2, enable static capacity supportHow was this change tested?
make presubmitFOCUS=StaticCapacity make az-upstream-e2etests. Proper CI integration for testing of this is best done via chore: incorporate upstream E2E tests #1363Does this change impact docs?
Release Note