Changes since 1.18.1
General:
- Nodes with small difference in available memory will now be considered similar for the purposes of balancing NodeGroup sizes. This should increase the reliability of NodeGroup balancing on some providers (#3124).
- Fixed a bug where Cluster Autoscaler incorrectly didn't correctly take into account resource requests of initContainers (#3224).
AWS:
- Fixed an issue where nodes with NodeWithImpairedVolumes could block scale-up.
Azure:
- Changed the label CA uses to detect nodes with GPUs from to "accelerator" (previously was "cloud.google.com/gke-accelerator").
- Various improvements to scale down scenarios.
Cluster API:
- Add autodiscovery for group and version of CAPI resources. This change removes a restriction that limited the CAPI version to v1alpha2 (#2932).
- Remove autoscaler panic conditions based on available Custom Resource Definitions. This change ensures that the autoscaler will emit error messages instead of panicing if it cannot find the CAPI resources (#3171).
- Failed Machines now have a fake Provider ID. This change ensures that the autoscaler will see these Machines as failed and not attempt to wait for them (#3172)
- Node Provider IDs are no longer normalized internally. This change improves the behavior for Nodes joining the cluster (#3175).
Images
k8s.gcr.io/autoscaling/cluster-autoscaler:v1.18.2