Changelog

Deprecation

NodeInfoProcessor interface has been deprecated and will be removed in version 1.24. (#4191)
- The interface is superseded by a new TemplateNodeInfoProvider interface.
- This deprecation is only relevant for users who build customized images of Cluster Autoscaler. There is no change in behavior when using the official CA image.

--expander flag now accepts a comma-separated list of expanders to use. Additional expanders will be used as fallback if the first expander selects multiple NodeGroups (#4233).
Added --feature-gates flag that allows specifying which Kubernetes feature gates should be enabled for the kube-scheduler code embedded by CA (#4539).
Fixed a bug where Kubernetes events were not published reliably for cluster-autoscaler-status configmap. The configmap should now have attached events for every scale-up (including failed scale-up attempts) and scale-down (#4328).
Improved CA performance in clusters with large number of pods using projected volumes (#4441).

Note: The merging of #4468 means that the default configuration of the CA will now require an extra IAM permission - ec2:DescribeInstanceTypes

Instance type list generation moved from JSON parsing to DescribeInstanceTypes API (#4468)
Scale-down options can now be configured per-ASG (#4238).
Reduced the number of API calls made by CA (#4073).
Added support for NVIDIA A100 GPU.

The CAPI_GROUP environment variable will now also affect the annotations keys for minimum and maximum node group size, the machine annotation, machine deletion, and the cluster name label (#4451).

Scale-down options can now be configured per-MIG (#4236).
SSD ephemeral-storage can now be specified via instance template, allowing scale-from-0 for pods requesting it (#4318).
Reduced the number of GCE API calls made by CA (#4423).