Changelog

Known issues

Importing Cluster Autoscaler 1.24.0 code via go mod may lead to compilation errors (#4875). This is only relevant to other projects importing CA code and has no impact on anyone using official CA image.

CA would no longer block scale-down in entire cluster while some pods are waiting for booting up nodes. This allows scale-down in some NodeGroups to happen, while other NodeGroups are scaling-up (#4661).
Introduced gRPC expander (#4452).
Introduced debugging snapshot feature (#4552).
CA can now continue working even if it is unable to remove some of the VMs that failed to register in the cluster (#4810).
Added --node-info-cache-expire-time flag that controls how long node templates are cached (#4669).
Introduced a --max-pod-eviction-time flag to allow configuration of MaxPodEvictionTime, i.e., the maximum time the cluster autoscaler tries to evict a pod (#4842).
Fixed an issue where large number of pods managed by a single controller with slightly different pod spec (ex. pods created by Jobs using completionMode: Indexed) could lead to extremely slow CA reaction (#4724).
Fixed multiple bugs related to CA not properly accounting for daemonsets in scale-up calculations (#4641, #4761).
Status ConfigMap now contains additional information about nodes with unready resources.

Abort early if ASG has no capacity. This allows CA retry a different ASG without waiting for 15 minute timeout (#4489).
Added support for colon (:) in the AWS ASG discovery tag keys (#4636)
Updated list of supported instance types (#4683).

Fixed a bug where CA would only delete one node, when it intended to remove multiple ones (#4634).

Refactored the cloud provider to support both plain Instance Pools and SKS Nodepools scaling (#4247).

Increased parallelism of fetching MIG instance templates, which should improve CA performance in clusters with very large number of MIGs (#4621).
Reduced the number of GCE API calls during large scale-downs (#4660).