Changes by Kind

Feature

DRA Improvements

For this release we focused DRA efforts on delivering a production-ready implementation of the basic DRA Structured Parameters feature set (a basic Cluster Autoscaler implementation shipped in 1.32.0.

• Use DeltaSnapshotStore for optimal processing (#8235, @mtrqq
• Handle node readiness for DRA after a scale-up (#8109, @bdelrahman882
• Patches based implementation for DRA snapshot (#8090, @mtrqq

General Features

• Added checkCapacityProvisioningRequestProcessorInstance option that allows to filter check capacity ProvisioningRequests to process by specific Cluster Autoscaler instance. (#7676, @macsko)
• Adding pods when creating a cluster snapshot for scale up is now parallelized, improving performance. Parallelization can be adjusted or disabled by setting the cluster-snapshot-parallelization flag. (#7630, @macsko)
• Allow setting Hetzner Cloud Cluster Config by providing a json config file, set via HCLOUD_CLUSTER_CONFIG_FILE (#7817, @karsten42)
• Cloud Provider using cluster-api is extended to handle DRA devices such as GPU in the scale-from-0-nodes. (#7804, @ttsuuubasa)
• Cluster-autoscaler kwok cloudprovider automatically annotates nodes with metrics-server path (#7339, @justinmir)
• Improved frequent loops when there is only one of scale up activity or ProvisioningRequest processing is productive. (#7679, @macsko)
• ProvisioningRequests can now be processed without delay if they have just been created. (#7688, @macsko)

Azure

• Enable VM node pools #8078, @wenxuan0923

AWS

• AWSCloudProvider should ignore unrecognized provider IDs #8047, @raykrueger

GCE

• pricing changes: updated z3 pricing information #8137, @DigitalVeer
• Improve resilience of diskTypes requests #7923, @Uladzislau97

Bug or Regression

• Add events for pods that are not triggering scale up because of reaching total max nodes. (#7667, @jbtk)
• Azure: fixed an issue where CAS crashes when VMSS is not present or has no nodes (#7708, @comtalyst)
• Bug fixes and improvements for Kamatera cluster autoscaler to align with latest Kamatera platform changes (#7767, @OriHoch)
• Fixed memory leak for GCE instance templates fetching (#7694, @mtrqq)
• New ScaleDownNoCandidates status emitted instead of existing ScaleDownInCooldown when there are no candidates.
• last_activity{activity=scaleDown} metric will be updated even when there are no candidates. (#7997, @damikag)
• On Cluster API, node groups that contain failed Machines can have their size decreased when the autoscaler determines that there is a difference between the node count and the instances in the node group. (#7950, @elmiko)
• Report machine phases in ClusterAPI provider to allow faster autoscaler decisions and avoid timeout delays. (#7989, @loick111)
• Set k8s.io/client-go version to v0.33.0. (#8165, @nickytd)
• The log messages related to the DecreaseTargetSize function have been updated to better reflect the names of the values. (#7929, @elmiko)

Other (Cleanup or Flake)

• Enabled logging time of scale down logic when it takes too long. (#8185, @x13n)
• Introduce new scale up status when cluster reaches max nodes total limit. (#7648, @jbtk)
• Misc logging improvements. (#7639, @x13n)
• Removed deprecated flags:
• max-autoprovisioned-node-group-count -> had a warning and deprecated in 1.31
• node-autoprovisioning-enabled -> had a warning and deprecated in 1.31
• gce-expander-ephemeral-storage-support -> no references, initialized with "_", planned to be removed in 1.30+
• max-empty-bulk-delete -> had a warning and deprecated in 1.29 (#7890, @YahiaBadr)
• Update imports to synchronize with Kubernetes v1.33.1 (#8237, @jackfrancis)

Uncategorized

• Deployments with topology spread constraints will no longer block scale down if the removable node changes the global minimum (#8164, @MenD32)
• Fixed a bug that prevents third party statefulset from being drained by cluster auto scaler (#8011, @jinglinliang)
• On GKE, Adding node labels that start with "clusterautoscaler-nodetemplate-resources-", with value equal to the amount of the resources, allows the cluster autoscaler to detect extended resources and scale up the node pools from zero. (#7604, @mu-soliman)
• The default cluster-autoscaler expander as been changed from random to least-waste to prevent misused expensive nodes. (#7631, @jfcoz)

Images

• registry.k8s.io/autoscaling/cluster-autoscaler:v1.33.0
• registry.k8s.io/autoscaling/cluster-autoscaler-arm64:v1.33.0
• registry.k8s.io/autoscaling/cluster-autoscaler-amd64:v1.33.0
• registry.k8s.io/autoscaling/cluster-autoscaler-s390x:v1.33.0