Cluster Autoscaler 1.27.1

Note: This is the first 1.27 release, we're starting with 1.27.1 to match the version of our synced k8s dependencies.

Changelog

General

• Fix an issue when not all DaemonSet pods would schedule on node group templates due to unsanitized taints. #5659
• Add annotation cluster-autoscaler.kubernetes.io/safe-to-evict-local-volumes which enables CA to ignore local storage volumes during scale-down. #5594
• Enable evicting pods with in-memory storage. #5551
• Add flag --allow-scale-down-on-custom-controller-owned-pods which allows scale-down of pods owned by custom controllers not recognised by CA. #5507, #5653
• Add flag --scale-down-unready-enabled which allows enabling/disabling scale-down of unready nodes. #5537
• Fixes and optimisations for parallel node drain. #5578, #5562, #5556, #5552, #5502
• Fix scale-down node deletion time metric calculation. #5472
• Limit maximum duration of binpacking simulation to prevent CA becoming unresponsive for huge scale-up scenarios. This behavior is controlled by --max-nodes-per-scaleup and --max-nodegroup-binpacking-duration flags. ​​#4970
• Add resource_name to scaled_up_gpu_node_total and scaled_down_gpu_nodes_total metrics. #5518
• Stop scale-down from considering registered upcoming nodes as candidates. #5488
• Add flag --force-ds which enables force scheduling Daemon Sets on node group templates created from existing nodes. #5444
• Add filtering out Daemon Set pods from scale-up, making them no longer emit scale-up events. #5442
• Add flags --memory-difference-ratio, max-free-difference-ratio and --max-allocatable-difference-ratio used to adjust difference ratios and configure the balancing of similar node groups. #5402
• Cap logs logged by HintingSimulator. #5380
• Add flags --kube-client-burst and --kube-client-qps which control how many requests CA can do against Kubernetes API. #5223

GCE

• Add MaxNodeProvisionTime override that can be set on the node group level. #5649
• Limit number of GCE API calls performed after big scale-down events. #5665
• Add pricing for the C3 machine family. #5572
• Fix MEMBER_ALREADY_EXISTS errors occasionally causing large scale-ups to fail. #5496

AWS

• Limit the number of InstanceRequirements API calls. #5550
• Fix incorrect endpoints for sku-client. #5603
• Add support for inf2 instance types. #5529
• Add support for m7g/r7g instance types. #5508
• Fix handling extended resources provided by ASG via tags. #5214
• Add support for AWS_MAX_ATTEMPTS to configure max retries. #5270

Azure

• Add support for VMSS Flex. #5602
• Add support for workload identity. #5390

Hetzner

• Add support for Hetzner Cloud Arm Server Types. #5677

Cluster API

• Add support for MachinePools. #4676
• Add support for supplying labels and taints when scaling to and from zero nodes. #5382
• Add support for ephemeral disk capacity annotation for scaling from zero: #5416
• Inherently ignored labels for balancing similar node groups with the Cluster API provider have been removed, please use the --balancing-ignore-label flag to add any labels which must be ignored. #5375

OCI

• Add support for OCI node pools. #5577

OVHCloud

• Fix deletion of multiple nodes at the same time. #5584
• Add support for nodepool templating features. #5586

Brightbox

• Add support for additional Brightbox server groups. #5407

Images

• registry.k8s.io/autoscaling/cluster-autoscaler:v1.27.1
• registry.k8s.io/autoscaling/cluster-autoscaler-arm64:v1.27.1
• registry.k8s.io/autoscaling/cluster-autoscaler-amd64:v1.27.1