Cluster Autoscaler 1.18.0

Changes since 1.17.0

General:

• Add cluster-api based cloudprovider
• Add /debug/pprof http handler
• Add topology.kubernetes.io labels to be ignored when comparing similar node groups
• Add logic for removing taints with specific prefix from template node
• Always filter out schedulable against upcoming nodes
• Uniformize node shuffling when finding nodes to remove
• Remove OldBinpackingEstimator
• Code cleanups; Notable ones:
• Simplify building equivalence groups in scale up
• Migrate CA to use scheduler framework
• Lots of bug fixes and performance imporovements

AWS:

• Update AWS instance type list
• Change the regex to match the providerID used by Fargate on EKS
• Add eks.amazonaws.com/nodegroup label to awsIgnoredLabels

Azure:

• Read subscription ID from instance metadata
• Fix VMSS scale down issues when updating target sizes
• Generate azure instance types
• Fix azure autodiscovery when max tags aren't specified
• Remove ACS support
• Decrease the number of calls to VMSS API
• Make VMSS info cache TTL configurable
• Add fallback check for 'Promo' instance types
• Fix parsing azure cloud provider config
• Add VM Cache
• Delete outdated cluster-autoscaler deployments
• Dont return error for nodes without providerId
• Support Azure user assigned identity

GCE:

• Add Windows nodes support
• Ignore node termination taint in scale up

Image: gcr.io/google-containers/cluster-autoscaler:v1.18.0