1.19.0 release for Cluster Autoscaler
General
- Support caching errors when fetching machine types.
AWS
- Fix AWS CA tests for InstanceType generation changes.
- Updated documentation.
- Ignore AWS NodeWithImpairedVolumes taint.
- Static Instance List updated.
Azure
- Fixed the label CA uses to detect nodes with GPUs from to "accelerator" (previously was an incorrect "cloud.google.com/gke-accelerator").
- Made Various improvements to scale down scenarios.
Cluster API
- Improved node deletion mechanics by adding locking around delete calls, checking minimum group size more accurately, and ensuring that node replica counts are not stale. This alleviates issues around erroneous deletes.
- Added documentation about CAPI requirements for running the autoscaler.
- Added an example manifest for deploying the autoscaler using CAPI.
GCE
- Added prices for all machine types. Also adjusted e2-small and e2-micro.
Huawei
- Added support for Huawei cloud.
Magnum
- Updated to support Magnum node groups (introduced in Magnum Train).
- Added node group autodiscovery based on the group's role proper.
- Cluster Autoscaler v1.19 and onwards will require Magnum Train or a later version.
- Note on compatibility.
Image: k8s.gcr.io/autoscaling/cluster-autoscaler:v1.19.0