Skip to content

Cluster Autoscaler 1.17.0
Compare
Choose a tag to compare
@losipiuk losipiuk released this 29 Nov 12:51
· 143 commits to cluster-autoscaler-release-1.17 since this release
cluster-autoscaler-1.17.0
477857a
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

Changes since 1.16.0

General:

  • Ignore kops.k8s.io/instancegroup label when camparing nodes for similarity
  • Ignore alpha.eksctl.io/nodegroup-name label when camparing nodes for similarity
  • Switch leader election mechanism to use lease objects (Note: requires updated RBAC rules)
  • Invalidate node instances cache after deleting failed and unregistered nodes
  • Allow for greater node memory mismatch on comparison for similarity (bumped to 256K from 128K)
  • Allow marking pods with cluster-autoscaler.kubernetes.io/daemonset-pod annotation. Marked pods will be treated as daemon set pods.
  • Code cleanups
  • BugFixes

AWS:

  • Update aws-sdk-go to 1.23.18
  • Bugfix; Fix case when ASG size could be descresed twice
  • Don't treat lack of queried node in cache as error
  • Bugfix; Allow arbitrary placeholder AWS instance names
  • Ensure we use valid AWS LaunchTemplate version
  • Introduce g4dn instance types
  • Allow automatic dicovery of EC2 instance types. Enabled by default. Use --aws-use-static-instance-list to change to static list.

Azure:

  • Add support for NV6Promo instance types
  • Allow for node template labels and taints to come from the VMSS's tags
  • Add node group autodiscover via --node-group-auto-discovery flag
  • Use List call which obtains the status for all the node pools at once. Improves performance
  • Use version from cluster autoscaler package instead of client-go for azure autorest

Baidu:

  • Update version of Baidu cloud provider (multple changes)
  • support multiple groups and GPU group

GCE:

  • Do not break CA loop if if finds Node which looks like it is coming from autoscaled MIG; but MIG itself cannot be found.
  • Increase default GCE operation wait timeout to 20s
  • More verbose logging of GCE instance create errors
  • Less aggresive errors reporting for created nodes in GCE (unknown error will not trigger node removal if VM is already UP)

Digital Ocean:

  • DOKS node pool autoscale API integration
  • Update godo to 1.22.0
  • Fixes

Image: gcr.io/google-containers/cluster-autoscaler:v1.17.0

Assets 2
Loading