[gardener-extension-provider-openstack]

Action Required

• [OPERATOR] The openstack provider extension is incompatible with Gardener version > v1.10.0 (if feature flag MountHostCADirectories is enabled on the Gardenlet) for Openstack Shoots with certain Kubernetes versions (>= 1.17.x, 1.18.x, 1.19.x without CSI migration complete). Please consult the compatibility notes under /docs/compatibility.md (#147, @danielfoehrKn)

Improvements

• [USER] An issue causing kube-controller-manager to panic when upgrading an OpenStack cluster from v1.18 to v1.19 is now fixed. (#145, @ialidzhikov)
• [USER] An issue preventing csi-driver-node Pods to be created when Shoot specifies .spec.kubernetes.allowPrivilegedContainers=false is now fixed. (#143, @ialidzhikov)
• [USER] The CSI StorageClasses (with provisioner cinder.csi.openstack.org) do now also specify WaitForFirstConsumer for volumeBindingMode. (#140, @ialidzhikov)
• [OPERATOR] The following options can now be configured through the extension's Helm chart values charts/gardener-extension-provider-openstack/values.yaml: (#159, @timuthy)
  • Health check worker count
  • minAllowed values for VPA
• [OPERATOR] CSI sidecar containers timeout is now increased to 3m. (#158, @kayrus)
• [OPERATOR] An issue causing CSI PV to do not have set spec.csi.fsType is now fixed. The csi-provisioner is now started with --default-fstype=ext4 which is the default fstype to be used when there is no fstype specified in the StorageClass. (#142, @ialidzhikov)
• [OPERATOR] The Webhook ensurer does not remove or add the /etc/ssl directory for kube apiserver deployments any more. This is done by the Gardenlet for version >= 1.10.0. (#141, @danielfoehrKn)

[machine-controller-manager]

Most notable changes

• [USER] Support for Spot Instances is available in AWS driver. If the spotPrice is empty, price is automatically set to the on-demand price so that Spot instance can launch immediately. (gardener/machine-controller-manager#481, @zuzzas)
• [OPERATOR] Introduced a backoff in re-enqueuing machines on creation/deletion failures. Avoids throttling APIServer & provider calls. (gardener/machine-controller-manager#523, @hardikdr)
• [OPERATOR] RBAC policies have to be updated to allow updating of node/status resources. (gardener/machine-controller-manager#492, @guydaichs)
• [OPERATOR] New flag delete-migrated-machine-class is introduced. When set to true (defaulted to false), deletes any provider-specific machine class (e.g. AWSMachineClass) that has the machine.sapcloud.io/migrated annotation set on it. (gardener/machine-controller-manager#484, @prashanth26)
• [DEVELOPER] Added migration logic for moving from provider-specific machine class to generic machine classes in out of tree code path. On migration, the machine.sapcloud.io/migrated annotation set on the old machine class. (gardener/machine-controller-manager#484, @prashanth26)
• [DEVELOPER] The machine controller adds finalizer only when machine reference is present, deletes it otherwise. (gardener/machine-controller-manager#484, @prashanth26)

Improvements

• [USER] Retry when secret is referred by machineClass is missing (gardener/machine-controller-manager#495, @AxiomSamarth)
• [USER] Node condition is added to the status of terminating nodes indicating the termination start time and reason (Unhealthy|ScaleDown) (gardener/machine-controller-manager#492, @guydaichs)
• [OPERATOR] Added a more comprehensive set of events to trigger machine class reconciliations. (gardener/machine-controller-manager#532, @prashanth26)
• [OPERATOR] Finalizers are added by default for all machine class objects. (gardener/machine-controller-manager#532, @prashanth26)
• [OPERATOR] AWS: Allow deletion of VMs even on list image or modify instance failure (gardener/machine-controller-manager#516, @prashanth26)
• [OPERATOR] All nodes under machine deployments being rolled-out are annotated with cluster-autoscaler.kubernetes.io/scale-down-disabled: "True" during the period of rolling-update. (gardener/machine-controller-manager#496, @hardikdr)
• [OPERATOR] A new command line flag autoscaler-scaldown-annotation-during-rollout is introduced to disable annotating the nodes with cluster-autoscaler annotation cluster-autoscaler.kubernetes.io/scale-down-disabled during rollout. (gardener/machine-controller-manager#496, @hardikdr)
• [DEVELOPER] Adds a new phase CrashLoopBackOff that is set due to machine creation failures. (gardener/machine-controller-manager#523, @hardikdr)

Docker Images

gardener-extension-provider-openstack: eu.gcr.io/gardener-project/gardener/extensions/provider-openstack:v1.12.0
gardener-extension-validator-openstack: eu.gcr.io/gardener-project/gardener/extensions/validator-openstack:v1.12.0