Operator Release Notes
January 27, 2023
- In Operator version 1.10.2, the Portworx pod was being scheduled on a random node because of a missing node name in the Portworx pod template. This issue is fixed in Operator version 1.10.3.
January 24, 2023
- Stork now uses
KubeSchedulerConfigurationfor Kubernetes version 1.23 or newer, so that pods are evenly distributed across all nodes in your cluster.
Dec 5, 2022
Added support for Kubernetes version 1.25, which includes:
PodSecurityPolicywhen deploying Portworx with Operator.
Upgraded the API version of
PodDisruptionBudgetfrom policy/v1beta1 to policy/v1
Added a UI option in the spec generator to configure Kubernetes version when you choose to deploy Portworx version 2.12.
Operator is now deployed without verbose log by default. To enable it, add the
--verboseargument to the Operator deployment.
For CSI deployment, the px-csi-ext pods now set Stork as a scheduler in the px-csi-ext deployment spec.
Operator now chooses
maxStorageNodesPerZone’s default value to efficiently manage the number of storage nodes in a cluster. For more details, see Manage the number of storage nodes.
Oct 24, 2022
- Pure1 integration has been re-architected to be more robust and use less memory. It is supported on Portworx version 2.12 clusters deployed with Operator version 1.10.
- To reduce memory usage, added a new argument
disable-cache-forto disable Kubernetes objects from controller runtime cache. For example,
- Operator now blocks Portworx installation if Portworx is uninstalled without a wipe and then reinstalled with a different name.
- For a new installation, Operator now sets the max number of storage nodes per zone, so that the 3 storage nodes in the entire cluster are uniformly spread across zones.
- Fixed a bug where DaemonSet migration was failing if the Portworx cluster ID was too long.
Sep 8, 2022
- Added support for Kubernetes version 1.24:
docker.ioprefix for component images deployed by Operator.
- To determine Kubernetes master nodes, Operator now uses the
control-planenode role instead of
- In Operator 1.9.0, when you enabled the CSI snapshot controller explicitly in the StorageCluster, the
csi-snapshot-controllersidecar containers might have been removed during an upgrade or restart operation. This issue is fixed in Operator 1.9.1.
Aug 1, 2022
- Daemonset to Operator migration is now Generally Available. This includes the following features:
- The ability to perform a dry run of the migration
- Migration for generic helm chart from Daemonset to the Operator
- Support for the
- Support for various configurations such as external KVDB, custom volumes, environment variables, service type, and annotations
- You can now use the
generic helm chartto install Portworx with the Operator. Note: Only AWS EKS has been validated for cloud deployments.
- Support for enabling
pprofin order to get Portworx Operator container profiles for memory, CPU, and so on.
- The Operator now creates example CSI storage classes.
- The Operator now enables the CSI snapshot controller by default on Kubernetes 1.17 and newer.
- Fixed an issue where KVDB pods were repeatedly created when a pod was in the
- When you upgrade Operator to version 1.9.0, the snapshot controller containers are removed from
px-csi-extdeployment when the
installSnapshotControllerflag is set to true explicitly in the StorageCluster spec.
Workaround: To fix this issue, either restart Operator or upgrade to a newer version.
June 22, 2022
- Added support for Operator to run on IPv6 environment.
- You can now enable CSI topology feature by setting the
truein the StorageCluster CRD, the default value is
false. The feature is only supported on FlashArray direct access volumes.
- Operator now uses custom SecurityContextConstraints
- You can now add custom annotations to any service created by Operator.
- You can now configure
ServiceTypeon any service created by Operator.
- Fixed pod recreation race condition during OCP upgrade by introducing exponential back-off to pod recreation when the
operator.libopenstorage.org/cordoned-restart-delay-secsannotation is not set.
- Fixed the incorrect CSI provisioner arguments when custom image registry path contains “:“.
Apr 14, 2022
- Daemonset to operator migration is in Beta release.
- Added support for passing custom labels to Portworx API service from StorageCluster.
- Operator now enables the Autopilot component to communicate securely using tokens when PX-Security is enabled in the Portworx cluster.
- Added field
preserveFullCustomImageRegistryin StorageCluster spec to preserve full image path when using custom image registry.
- Operator now retrieves the version manifest through proxy if
- Stork, Stork scheduler, CSI, and PVC controller pods are now deployed with
topologySpreadConstraintsto distribute pod replicas across Kubernetes failure domains.
- Added support for installing health monitoring sidecars from StorageCluster.
- Added support for installing snapshot controller and CRD from StorageCluster.
- The feature gate for CSI is now deprecated and replaced by setting
- Added support to enable hostPID to Portworx pods using the annotation
- Operator now sets
fsGroupPolicyin the CSIDriver object to
File. Previously it was not set explicitly, and the default value was
skip-resourceannotation to PX-Security Kubernetes secrets to skip backing them to the cloud.
- Operator now sets the dnsPolicy of Portworx pod to
- When using Cloud Storage, Operator validates that the node groups in StorageCluster use only one common label selector key across all node groups. It also validates that the value matches
spec.cloudStorage.nodePoolLabelif a is present. If the value is not present, it automatically populates it with the value of the common label selector.
- Fixed Pod Disruption Budget issue blocking Openshift upgrade on Metro DR setup.
- Fixed Stork scheduler’s pod anti-affinity by adding the label
name: stork-schedulerto Stork scheduler deployments.
- When a node level spec specifies a cloud storage configuration, we no longer set the cluster level default storage configuration. Before this fix, the node level cloud storage configuration would be overwritten.