v1.2.1

This releases fixes some crashes when using multiple S3 remotes.

---

Fixed

• Fix crashes when listing backups

v1.2.0

This release brings two new features for S3 snapshots:

First, LINSTOR CSI will now also delete the S3 part of the backup when the VolumeSnapshot is deleted.

Secondly, you can now use snap.linstor.csi.linbit.com/delete-local: "true" in the snapshot parameters to automatically delete the local snapshot after a S3 backup completes.

---

Added

• Delete S3 backups in delete snapshot requests.
• Delete local snapshots when creating S3 backups when new "delete-local" parameter is true.

v1.1.1

This release fixes an issue with the way read-only volumes are handled, letting LINSTOR CSI pass the Kubernetes E2E tests for ROX volumes.

---

Changed

• Removed unnecessary checks for read-only publishing by the controller.

v1.1.0

This release fixes an issue when using replicasOnSame and replicasOnDifferent with kubernetes label values when deployed with Operator v2.

It also removes a work around for unexpected quorum loss when running DRBD < 9.1.12,

---

Changed

• When using another property namespace (other than Aux/), also use it to prefix replicasOnSame
  and replicasOnDifferent. Values already prefixed with Aux/ will be left unchanged.
• Starting with 0.13.1, the CSI driver tried to work around an issue in quorum calculation
  by forcing the creation of a diskful instead of the normal diskless resource. Since DRBD 9.1.12/9.2.0,
  this is no longer necessary, as quorum is now inherited from the diskful counterparts.

v1.0.1

This release improves the reported storage capacity to take into account evicted or evacuated nodes. This prevents some issues when using strict volume scheduling. Note that there are still scenarios in which a Pod is scheduled to a node where LINSTOR may not be able to create a volume.

---

Added

• Ignore storage capacity from "unschedulable" nodes. Those are nodes that are either evicted/evacuated or have the
  AutoplaceTarget property set to false.

v1.0.0

With this release, we finally switch away from zero-versioning: the driver has been stable for a while, so using 1.0.0 seems appropriate.

The only change since 0.22.1: LINSTOR CSI automatically adds the _netdev mount option, which triggers unmounting during node shutdown while the main network is still available: this fixes a potential issue where DRBD devices would get suspended before systemd tries to unmount remaining mounts.

---

Changed

• Add _netdev mount option to all mounts. This ensures unmounting on system shutdown happens before the
  network is unconfigured, which might cause DRBD to lose quorum.

v0.22.1

Changed

• Use golinstor based initialization of LINSTOR client. This enables checking the server TLS without requiring a
  mutual TLS config, as well as using the new built-in User-Agent and Bearer Token configs.

v0.22.0

As we skipped the release notes for 0.21.0, here are the notes for 0.21.0 and 0.22.0 combined:

Added

• Support for live-migration in kubevirt by setting the allow-two-primaries property on demand. This only affects RWX volumes in block mode. If used for anything other than kube-virt, results cannot be guaranteed.
• Add option to set property namespace used to determine a nodes topology. Used for Piraeus Operator v2.
• Add option to skip labelling nodes based on configured storage pools. Used for Piraeus Operator v2.

Fixed

• LINSTOR may return more backups than requested when searching for a specific snapshot ID. The results are now filtered, ensuring we really restore from the expected backup.

v0.20.0

Added

• A new check ensures snapshots are only attempted when the resource is completely in sync. This prevents an issue
  with LINSTOR creating too many error reports in case a node is (temporarily) offline.
• Support CSI Volume Conditions

Changed

• Update CSI Spec to 1.6.0

v0.19.1

Just a small release, improving interaction with external components and clusters without quorum configuration.

---

Added

• Store access policy in volume context, enabling volume context parsing without access to the storage class.
  This improves the situation for LINSTOR Scheduler and Affinity Controller, which need to access this value.

Changed

• The attach logic was updated to take the quorum configuration into account. Previously, the CSI driver would
  create a diskful resource in the assumption that it is required to achieve quorum, even if the cluster was
  configured without quorum.