Skip to content

Releases: piraeusdatastore/linstor-csi

v1.6.3

28 Jun 05:40
v1.6.3
Compare
Choose a tag to compare

This release fixes an issue with volume provisioning not placing the resources on the intended node when the intended node itself does not have a storage pool. The eligible nodes should then be determined by the remoteAccessPolicy. Due to a bug introducing with caching back in release 1.2.3, this instead allowed placement on all nodes.


Fixed

  • Fixed node filtering when initial placement was not possible. This now takes the remoteAccessPolicy into account like
    intended.

v1.6.2

19 Jun 07:00
v1.6.2
Compare
Choose a tag to compare

Small fix for the previous release. LINSTOR does not always return a device path immediately after creating a resource, so we check that the device path is set before passing it to the csi node agent.


Fixed

  • Fix device path passdown when LINSTOR does not report a device path.

v1.6.1

16 May 12:13
v1.6.1
Compare
Choose a tag to compare

A small release that fixes an issue introduced by the recent caching changes. The fix also means that the csi-node plugin does not need to query LINSTOR for volume information.


Added

  • Pass device paths down from CSI Controller on publish, reducing LINSTOR API requests from the CSI Node.

v1.6.0

02 May 14:10
v1.6.0
Compare
Choose a tag to compare

This release brings support for a new snapshot type Linstor: This enables users to use Linstor-to-Linstor backup shipping using the CSI driver. This is currently limited to the creation of backups, restoring requires manual interaction.


Added

  • Support for creating backups with LINSTOR-to-LINSTOR shipping.

v1.5.0

09 Apr 10:11
v1.5.0
Compare
Choose a tag to compare

Update the base image to have access to newer FS utilities. This fixes a bug when running on new kernels that make use of "new" ext4 and XFS features.


Added

  • Support specifying multiple storage pools in the parameters, separated by spaces.

Changed

  • Update base image to Debian 12.

v1.4.0

01 Feb 09:04
v1.4.0
Compare
Choose a tag to compare

Remove support for Openflex layers: these caused LINSTOR 1.26.0 to return errors in some edge cases.


Breaking

  • Update golinstor to support LINSTOR 1.26.0, removing support for Openflex layers.

v1.3.1

30 Jan 12:53
v1.3.1
Compare
Choose a tag to compare

This release contains some fixes when attaching volumes (i.e. when a volume is mounted on behalf of a Pod).


Changed

  • Default access policy for resources without replication layer is now "local only".

Fixed

  • Do not try to create diskless resource if there is no compatible diskless layer (DRBD or NVMe) available.
  • Do not allow attaching a volume that has no existing replica.

v1.3.0

15 Nov 12:47
v1.3.0
Compare
Choose a tag to compare

This release improves the handling of some situations you may encounter with advanced setup.

If you restore from a remote backup in Kubernetes, and the SnapshotClass used sets delete-local: "true", the local copy of the snapshot will be deleted after restoring the volume.

On the topic of remote backups, LINSTOR CSI now randomizes the node used to download the backup if no guidance is given from the provisioner. For Kubernetes, this requires disabling the topology feature.

Last but not least, a new parameter overProvision is introduced: this influences the free space reported by LINSTOR CSI: if set to a value (floating point), the free space reported will be calculated by StoragePool capacity * overProvision - reserverd capacity. Reserved capacity is the potential size of all volumes already deployed in the storage pool. This may influence components, such as the Kubernetes scheduler, to better distribute volumes.


Added

  • Support to delete downloaded backups after restore operation (k8s only).
  • New parameter overProvision: when set available capacity on a node is calculated by taking into account
    the reserved capacity in the pool based on existing volumes.

Changed

  • When not using topology, select a random node to download the backup. This should prevent the same node
    being used to download all backups.

v1.2.3

01 Sep 06:44
v1.2.3
Compare
Choose a tag to compare

This release is focused on reducing the load generated by LINSTOR CSI on the LINSTOR cluster. This is especially important for the "status" calls, such as when collecting volume stats on the node or when fetching storage capacity, as those are generally called once per volume/(storage class x node), so have the potential to get called often in bigger clusters.


Changed

  • Reduced load on LINSTOR when attaching volumes.
  • Reduced load on LINSTOR when detaching volumes.
  • Reduced load on LINSTOR when checking node stats.
  • Reduced load on LINSTOR when listing volumes with status.
  • Reduced load on LINSTOR when fetching storage capacity.

v1.2.2

13 Jul 07:41
v1.2.2
Compare
Choose a tag to compare

This releases fixes an issue around restoring snapshots in volumes smaller than the original. This is not an operation we support, but in some cases we allowed it, leading to broken volumes. This has now been fixed.


Fixed

  • Prevent shrinking of volumes. While this is enforced by the CSI Provisioner already, when
    restoring from backups (which do not have a reported size) these limits are not enforced.