Another small release, fixing issues with evacuating multiple nodes at once and a small issue with container ports colliding.

Fixed

• Name collision for CSI Node container ports
• Do not block evacuation on other satellites being evacuated.

Changed

• Updated images:
  • LINSTOR CSI 1.10.5

Another patch release for the 2.10 release. We have another update for the LINSTOR CSI driver that fixes a issues with the new snapshot naming scheme. We have also fixed the RBAC resources for the CSI Resizer sidecar.

There are also small updates to the HA Controller and DRBD Shutdown Guard that fix some unwanted warning being generated in regular use.

Fixed

• Update RBAC for csi-resizer sidecar.

Changed

• Updated images:
  • LINSTOR CSI 1.10.4
  • HA Controller 1.3.2
  • DRBD Shutdown Guard 1.1.2

This release updates the DRBD version to 9.2.16, along with an update to LINSTOR CSI to fix issues with the new fsck behaviour.

We also added configurable timeouts for the new evacuation process.

Finally, we fixed an issue where the CSI Driver would report SELinux capabilities, but the mount command stripped them because it saw SELinux as disabled.

Added

• Configurable timeouts for how long the Operator waits for volume to (re-)attach during evacuation.

Fixed

• Ensure mount commands pass along SELinux options by mounting SELinux directories.

Changed

• Updated images:
  • DRBD 9.2.16
  • LINSTOR CSI 1.10.3
  • Latest CSI sidecars

A small bug fix release.

We noticed that with the more verbose status messages, we accidentally caused the Operator to run in a loop continuously.

We also noticed that snapshotting stopped working because we unconditionally enabled the VolumeGroupSnapshot feature, which caused the CSI sidecar to stop working on clusters where this resource has not been enabled.

Finally, the Operator stopped working on cluster managed by Cluster API, where the Cluster API itself is not installed in the cluster.

All these issues have now been fixed.

Added

• Liveness probes for CSI sidecars

Changed

• Disable group snapshots if we cannot be sure they are supported, otherwise the csi-snapshot silently breaks.

Fixed

• Status update on LinstorCluster and LinstorSatellite resources no longer causes infinite reconciles.

Piraeus Operator 2.10 is our biggest release since releasing "v2" back in 2023. With release 2.10.0 we are proud to announce support for ReadWriteMany (RWX) filesystem volumes!

RWX support is achieved by creating a highly available NFS export for every volume. This comes with the advantage of using NFS as a tried and tested technology for RWX filesystem, with the drawback being: it's NFS. We strongly recommend any users to rely on RWO volumes for performance, only using RWX where it is absolutely necessary.

Apart from the RWX support, we now also integrate with Cluster API to enable graceful node shutdown: Set the deletionPolicy of your LinstorSatellites to Evacuate to enable this in your cluster.

There is also support for Volume Group Snapshots within LINSTOR CSI, along with other small improvements.

Lastly, if you use kubectl get linstorcluster, kubectl get linstorsatellite and co, you will get improved status information in the output.

Added

• Deploy LINSTOR Affinity Controller as part of Piraeus Datastore.
• Add option to set replica count for Deployment components.
• Integration with ClusterAPI to enable graceful node evacuation in case of rolling machine updates.
• Additional status information on LinstorCluster, LinstorSatelliteConfiguration and LinstorSatellite resources.
• Sign container images using keyless cosign signatures
• Support for Volume Group Snapshots
• Support for RWX filesystems using regular LINSTOR CSI volumes.

Changed

• Set shareProcessNamespace: true on workloads to automatically reap orphaned processes.
• LINSTOR CSI now use hostNetwork: true so RWX mounts use a stable IP when connecting to NFS.
• Updated images:
  • LINSTOR 1.32.3
  • LINSTOR CSI 1.10.1
  • DRBD Reactor 1.10.0
  • DRBD Shutdown Guard: 1.1.1
  • LINSTOR HA Controller 1.3.1
  • Latest CSI sidecars

This release fixes a few issues introduced with the scheduling changes introduced with v2.9.0.

We also update to the latest component versions, delivering various fixes and improvements.

Changed

• Image configuration for RHEL10 clones.
• Do not remove existing satellites for *NoSchedule taints.
• Updated images:
  • LINSTOR 1.32.1
  • LINSTOR CSI 1.9.0
  • DRBD 9.2.15
  • DRBD Reactor 1.9.0
  • kTLS-utils 1.2.1
  • Latest CSI sidecars

Fixed

• Fix a crash in the validation webhook when trying to change the type of storage pool.

Removed

• Removed image configuration for EOL distributions:
  • Ubuntu 18.04 ("Bionic Beaver")
  • Ubuntu 20.04 ("Focal Fossa")
  • CentOS 7
  • Debian 10 ("Buster")

This release brings a few new features and fixes.

On the operator front, the two major changes are centred around tolations and Satellite deletions.

First, previous versions of Piraeus Operator would automatically try to evacuate a node when the LinstorSatellite resource was deleted. This caused many users issues, as often these deletions happend "by accident" due to misconfiguration, missing node labels and so on. This behaviour is now no longer the default, but is controlled by the deletionPolicy.

Secondly, previous versions of the Operator basically ignored the usual node taintsm trying to run the Satellite workloads on all nodes not excluded by some affinity or label selector. Control plane nodes are usually only excluded from scheduling based on their taints, which again caused confusion in many users. With this release we take the taints of a node into account, and won't try to schedule Satellites on tainted nodes, unless these taints are tolerated.

On the component side, LINSTOR CSI now automatically tries to reconcile VolumeSnapshotClass resources with LINSTOR Remotes, which bridges a gap when restoring snapshots from a remote location. DRBD 9.2.14 fixes an issue that was most prevalent in Kubernetes setups, where restoring volumes from a snapshot in a typical 2 diskful + 1 tiebreaker configuration could lead to corrupted filesystems on the restored volume.

Added

• Support for different deletion policies on satellites.
• Option to select taints that should be tolerated by Piraeus Datastore.
• Allow pinning images in the image configuration to a specific digest.
• Support overriding image configuration directly on the Operator Deployment using RELATED_IMAGE_* environment variables.

Changed

• Deletion of LinstorSatellite no longer causes a node evacuation by default. See spec.deletionPolicy.
• Piraeus Datastore no longer runs on control plane nodes by default.
• "Inconsistent" resources no longer cause AlertManager notifications, as long as there is resync progress.
• Updated images:
  • LINSTOR 1.31.2
  • LINSTOR CSI 1.8.0
  • DRBD 9.2.14
  • kTLS-utils 1.1.0
  • Latest CSI sidecars

This release fixes a memory leak in the operator, that also meant that the LINSTOR API cache was not working as intended. There are also the usual number of image updates.

Added

• Option to set the maximum time between reconciliations.

Changed

• Updated default properties applied on controller level to match LINSTOR release 1.31.0
• Updated images:
  • LINSTOR 1.31.0
  • LINSTOR CSI 1.7.1
  • DRBD 9.2.13
  • Latest CSI sidecars

Fixed

• Fix Operator caches not being index by a comparable key, leading to cache misses and a memory leak.
• Fix missing ports: section on LINSTOR Satellites, need for certain network mesh solutions.

This release improves support for reusing existing LINSTOR clusters to provide storage for Kubernetes in a few different ways.

First, you can now specify multiple controller URLs when setting up the connection. This is commonly used when setting up a highly available LINSTOR Controller with DRBD Reactor: specify the possible IP addresses where the Controller is running, and the Operator and components will automatically connect to the right node.

Secondly, the Operator will now automatically try to determine the LINSTOR version for external controller, and deploy the matching Satellite version by default.

With the usual update of image versions, we also now support the x-replicas-on-different setting on StorageClasses. For example, to deploy 2 replicas per zone, you can now set:

parameters:
  linstor.csi.linbit.com/xReplicasOnDifferent: |
    topology.kubernetes.io/zone: 2

Added

• Automatically set the LINSTOR Satellite image version when using an external LINSTOR Controller.

Changed

• Disable debug logging by default for the Operator.
• Allow specifying multiple controller URLs for external LINSTOR clusters.
• Updated images:
  • LINSTOR 1.30.4
  • LINSTOR CSI 1.7.0
  • DRBD Reactor 1.8.0
  • LINSTOR HA Controller 1.3.0
  • Latest CSI sidecars

This patch release bumps the images for DRBD and DRBD Reactor.

The new DRBD release brings compatibility for Linux 6.11 and 6.12, which also restore compatibility for recent Debian Bookworm kernels. It also contains the usual round of bug fixes.

The new DRBD Reactor release only new feature is related to systemd, and so mostly irrelevant for Piraeus. It however also fixes repeated, harmless, warn-level messages about an unknown keyword in the events2 stream.

Changed

• Updated images:
  • DRBD 9.2.12
  • DRBD Reactor 1.6.0