Release 3.15.3

Enhancement

AMD dual socket support - all previous AMD dual socket support-related limitations have been lifted in this release.

LBM1-43382

Enhancement

Changed AEN for deleted volumes to only notify the necessary clients - thus reducing the overhead and load on the servers/nodes as well as the client machines following delete volume operations.

LBM1-39262

Enhancement

Enabled encrypted clones from unencrypted base snapshots, allowing derived volumes to use unique keys. Supported on clusters configured with encryption.

LBM1-42408

Major

Fixed a bug related to incorrect error handling that might occur during the handling of a change in a volume's protection state, which under rare certain circumstances can lead to service loss.

LBM1-39742

Major

Fixed a bug where a node could crash when powering up from an abrupt failure in the rare case where the volume containing the most recently written data is deleted just before an NVMe device failure. The system also completes the full rebuild before any new writes are issued to any volume replicated on that node.

LBM1-41068

Major

Fixed a bug where under an extremely rare race condition that can occur during background garbage collection while two successive snapshots are deleted, it is possible for data from an older volume snapshot to overwrite more recent data.

LBM1-40626

Major

Fixed an issue where a volume rebuild may fail to complete following an internal error in the handling of create new snapshot. When a very specific portion of the handling of a create snapshot task occurs exactly as the cluster manager service is switched over, this volume and other volumes that share the same protection group may get into an inconsistent state that will prevent the completion of a volume rebuild.

LBM1-40208

Major

node-manager & etcd: Reduced Etcd's leader election timeout for new installations. In rare cases, if the Etcd leader experiences slowness (e.g., due to a malfunctioning disk), lightbits-nodes could miss heartbeats and become inactive before Etcd followers trigger a new leader election. By lowering the leader election timeout, leader elections now occur promptly, preventing nodes from incorrectly turning inactive.

LBM1-41460

Minor

cluster-manager: In rare cases, a deprecated event key in etcd could cause the event cleaner to exit, leading to event accumulation and potential OOM during CM switchover.

LBM1-42614

Minor

Fixed an issue when cluster encryption is enabled, KEK is rotated, and the cluster manager service is switched over - that could lead to a failure to process new API calls, and actively react to changes in the cluster. If the cluster manager fail-over sequence coincides exactly with a very specific time window in the KEK rotation process, the cluster manager service may fail to start again, preventing successful handling of new API requests or changes of state in cluster.

LBM1-42309

Minor

Fixed an issue where a volume protection state might be reported incorrectly in the API as fully protected instead of degraded or read only, following a permanent failure re-balance that fails.

LBM1-42282

Minor

Resolved an issue where a snapshot deleted while a node was inactive and was not removed from the node after it recovered. This inconsistency could potentially cause rebuild or migration failures when the node later assumed the primary role.

LBM1-41162

Minor

userlbe: Contributed to data integrity and system reliability by fixing a rare issue where a volume could get stuck in a "Deleting" state. This could occur if a volume was deleted and then quickly recreated before the deletion fully completed, causing the affected volume replica to silently ignore all incoming writes. This fix ensures that the volume state is correctly handled, preventing data inconsistency.

LBM1-40243

Minor

userlbe: Enabled graceful recovery even when some required cores or workers are unavailable, preventing fallback to abrupt recovery as long as at least one graceful recovery worker can be allocated.

LBM1-40871