Release 3.18.3

Enhancement

Improved rebuild responsiveness: a placement group (PG) rebuild now stops promptly when a read or write error occurs during the rebuild, instead of prolonging the operation.

LBM1-41399

Enhancement

NM and CM will automatically pick up new certificates when they are placed in the expected location(s).

LBM1-44917

Major

A data-integrity safeguard has been added to prevent a read from completing with incorrect data in the rare case where an SSD fails to write into the read buffer due to an IOMMU fault.

LBM1-45131

Major

Fixed a bug where deleting a snapshot during a rebuild could cause a damaged object to be read on the primary, triggering data-integrity alerts and preventing the affected data from being correctly persisted on the secondary.

LBM1-47343

Major

Fixed an edge case where the frontend could hang for up to 120 seconds due to a missing timeout on the replicator connection.

LBM1-46486

Major

Fixed an issue where aborted client-kernel NVMe commands triggered unexpected node down/up state transitions in VMware environments.

LBM1-45998

Major

Hardened the etcd-based leader election so that, in a rare combination of node failures, etcd disruptions, and network issues, two management instances can no longer be active simultaneously.

LBM1-44972

Major

Strengthened rebuild handling against a rare timing condition where a rebuild to a newly added empty node failed within a few seconds of starting; a subsequent rebuild attempt could then skip part of the data destined for that node. Rebuilds now transfer all data to the new node in this scenario.

LBM1-45601

Minor

Node Manager now shuts down faster when it is no longer needed, reducing wait times.

LBM1-45245

Minor

Fixed an issue where a secondary node failure could cause outstanding client writes to hang for longer than the client I/O timeout. Replications waiting on a pending connection are now canceled when the connecting peer is removed, allowing failover to complete promptly.

LBM1-45904

Minor

Fixed incorrect values (namespace ID, QoS limits, size) reported in the frontend log upon volume update completion, though the volume updates themselves were applied correctly.

LBM1-46178

Minor

Improved offline installation robustness: installation now completes successfully on hosts that have additional repositories enabled under /etc/yum.repos.d.

LBM1-43390

Minor

Improved the resiliency of cluster encryption key (KEK) rotation, which now completes reliably under a high volume of concurrent volume and snapshot operations.

LBM1-44673

Minor

Improved the resiliency of internal management workflows (FSMs): previously, a workflow that could not determine a valid state would exit silently, leaving its object (for example, a pending volume update) unresponsive until the management service was restarted, while the rest of the cluster continued operating normally. Workflows now pause after the maximum retry attempts and resume automatically on the next trigger, once the cluster state has settled.

LBM1-44465

Minor

Node manager was ending its shutdown sequence before the frontend had finished shutting down cleanly; the frontend now gets dedicated time within node manager's shutdown sequence.

LBM1-44334

Minor

Reduced Prometheus metric cardinality for frontend replication metrics by aggregating them per core/shard and instance (outgoing metrics labeled by peer UUID, incoming by peer IP) and removing irrelevant metrics, preventing excessive metric growth on large clusters.

LBM1-46313

Minor

The tar and gzip packages are now installed on management hosts as a requirement for lb-support, ensuring that log collection succeeds on hosts where they were absent.

LBM1-43401