Release 3.9.3

1

backend: Fixed a potential node crash that could happen when a disk fails or is removed during system operation.

LBM1-33893

2

backend: Fixed a rare situation in which a node could get stuck when a disk fails.

LBM1-30413

3

backend: Fixed a bug where the backend crashed following a detection of disk failure during node rebuild.

LBM1-33690

4

cluster-manage r: Fixed an issue where control operations could be unavailable, following the loss of an active cluster manager. For an issue to occur, a cluster-manager instance must first experience some other error, such as a network or load issue that coincides exactly with a very specific and extremely small window of time, for updating internal resources with information on migration completion.

LBM1-33339

5

cluster-manager: Fixed an issue that could cause a volume to be stuck in Migrating state following an earlier migration attempt.

LBM1-33366

6

discovery-service: Fixed a potential loss of discovery-service while reconnecting, following an initial failure to connect.

LBM1-33586

7

duroslight: Fixed a potential issue that could cause duroslight slowness/unresponsiveness, which would in turn lead to a node becoming inactive. The fix for this was an update of the code to use the mlockall (--lock-memory option) to keep duroslight pages from eviction out of page cache. This prevented an abnormal kernel behavior observed in el9_3 kernels that kept evicting R/O pages and made the process very slow, due to constant major page faults that re-read from the executable file.

LBM1-33171

8

duroslight: Fixed a rare race that could occur during a series of consecutive starts of rebuild and aborted rebuild (i.e., when a node jitters between Active/Inactive states or in and out of Read Only state). This issue could result in the affected volumes remaining in Degraded state.

LBM1-34173

9

node-manager/cluster-manager: Fixed an issue where - due to some rare internal races and internal data inconsistencies - services logs could be flooded.

LBM1-33773

10

node-manager: Enhanced auto revive support to attempt a graceful shutdown of the backend service, once auto-revive logic was triggered (auto-revive logic detected an unresponsive service, or a fatal error in a service).

LBM1-33628

11

node-manager: Fixed an issue in the calculation and allocation of memory space for Lightbits services. node-manager now takes the page-table overhead (0.2%) into account when calculating the free RAM available. This is to prevent this overhead from depleting the systemReservedRam specified in the system-profile.yaml on systems with large RAM, where 0.2% is the same order of magnitude as systemReservedRam.

LBM1-33380

12

lb-csi: Fixed an issue of trying to unmount a volume on unpublish and unstage - even if the verify-is-mount-point fails on io-error - to enable clean detaching of a volume that reported I/O errors. Otherwise the filesystem could report an IO error and will not be usable by any workload.

LBM1-33573

13

userlbe: Fixed an issue that caused abrupt recovery to fail with assertion failure in md_recovery.c:644, leaving a server unable to power up after abrupt shutdown. This can happen following the deletion of more than one snapshot of a volume, where one of the deleted snapshots has the SnapshotNsidVer value matching a certain pattern (two values out of each 64, bits [5:1]=0x1f). The snapshots must contain data prior to deletion (non-zero values of userWritten/physicalCapacity).

LBM1-34586