RATATOSKRATATOSK
로그인

Longhorn

v1.12.2Storage & Data
2026년 11월 9일

ACTION 1CHECK 4OTHER 74

데이터 손실 위험, 런타임 중단, 마운트 지연과 장치 상태 처리 문제를 다룬 안정성 중심의 유지보수 릴리스입니다. 네트워크와 Helm 차트 개선이 포함되며, 업그레이드 전 조건과 V2 운영 제약, 요청 로그의 비밀 정보 정제 문제도 함께 다뤄집니다.

조치 필요 (1)

  • securityCSI 요청 로그의 비밀 정보 정제 누락 수정

    CSI 메서드 수준 요청 로그가 비밀 정보 정제를 우회하던 문제를 수정했습니다.

    13727

영향 확인 (4)

  • breakingV2 Data Engine 실시간 업그레이드 제한

    v1.12.2에서 v1.13.0으로 V2 Data Engine을 실시간 업그레이드하는 경우에 적용됩니다.

    v1.12.2에서 v1.13.0으로 V2 Data Engine을 실시간 업그레이드하는 방식은 지원되지 않습니다.

  • breaking레거시 linked-clone 볼륨 작업 제한

    v1.12.1 이상으로 업그레이드하는 경우에 적용됩니다.

    legacy linked-clone volumes는 분리와 삭제 외에는 작업할 수 없습니다.

  • breakingV2 Data Engine CPU 코어 최소 기준

    V2 Data Engine에 CPU 코어를 2개보다 적게 설정하는 경우에 적용됩니다.

    안정성을 위해 V2 Data Engine에 CPU 코어를 2개 이상 설정해야 합니다. 다중 코어 마스크를 사용하면 한 코어가 admin reactor에 예약되어 입출력에 쓰이는 코어가 하나 줄어듭니다.

  • breakingLonghorn 설치 전 Kubernetes 버전 조건

    Kubernetes가 v1.25보다 이른 클러스터에 Longhorn v1.12.2를 설치하는 경우에 적용됩니다.

    설치 전에 클러스터가 Kubernetes v1.25 이상을 실행하고 있어야 합니다.

그 외 기록된 변경 74건 전체fixes 60 · value changes 8 · additions 4 · constraints 2

fixes (60)

  • Fixes a bug in which acknowledged writes could be silently lost during a replica rebuild.
  • Fixes replica rebuilds that randomly failed on busy networks because a single lost TCP keepalive terminated the idle sync-agent gRPC connection after 20 seconds.
  • the rebuild switched the parent of the destination replica head while it was still serving I/O
  • spdk_tgt could abort with double free or corruption (fasttop) during replica rebuilds.
  • Large reads (for example, 1 MiB at a high queue depth) could flood the NVMe-oF TCP target with flush retries
  • Deleting a snapshot that is smaller than its child and not a multiple of 512 MiB could leave lvol metadata that SPDK fails to load
  • Fixes a critical issue in which a failure while setting up a rebuild on the destination replica faulted the entire engine
  • Fixes a lost wakeup in the SPDK erasure-coding bdev that could stall writes and leave the SPDK target unresponsive
  • spdk_tgt crashes with "double free or corruption (fasttop)" around replica rebuilds
  • Concurrent unaligned writes during replica rebuild lose data
  • v2 instance-manager is crashed by disk recovery failure in SPDK
  • a rebuilding replica sometimes disappears in the Engine.Status.ReplicaModeMap
  • Concurrent volume attachments can create multiple NVMe-oF transports and crash spdk_tgt
  • V2 instance manager live upgrade stuck in relocating-engines
  • V2 instance-manager spdk_tgt allocates its hugepages from another pod's hugetlbfs mount
  • A stalled NVMe device makes engine frontends of unrelated v2 volumes fail validation and be recreated
  • nvme list zombie processes leak when NVMe admin queue returns EAGAIN
  • SystemBackup stuck in CreatingVolumeBackups
  • v2 spdk_tgt aborts with assertion in blob_insert_cluster_msg due to an extent page race
  • Fast replica rebuild skips a snapshot file that was changed by purge/coalesce
  • Transient status RPC failure marks healthy replica failed and can fault a volume
  • Disk health data never collected for V1 filesystem disks whose path is not itself a mount point
  • Force-deleted pod VolumeAttachment cleanup ignores "Waiting for detach" events
  • csi-resizer fails to start with "unrecognized feature gate: RecoverVolumeExpansionFailure"
  • V2 DR volume restore does not retry when backupstore lock is held
  • All v1 volume I/O stalls silently with longhorn-engine v1.7.1 through v1.9.2 on Rocky Linux 9.7 / RKE2
  • Sharded/EC volume: writes stall to ~0 and wedge the SPDK target on arm64/AIO
  • Volume expansion snapshot CR name expand-<size> collides across volumes
  • Longhorn v2 engine crashes on large-block reads in interrupt mode
  • v2 volume gets stuck in Attaching in an IPv6 storage network environment
  • v1.11.3 regenerates the webhook CA on upgrade from an ECDSA (dynamiclistener) CA, deadlocking every longhorn-manager
  • SystemBackup fails when a snapshot has an empty status.creationTime
  • invalid block disks should not block Longhorn system uninstallation
  • deleting nodes.longhorn.io failed due to KubernetesNodeGone
  • v1.12.1 chart: new default-on networkPolicies.restrictInternalTraffic silently breaks Prometheus scraping of longhorn-manager
  • V2 replica rebuild destination failure faults the entire engine
  • NodeStageVolume can trigger destructive mkfs on existing volume during temporary I/O errors, risking data loss
  • instance-manager crashes with "concurrent map iteration and map write" in ProcessList
  • SystemRestore job ignores the registry-secret setting
  • DR replicas break
  • Cloned volume stuck "not ready for workloads" after the clone finishes
  • Runtime container images contain development (-devel) packages
  • Adding V2 disk using the "Default Data Path" /dev/disk/by-id/scsi-* path fails
  • Request Schema in Upgrade Responder is Outdate
  • v2 volume expansion is reported complete before the frontend dm device is resized
  • Terminal workload pods cause NodePublishVolume to abort with "no Pending workload pods"
  • v1 engine never starts: a stale stopped process record makes createInstance a silent no-op
  • v1.12.1 default NetworkPolicies miss legitimate non-pod-labeled and host-originated callers, silently breaking CSI attach/detach
  • Recurring-job pods report success after startup or volume execution errors
  • V1 volumes not rebuilding after cluster shutdown
  • v2: spdk_tgt aborts on bdev_channel_destroy_resource assertion when rebuild finish sets the dst replica parent under I/O
  • V2 instance-manager liveness probe errors ('test: -eq: unary operator expected') -> self-kill -> node-wide replica fault under rebuild load
  • V2 Instance Manager panics in backupstore when S3 volume.cfg is missing, taking all node volumes offline
  • Admission webhook calls for one resource type (backups.longhorn.io) fail with instant EOF
  • Recurring snapshot-cleanup aborts the entire run when any volume's replica is rebuilding
  • - [BACKPORT][v1.12.2][BUG] V2 backup readers race snapshot cleanup and panic the instance manager [14021](https://github.com/longhorn/longhorn/issues/14021) - @yangchiu
  • - [BACKPORT][v1.12.2][BUG] Live migration and replica rebuild enter circular wait after nodes are uncordoned [14154](https://github.com/longhorn/longhorn/issues/14154) - @chriscchien @carterli0407-cell
  • - [BACKPORT][v1.12.2][BUG] instance-manager panics (nil pointer) in sync.(*Task).HashSnapshotStatus when a replica is rebuilding: shadowed err in isRebuilding branch [14252](https://github.com/longhorn/longhorn/issues/14252) - @chriscchien
  • - [BACKPORT][v1.12.2][BUG] RWX volume mounts are delayed up to 2 minutes when the share manager is stopped and started repeatedly (e.g. by a CronJob) [14178](https://github.com/longhorn/longhorn/issues/14178) - @roger-ryao
  • - [BACKPORT][v1.12.2][BUG] Longhorn block disk remains Ready=False and Schedulable=False when NVMe device is bound to vfio-pci [13934](https://github.com/longhorn/longhorn/issues/13934) - @derekbit @chriscchien

value changes (8)

  • Skip the filesystem sync when suspending the frontend for replica rebuilding
  • Sync v2 volume and engine image fields after upgrade
  • CSI ignores EC StorageClass parameters (silent replicated fallback)
  • CSI Resize may idle for a long time after volume clone or restore
  • (chart) Precedence of image.*.*.registry
  • nsmounter: kubelet PID lookup on Talos takes seconds per call
  • - [BACKPORT][v1.12.2][IMPROVEMENT] Avoid re-sending an unchanged Node status update [14145](https://github.com/longhorn/longhorn/issues/14145) - @hookak @chriscchien
  • - [BACKPORT][v1.12.2][TASK] Add Helm chart image HIP-0015 annotations [14204](https://github.com/longhorn/longhorn/issues/14204) - @roger-ryao @carterli0407-cell

additions (4)

  • You can now limit the number of I/O queues the kernel NVMe-TCP initiator creates for V2 volumes
  • Enable CPU Isolation for V2 Data Engine
  • backup target availability status metric
  • - [BACKPORT][v1.12.2][IMPROVEMENT] Chart value to allow metrics scrapers through the longhorn-manager NetworkPolicy [13989](https://github.com/longhorn/longhorn/issues/13989) - @yangchiu

constraints (2)

  • Previously they always used IPv4.
  • Longhorn may not reject expansion requests when the volume clone is in-progress
Longhorn 스택에 추가

조치가 필요한 릴리스가 나왔을 때 주간 메일로 알려드립니다. 이번 릴리스의 보안 패치와 브레이킹 체인지 같은 것들입니다.

스택에 추가 →
공식 릴리스 노트 ↗Longhorn 릴리스 전체 →