RATATOSKRATATOSK
로그인

Longhorn

v1.13.0Storage & Data
2026년 9월 29일

ACTION 1CHECK 5OTHER 118

V2 Data Engine, 스케줄링, 스냅샷, 메트릭, 운영 기능을 넓히고 볼륨, 백업, 네트워킹, 복원력, 안정성 관련 문제를 폭넓게 수정한 릴리스입니다. 업그레이드 전 사전 조건을 확인해야 하며, 기존 기본값을 바꾸지 않는 선택적 복원력 및 CPU 관리 개선과 보안 수정이 포함되어 있습니다.

조치 필요 (1)

  • securityCSI 요청 로그의 비밀값 정제 우회 수정

    CSI 메서드 수준 요청 로그가 비밀값 정제를 우회하던 보안 문제가 수정되었습니다. 관련 이슈는 13613입니다.

    13613

영향 확인 (5)

  • breakingKubernetes 최소 버전 변경

    Longhorn v1.13.0을 설치하거나 업그레이드할 때 적용됩니다.

    CSI external-provisioner가 v6.3.0으로 업그레이드되므로, 모든 클러스터가 Kubernetes v1.34 이상을 실행해야 합니다.

  • breaking이전 V2 연결 복제 볼륨의 작업 제한

    Longhorn v1.12.0 이하에서 생성된 V2 linked-clone volumes를 사용하는 경우에 적용됩니다.

    v1.12.0 이하에서 생성된 V2 linked-clone volumes는 v1.12.1부터 분리하거나 삭제하는 작업만 지원합니다.

  • breakingV2 실시간 업그레이드 경로 제한

    적용 대상이 릴리스 노트에 명시되지 않았습니다.

    V2 live upgrade는 Longhorn v1.12.2에서만 v1.13.0으로 직접 지원됩니다. v1.12.0 또는 v1.12.1에서 업그레이드하거나 사전 조건을 충족하지 못한 경우에는 업그레이드 전에 모든 V2 볼륨을 분리하고 해당 복제본을 중지해야 합니다.

    prerequisites

  • breakinglonghorn-global-manager Deployment의 기본 복제본 수 변경

    적용 대상이 릴리스 노트에 명시되지 않았습니다.

    설치와 업그레이드 과정에서 longhorn-global-manager Deployment가 기본적으로 복제본 세 개로 생성됩니다. 업그레이드 전에 해당 파드 중 하나 이상을 스케줄링할 수 있어야 합니다.

  • breaking인스턴스 관리자 gRPC 서비스의 mTLS 적용 범위 확대

    longhorn-grpc-tls 시크릿을 설정한 경우에 적용됩니다.

    longhorn-grpc-tls 시크릿이 설정되면 모든 instance manager gRPC services에 mTLS가 적용됩니다. 따라서 모든 gRPC 포트에서 유효한 클라이언트 인증서를 요구합니다.

그 외 기록된 변경 118건 전체fixes 78 · additions 28 · value changes 9 · constraints 2 · defaults 1

fixes (78)

  • Longhorn v1.13.0 fixes a V2 linked-clone backup restore issue that could produce a corrupted volume when the source volume or the snapshot the clone was created from no longer existed.
  • Longhorn v1.13.0 fixes a CSI volume clone issue where cloning a volume with dataLocality: strict-local could fail with hard affinity cannot be satisfied.
  • Longhorn v1.13.0 fixes a node cleanup issue where a Longhorn node could not be removed after its Kubernetes node had been deleted without eviction.
  • Longhorn v1.13.0 fixes a backing image issue where copies were always transferred over IPv4, so on IPv6 single-stack and IPv6-first dual-stack clusters a backing image stayed at one copy.
  • [IMPROVEMENT] [CLI] preflight checks on Elemental3-based systems fails
  • [BUG] Runtime container images contain development (-devel) packages
  • [BUG] DR replicas break
  • [BUG] Longhorn may not reject expansion requests when the volume clone is in-progress
  • [BUG] Request Schema in Upgrade Responder is Outdate
  • [BUG] Disconnect Volume Node Network For More Than Pod Eviction Timeout While Workload Heavy Writing With RWX Fast Failover Disabled fails on v2 data engine
  • [BUG] deleting nodes.longhorn.io failed due to KubernetesNodeGone
  • [BUG] Cloned volume stuck "not ready for workloads" after the clone finishes (v1.12.0, v1 engine)
  • [BUG] SystemRestore job ignores the registry-secret setting
  • [BUG] After host restart one node no longer abe to start any replica
  • [BUG] v2 workload IO gets stuck in Continuous IO Test
  • [BUG] Terminal workload pods cause NodePublishVolume to abort with "no Pending workload pods"
  • [BUG] v2 volumes might get stuck in deleting state
  • [UI][BUG] The restore of V2 Linked Clone volumes misses src volume handling
  • [BUG] Sharded/EC volume: writes stall to ~0 and wedge the SPDK target on arm64/AIO, blocking volume deletion; plus CSI drops EC StorageClass params
  • [BUG] v2 volume gets stuck in Attaching in an IPv6 storage network environment
  • [BUG] CSI volume cloning doesn't respect node-/diskSelector
  • [BUG] [v1.13.0-rc1] V2 volume DD easy to stuck in V2 interrupt mode
  • [BUG] v1.12.1 default NetworkPolicies miss legitimate non-pod-labeled and host-originated callers, silently breaking CSI attach/detach
  • [BUG] Replica rebuiliding may fail with Input/output error on v2 volumes
  • [BUG] Creating default disks with createDefaultDiskLabeledNodes setting and create-default-disk label doesn't work with BDF
  • [BUG] invalid block disks should not block Longhorn system uninstallation
  • [BUG] The manifests in the /deploy directory of the longhorn/cli repository do not work due to an error in the command: field.
  • [BUG] V2 replica rebuild destination failure faults the entire engine
  • [BUG] v1 engine never starts: a stale stopped process record makes createInstance a silent no-op, volume stuck in attaching forever
  • [BUG] SPDK initialization fails on Talos OS with "nsenter: operation not permitted"
  • [BUG] V2 expansion can report success while the engine remains at the old size
  • [BUG] Recurring trim job fails with deadlock
  • [BUG] Kernel Workqueue Lockup and Unstable RKE2 Service After Enabling LH V2 in Harvester
  • [BUG] Failed to add v2 block disk with virtio-scsi BDF path
  • [BUG] Longhorn 1.12.0: AWS chunked encoding not supported with OCI S3 buckets
  • [BUG] Host OS nvmf-autoconnect connects kernel initiators to v2 replica subsystems, stalling volume attach/detach for minutes
  • [BUG] Adding V2 disk using the "Default Data Path" /dev/disk/by-id/scsi-* path fails
  • [BUG] backupvolume.longhorn.io not updated on cluster with nodes drained
  • [BUG] Volume expansion snapshot CR name expand-<size> collides across volumes
  • [BUG] System managed components don't respect System Managed Components Node Selector
  • [BUG] Recurring-job pods report success after startup or volume execution errors
  • [BUG] backupvolume.longhorn.io not created on cluster have nodes drained
  • [BUG] Backup not started on cluster have nodes drained
  • [BUG] v1 replica rebuild fails after a single gRPC TCP keepalive is lost
  • [BUG] v1.12.1 chart: new default-on networkPolicies.restrictInternalTraffic silently breaks Prometheus scraping of longhorn-manager (metrics/alerting blackout on patch upgrade)
  • [BUG] Longhorn UI is not running with PID 1
  • [BUG][v1.6.2-rc2] Volume failed to recover after nodes reboot, pod failed to mount the volume with Input/output error
  • [BUG] Snapshot CR is missing for backup CR of recurring job
  • [BUG] V1 volumes not rebuilding after cluster shutdown
  • [BUG] Incorrect Web Link in GUI
  • [BUG] V2 backup/snapshot can leave NVMe/TCP frontend or dm device stale, causing pod EIO on attached volumes
  • [BUG] Longhorn Helm Chart NetworkPolicies do not honor new RKE2 "rke2-traefik" ingress controller
  • [BUG] v2 volume repeated replica reuse failure
  • [BUG] Longhorn may try to attach volumes to a node without valid IM pod during the clone
  • [BUG] Backup Listing With More Than 1000 Backups fails on v2 volume due to an empty replica address in the backup status
  • [BUG] Interrupt Mode V2 stuck high CPU after workload completed
  • [BUG] Encrypted V2 volume size is 16MB short of the claimed size
  • [BUG] Transient SPDK lvol metadata failure can permanently fault a healthy v2 replica
  • [BUG] CSI pods do not respect anti-affinity preset update
  • [BUG] V2 encrypted volume keeps switching between Attaching and Detaching state after expand operation
  • [BUG] Fail to restore a v2 volume from a full backup if a previous backup is corrupted
  • [BUG] V2 Data Engine: UBLK fails with EINVAL on Linux kernel 6.17.0
  • [BUG] csi.*ReplicaCount Helm values silently no-op on existing csi-* deployments (only applied at first creation)
  • [BUG] v2 volume may crash again after the auto reattachment
  • [BUG] V2 Encrypted Volume Restore Fails
  • [BUG] (chart) ArgoCD OutOfSync when using Gateway API
  • [BUG] volume expansion stuck
  • [BUG] expanding the volume fails
  • [BUG] System Backup RecurringJob retention prunes newest CR — sorts by Status.CreatedAt (zero for Error/racing CRs)
  • [BUG] CSI components may have 0 running replica during upgrade
  • [BUG] Misconfigured and deleted Talos node causes Driver Deployer to constantly restart
  • [BUG] V2 instance-manager liveness probe errors ('test: -eq: unary operator expected') -> self-kill -> node-wide replica fault under rebuild load
  • [BUG] V2 Instance Manager panics in backupstore when S3 volume.cfg is missing, taking all node volumes offline
  • [BUG] V2 backup readers race snapshot cleanup and panic the instance manager
  • [BUG] Recurring snapshot-cleanup aborts the entire run when any volume's replica is rebuilding
  • [BUG] Longhorn block disk remains Ready=False and Schedulable=False when NVMe device is bound to vfio-pci
  • - [BUG] v2 volume stuck attaching with Storage Network enabled because the EngineFrontend target uses the engine pod IP instead of StorageIP [13351](https://github.com/longhorn/longhorn/issues/13351) - @yangchiu @c3y1huang
  • [BUG] Migration Engine Can Be Unexpectedly Deleted If the Target Node Is Still in Readiness Transition

additions (28)

  • Longhorn v1.13.0 supports live upgrade of V2 volumes, so the V2 instance managers can be upgraded without detaching the volumes, preventing disruption during Longhorn upgrades.
  • Starting with v1.13.0, V2 linked-clone volumes also support restore operations.
  • Introduced in v1.12.1, storage sharding is an experimental V2 Data Engine feature.
  • the data-engine-iobuf-large-pool-size and data-engine-iobuf-small-pool-size settings control the size of the SPDK buffer pools used by the V2 Data Engine.
  • Longhorn v1.13.0 makes it easier to protect related volumes by letting you snapshot a group of volumes with a single request from the Longhorn UI, kubectl, or Kubernetes VolumeGroupSnapshot objects through CSI.
  • Longhorn v1.13.0 adds an age-based retention policy for snapshot, backup, and system backup recurring jobs
  • Longhorn v1.13.0 adds the volumeTopology StorageClass parameter (any, zonal, or regional)
  • Longhorn v1.13.0 improves pod scheduling by allowing kube-scheduler to check actual Longhorn disk capacity and, when possible, place a restarted pod on the node that already holds all of its replicas.
  • Since v1.12.1, Longhorn creates ingress NetworkPolicy resources for its internal components by default.
  • - networkPolicies.v1DataEngineInitiatorSourceCIDRs and networkPolicies.recoveryBackendAdditionalIngressPorts allow traffic that the v1.12.1 policies blocked.
  • - networkPolicies.metricsScrapeSources lets Prometheus and other scrapers reach longhorn-manager metrics on TCP port 9500.
  • Longhorn v1.13.0 runs the CSI controller sidecars under a dedicated longhorn-csi-service-account instead of the shared longhorn-service-account.
  • You can turn this off with the csi.allowControllerSecretAccess Helm value after removing csi.storage.k8s.io/provisioner-secret-* parameters from Longhorn StorageClasses.
  • [FEATURE] Allow configuring Gateway API filters
  • [IMPROVEMENT] backup target availability status metric
  • [IMPROVEMENT] Add longhorn-cli command to check if the user's environment supports IOMMU
  • [IMPROVEMENT] Make ignoreSigningHeaders configurable
  • [IMPROVEMENT][Helm] Allow setting spec.trafficDistribution
  • [IMPROVEMENT] Add metrics to collect information about V2 data engine usage
  • [IMPROVEMENT] Add a pod filter to the Workload/Pod detail popup
  • [IMPROVEMENT] Expose topologySpreadConstraints for longhorn-ui in the Helm chart
  • [IMPROVEMENT] v2 volumes: allow configuring the NVMe-TCP initiator I/O queue count (nr-io-queues)
  • [IMPROVEMENT] Add usage metrics for volume size
  • [IMPROVEMENT] Add metrics to collect information about LONGHORN_DISTRO
  • [IMPROVEMENT] Exporting a volume from a single replica via crictl command
  • [IMPROVEMENT] Chart value to allow metrics scrapers through the longhorn-manager NetworkPolicy
  • [IMPROVEMENT] Help large replicas on full disks reclaim the space and squash all data into the heads
  • - [IMPROVEMENT] Add opt-in PodDisruptionBudget for the Longhorn UI Deployment [13466](https://github.com/longhorn/longhorn/issues/13466) - @yangchiu @somaz94

value changes (9)

  • interrupt mode for the V2 Data Engine is now fully event-driven. SPDK reactors wait for I/O events instead of polling continuously, with only low-frequency background checks remaining, significantly reducing CPU usage for idle or low I/O workloads.
  • [IMPROVEMENT] v2 data engine: gate the replica host-ACL restriction on cluster-wide instance manager version
  • [IMPROVEMENT] CSI ignores EC StorageClass parameters (silent replicated fallback)
  • [IMPROVEMENT] (chart) Precedence of image.*.*.registry
  • [IMPROVEMENT] Truncate the attached-to pod list in the volume table for volumes with many pods
  • [IMPROVEMENT] Steer host RPS away from SPDK reactor cores
  • [IMPROVEMENT] updateBackupCompressionMethod may write the Volume even when the method is unchanged
  • [IMPROVEMENT] Improving error transparency for volume attachment failure
  • [IMPROVEMENT] V2 volume write I/O stalls(~10s) when a replica is removed during migration

constraints (2)

  • [IMPROVEMENT] Support mTLS encrypted communication for remaining gRPC services in instance manager
  • - [FEATURE] Support Kubernetes CPU Manager for Longhorn V2 instance-manager SPDK CPU assignment [13248](https://github.com/longhorn/longhorn/issues/13248) - @yangchiu @mantissahz

defaults (1)

  • Longhorn v1.13.0 sets data-engine-cpu-isolation-enabled to {"v2":"true"} by default for new installations
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 →