Deploy - Important Notes - 《Longhorn v1.6.1 Documentation》

General
V2 Data Engine

Important Notes

This page lists important notes for Longhorn v1.6.1. Please see here for the full release note.

General
V2 Data Engine

General

Supported Kubernetes Versions

Please ensure your Kubernetes cluster is at least v1.21 before upgrading to Longhorn v1.6.1 because this is the minimum version Longhorn v1.6.1 supports.

Offline Upgrade Required To Fully Prevent Unexpected Replica Expansion

Longhorn v1.6.0 introduces a new mechanism to prevent unexpected replica expansion. This mechanism is entirely transparent. However, a volume is only protected if it is running a new version of longhorn-engine inside a new version of longhorn-instance-manager and managed by a new version of longhorn-manager. The live upgrade process results in a volume running a new version of longhorn-engine in an old version of longhorn-instance-manager until it is detached (by scaling its consuming workload down) and reattached (by scaling its consuming workload up). Consider scaling workloads down and back up again as soon as possible after upgrading from a version without this mechanism (v1.5.1 or older) to v1.6.1.

Default Priority Class

Longhorn v1.6.0 introduces the default Priority Class longhorn-critical, which has the highest value and ensures that Longhorn pods are not evicted by kube-scheduler when system resources are low.

During upgrades, Longhorn applies the default Priority Class to components depending on specific settings.

When all volumes are detached and you did not specify a value for the global Priority Class setting priority-class, the default Priority Class is applied to all Longhorn components. priority-class is updated.
When all volumes are detached and you specified a value for the global Priority Class setting priority-class, the default Priority Class is applied only to user-deployed components. priority-class is not updated.
When one or more volumes are attached and you did not specify a value for PriorityClass in the chart/value.yaml or longhorn/deploy/longhorn.yaml, the default Priority Class is applied only to user-deployed components. priority-class is not updated.

If you want to apply the default Priority Class to system-managed components, you must detach all volumes and change the Priority Class default setting value after the upgrade is successfully completed.

You can change these behaviors by customizing the following before starting the upgrade process:

For user deployed components: priorityClass parameters for each component in the values.yaml file of the Longhorn Helm chart
For system managed components: defaultSetting.priorityClass in the values.yaml file of the Longhorn Helm chart

New Node Drain Policies Added

There are two new options for the Node Drain Policy setting. Both Block For Eviction and Block for Eviction If Contains Last Replica automatically evict replicas from draining nodes in addition to preventing drain completion until volume data is sufficiently protected. Block for Eviction maintains maximum data redundancy during maintenance operations, and both new options enable automated cluster upgrades when some volumes have only one replica. See the new Node Drain Policy Recommendations section for help deciding which policy to use.

Custom Resource Fields Deprecated

Starting in v1.6.0, the following custom resource fields are deprecated. They will be removed in v1.7.0:

Volume.status.evictionRequested

Pod Security Policies Disabled & Pod Security Admission Introduction

Longhorn pods require privileged access to manage nodes’ storage. In Longhorn v1.3.x or older, Longhorn was shipping some Pod Security Policies by default, (e.g., link). However, Pod Security Policy has been deprecated since Kubernetes v1.21 and removed since Kubernetes v1.25, link. Therefore, we stopped shipping the Pod Security Policies by default. For Kubernetes < v1.25, if your cluster still enables Pod Security Policy admission controller, please do:
- Helm installation method: set the helm value enablePSP to true to install longhorn-psp PodSecurityPolicy resource which allows privileged Longhorn pods to start.
- Kubectl installation method: need to apply the podsecuritypolicy.yaml manifest in addition to applying the longhorn.yaml manifests.
- Rancher UI installation method: set Other Settings > Pod Security Policy to true to install longhorn-psp PodSecurityPolicy resource which allows privileged Longhorn pods to start.
As a replacement for Pod Security Policy, Kubernetes provides a new mechanism, Pod Security Admission. If you enable the Pod Security Admission controller and change the default behavior to block privileged pods, you must add the correct labels to the namespace where Longhorn pods run to allow Longhorn pods to start successfully (because Longhorn pods require privileged access to manage storage). For example, adding the following labels to the namespace that is running Longhorn pods:
```
apiVersion: v1
kind: Namespace
metadata:
  name: longhorn-system
  labels:
    pod-security.kubernetes.io/enforce: privileged
    pod-security.kubernetes.io/enforce-version: latest
    pod-security.kubernetes.io/audit: privileged
    pod-security.kubernetes.io/audit-version: latest
    pod-security.kubernetes.io/warn: privileged
    pod-security.kubernetes.io/warn-version: latest
```

Updating CSI Snapshot CRD `v1beta1` to `v1`, `v1beta1` Removed

Support for the v1beta1 version of CSI snapshot CRDs was previously deprecated in favor of the v1 version. The CSI components in Longhorn v1.6.1 only function with the v1 version. Please follow the instructions at Enable CSI Snapshot Support to update CSI snapshot CRDs and the CSI snapshot controller. If you have Longhorn volume manifests or scripts that are still using v1beta1 version, you must upgrade them to v1 as well.

Engine Upgrade Enforcement

Beginning with version v1.6.0, Longhorn is implementing mandatory engine upgrades. See the release note for information about the minimum supported engine image version.

When upgrading through Helm, a component compatibility check is automatically performed. If the new Longhorn is not compatible with the engine images that are currently in use, the upgrade path is blocked through a pre-hook mechanism.

If you installed Longhorn using the manifests, engine upgrades are enforced by the Longhorn Manager. Attempts to upgrade Longhorn Manager may cause unsuccessful pod launches and generate corresponding error logs, although it poses no harm. If you encounter such errors, you must revert to the previous Longhorn version and then upgrade the engines that are using the incompatible engine images before the next upgrade.

Warning: Whenever engine upgrade enforcement causes upgrade failure, Longhorn allows you to revert to the previous version because Longhorn Manager will block the entire upgrade. However, Longhorn prohibits downgrading when an upgrade is successful. For more information, see Upgrade Path Enforcement.

You can determine the versions of engine images that are currently in use with the following script:

#!/bin/bash
namespace="longhorn-system"
engine_images=$(kubectl -n $namespace get engineimage -o=jsonpath='{.items[*].metadata.name}')
for engine_image in $engine_images; do
    cli_api_version=$(kubectl -n $namespace get engineimage $engine_image -o=jsonpath='{.status.cliAPIVersion}')
    controller_api_version=$(kubectl -n $namespace get engineimage $engine_image -o=jsonpath='{.status.controllerAPIVersion}')
    echo "EngineImage: $engine_image | cliAPIVersion: $cli_api_version | controllerAPIVersion: $controller_api_version"
done

Once you successfully upgrade to version v1.6.0, you will be able to view information about engine image versions on the UI.

Danger Zone Setting Configuration

Starting with Longhorn v1.6.0, Longhorn allows you to modify the Danger Zone settings without the need to wait for all volumes to become detached. Your preferred settings are immediately applied in the following scenarios:

No attached volumes: When no volumes are attached before the settings are configured, the setting changes are immediately applied.
Engine image upgrade (live upgrade): During a live upgrade, which involves creating a new Instance Manager pod, the setting changes are immediately applied to the new pod.

Settings are synchronized hourly. When all volumes are detached, the settings in the following table are immediately applied and the system-managed components (for example, Instance Manager, CSI Driver, and engine images) are restarted. If you do not detach all volumes before the settings are synchronized, the settings are not applied and you must reconfigure the same settings after detaching the remaining volumes.

Setting	Additional Information	Affected Components
Kubernetes Taint Toleration	Taints and Tolerations	System-managed components
Priority Class	Priority Class	System-managed components
System Managed Components Node Selector	Node Selector	System-managed components
Storage Network	Storage Network	Instance Manager and Backing Image components
V1 Data Engine		Instance Manager component
V2 Data Engine	V2 Data Engine (Preview Feature)	Instance Manager component
Guaranteed Instance Manager CPU		Instance Manager component
Guaranteed Instance Manager CPU for V2 Data Engine		Instance Manager component

For V1 and V2 Data Engine settings, you can disable the Data Engines only when all associated volumes are detached. For example, you can disable the V2 Data Engine only when all V2 volumes are detached (even when V1 volumes are still attached).

Longhorn PVC with Block Volume Mode

Starting with v1.6.0, Longhorn is changing the default group ID of Longhorn devices from 0 (root group) to 6 (typically associated with the “disk” group). This change allows non-root containers to read or write to PVs using the Block volume mode. Note that Longhorn still keeps the owner of the Longhorn block devices as root. As a result, if your pod has security context such that it runs as non-root user and is part of the group id 0, the pod will no longer be able to read or write to Longhorn block volume mode PVC anymore. This use case should be very rare because running as a non-root user with the root group does not make much sense. More specifically, this example will not work anymore:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: longhorn-block-vol
spec:
  accessModes:
    - ReadWriteOnce
  volumeMode: Block
  storageClassName: longhorn
  resources:
    requests:
      storage: 2Gi
---
apiVersion: v1
kind: Pod
metadata:
  name: block-volume-test
  namespace: default
spec:
  securityContext:
    runAsGroup: 1000
    runAsNonRoot: true
    runAsUser: 1000
    supplementalGroups:
    - 0
  containers:
    - name: block-volume-test
      image: ubuntu:20.04
      command: ["sleep", "360000"]
      imagePullPolicy: IfNotPresent
      volumeDevices:
        - devicePath: /dev/longhorn/testblk
          name: block-vol
  volumes:
    - name: block-vol
      persistentVolumeClaim:
        claimName: longhorn-block-vol

From this version, you need to add group id 6 to the security context or run container as root. For more information, see Longhorn PVC ownership and permission

V2 Data Engine

Longhorn System Upgrade

Longhorn currently does not support live upgrading of V2 volumes. Ensure that all V2 volumes are detached before initiating the upgrade process.

Changing Default Huge Page Size to 2 GiB

The default huge page size for the V2 Data Engine has been raised to 2 GiB, allowing the creation of more V2 volumes and enhancing the overall user experience. Before upgrading to v1.6.0, ensure that the configured huge page size on each node is 2 GiB.

Support for ARM64 Platform

As of Longhorn v1.6.0, volumes using the V2 Data Engine support the ARM64 platform. For more information, see Prerequisites.

Replacing `backendStoreDriver` with `dataEngine`

The attribute backendStoreDriver, which is defined in the parameters of StorageClasses and other Longhorn resources (for example, volumes, engines, and replicas), has been replaced with dataEngine. You must remove the existing StorageClasses for V2 volumes and create new ones that use dataEngine.

Updating the Linux Kernel on Longhorn Nodes

Host machines with Linux kernel 5.15 may unexpectedly reboot when volume-related IO errors occur. Update the Linux kernel on Longhorn nodes to version 5.19 or later to prevent such issues. For more information, see Prerequisites.