Ceph

    PLEASE NOTE: This document applies to v1.4 version and not to the latest stable release v1.7

    Documentation for other releases can be found by using the version selector in the top right of any doc page.

    Rook-Ceph Upgrades

    This guide will walk you through the steps to upgrade the software in a Rook-Ceph cluster from one version to the next. This includes both the Rook-Ceph operator software itself as well as the Ceph cluster software.

    Upgrades for both the operator and for Ceph are nearly entirely automated save for where Rook’s permissions need to be explicitly updated by an admin or when incompatibilities need to be addressed manually due to customizations.

    We welcome feedback and opening issues!

    Supported Versions

    This guide is for upgrading from Rook v1.3.x to Rook v1.4.x.

    Please refer to the upgrade guides from previous releases for supported upgrade paths. Rook upgrades are only supported between official releases. Upgrades to and from master are not supported.

    For a guide to upgrade previous versions of Rook, please refer to the version of documentation for those releases.

    Considerations

    With this upgrade guide, there are a few notes to consider:

    • WARNING: Upgrading a Rook cluster is not without risk. There may be unexpected issues or obstacles that damage the integrity and health of your storage cluster, including data loss.
    • The Rook cluster’s storage may be unavailable for short periods during the upgrade process for both Rook operator updates and for Ceph version updates.
    • We recommend that you read this document in full before you undertake a Rook cluster upgrade.

    Before you Upgrade

    Rook v1.4 has a breaking change that should be considered before upgrading.

    1. Make sure the lvm2 package is installed on the host where OSDs are running. If not, the prepare job will fail and the upgrade of OSDs will be blocked.
      • This only applies to non-PVC OSDs. Beginning in v1.3, OSDs created on PVCs no longer rely on LVM.
    2. CSI Snapshots: See the next section if you are currently using Ceph-CSI snapshots.

    CSI Snapshots

    CSI snapshots have moved from Alpha to Beta and are not backward compatible. The snapshots created with the Alpha version must be deleted before the upgrade.

    If you desire to continue using Ceph-CSI 2.x with Alpha snapshots, then:

    • Skip this section
    • Set the CSI version to 2.x as described below in the CSI Version section.

    To continue with the Ceph-CSI v3.0 driver that is recommended with v1.4:

    1. List all the volumesnapshots created
    kubectl get volumesnapshot
    NAME               AGE
    rbd-pvc-snapshot   22s
    
    1. Delete all volumesnapshots
    kubectl delete volumesnapshot rbd-pvc-snapshot
    volumesnapshot.snapshot.storage.k8s.io "rbd-pvc-snapshot" deleted
    
    1. List all volumesnapshotclasses created
    kubectl get volumesnapshotclass
    NAME                      AGE
    csi-rbdplugin-snapclass   86s
    
    1. Delete all volumesnapshotclasses
    kubectl delete volumesnapshotclass csi-rbdplugin-snapclass
    volumesnapshotclass.snapshot.storage.k8s.io "csi-rbdplugin-snapclass" deleted
    

    Note: The underlying snapshots on the storage system will be deleted by ceph-csi

    Delete the Alpha CRDs

    As we are updating the snapshot resources from Alpha to Beta we need to delete the old alphav1 snapshot CRD created by external-snapshotter sidecar container

    Check if we have any v1alpha1 CRD created in our kubernetes cluster

    kubectl get crd volumesnapshotclasses.snapshot.storage.k8s.io -o yaml |grep v1alpha1
      - name: v1alpha1
      - v1alpha1
    kubectl get crd volumesnapshotcontents.snapshot.storage.k8s.io -o yaml |grep v1alpha1
      - name: v1alpha1
      - v1alpha1
    kubectl get crd volumesnapshots.snapshot.storage.k8s.io -o yaml |grep v1alpha1
      - name: v1alpha1
      - v1alpha1
    

    As we have v1alpha1 CRD created in our kubernetes cluster, we need to delete the Alpha CRD

    kubectl delete crd volumesnapshotclasses.snapshot.storage.k8s.io volumesnapshotcontents.snapshot.storage.k8s.io volumesnapshots.snapshot.storage.k8s.io
    
    customresourcedefinition.apiextensions.k8s.io "volumesnapshotclasses.snapshot.storage.k8s.io" deleted
    customresourcedefinition.apiextensions.k8s.io "volumesnapshotcontents.snapshot.storage.k8s.io" deleted
    customresourcedefinition.apiextensions.k8s.io "volumesnapshots.snapshot.storage.k8s.io" deleted
    

    Finally, if you desire to use the beta snapshots, check that the prerequisites are met.

    Upgrading the Rook-Ceph Operator

    Patch Release Upgrades

    Unless otherwise noted due to extenuating requirements, upgrades from one patch release of Rook to another are as simple as updating the image of the Rook operator. For example, when Rook v1.4.9 is released, the process of updating from v1.4.x is as simple as running the following:

    kubectl -n rook-ceph set image deploy/rook-ceph-operator rook-ceph-operator=rook/ceph:v1.4.9
    

    Helm Upgrades

    If you have installed Rook via the Helm chart, Helm will handle some details of the upgrade for you. In particular, Helm will handle updating the RBAC and trigger the operator update and restart.

    Upgrading from v1.3 to v1.4

    Rook releases from master are expressly unsupported. It is strongly recommended that you use official releases of Rook. Unreleased versions from the master branch are subject to changes and incompatibilities that will not be supported in the official releases. Builds from the master branch can have functionality changed and even removed at any time without compatibility support and without prior notice.

    Prerequisites

    We will do all our work in the Ceph example manifests directory.

    cd $YOUR_ROOK_REPO/cluster/examples/kubernetes/ceph/
    

    Unless your Rook cluster was created with customized namespaces, namespaces for Rook clusters created before v0.8 are likely to be:

    • Clusters created by v0.7 or earlier: rook-system and rook
    • Clusters created in v0.8 or v0.9: rook-ceph-system and rook-ceph
    • Clusters created in v1.0 or newer: only rook-ceph

    With this guide, we do our best not to assume the namespaces in your cluster. To make things as easy as possible, modify and use the below snippet to configure your environment. We will use these environment variables throughout this document.

    # Parameterize the environment
    export ROOK_SYSTEM_NAMESPACE="rook-ceph"
    export ROOK_NAMESPACE="rook-ceph"
    

    In order to successfully upgrade a Rook cluster, the following prerequisites must be met:

    • The cluster should be in a healthy state with full functionality. Review the health verification section in order to verify your cluster is in a good starting state.
    • All pods consuming Rook storage should be created, running, and in a steady state. No Rook persistent volumes should be in the act of being created or deleted.

    Helm

    • Your Helm version should be newer than v3.2.0 to avoid this issue.
    • For a Rook cluster already deployed with Helm older than v3.2.0, also execute the following commands.
    KIND=ClusterRole
    NAME=psp:rook
    RELEASE=your-apps-release-name
    NAMESPACE=your-apps-namespace
    kubectl annotate $KIND $NAME meta.helm.sh/release-name=$RELEASE
    kubectl annotate $KIND $NAME meta.helm.sh/release-namespace=$NAMESPACE
    kubectl label $KIND $NAME app.kubernetes.io/managed-by=Helm
    

    Health Verification

    Before we begin the upgrade process, let’s first review some ways that you can verify the health of your cluster, ensuring that the upgrade is going smoothly after each step. Most of the health verification checks for your cluster during the upgrade process can be performed with the Rook toolbox. For more information about how to run the toolbox, please visit the Rook toolbox readme.

    See the common issues pages for troubleshooting and correcting health issues:

    Pods all Running

    In a healthy Rook cluster, the operator, the agents and all Rook namespace pods should be in the Running state and have few, if any, pod restarts. To verify this, run the following commands:

    kubectl -n $ROOK_NAMESPACE get pods
    

    Status Output

    The Rook toolbox contains the Ceph tools that can give you status details of the cluster with the ceph status command. Let’s look at an output sample and review some of the details:

    TOOLS_POD=$(kubectl -n $ROOK_NAMESPACE get pod -l "app=rook-ceph-tools" -o jsonpath='{.items[0].metadata.name}')
    kubectl -n $ROOK_NAMESPACE exec -it $TOOLS_POD -- ceph status
    
      cluster:
        id:     a3f4d647-9538-4aff-9fd1-b845873c3fe9
        health: HEALTH_OK
    
      services:
        mon: 3 daemons, quorum b,c,a
        mgr: a(active)
        mds: myfs-1/1/1 up  {0=myfs-a=up:active}, 1 up:standby-replay
        osd: 6 osds: 6 up, 6 in
        rgw: 1 daemon active
    
      data:
        pools:   9 pools, 900 pgs
        objects: 67  objects, 11 KiB
        usage:   6.1 GiB used, 54 GiB / 60 GiB avail
        pgs:     900 active+clean
    
      io:
        client:   7.4 KiB/s rd, 681 B/s wr, 11 op/s rd, 4 op/s wr
        recovery: 164 B/s, 1 objects/s
    

    In the output above, note the following indications that the cluster is in a healthy state:

    • Cluster health: The overall cluster status is HEALTH_OK and there are no warning or error status messages displayed.
    • Monitors (mon): All of the monitors are included in the quorum list.
    • Manager (mgr): The Ceph manager is in the active state.
    • OSDs (osd): All OSDs are up and in.
    • Placement groups (pgs): All PGs are in the active+clean state.
    • (If applicable) Ceph filesystem metadata server (mds): all MDSes are active for all filesystems
    • (If applicable) Ceph object store RADOS gateways (rgw): all daemons are active

    If your ceph status output has deviations from the general good health described above, there may be an issue that needs to be investigated further. There are other commands you may run for more details on the health of the system, such as ceph osd status. See the Ceph troubleshooting docs for help.

    Rook will prevent the upgrade of the Ceph daemons if the health is in a HEALTH_ERR state. If you desired to proceed with the upgrade anyway, you will need to set either skipUpgradeChecks: true or continueUpgradeAfterChecksEvenIfNotHealthy: true as described in the cluster CR settings.

    Container Versions

    The container version running in a specific pod in the Rook cluster can be verified in its pod spec output. For example for the monitor pod mon-b, we can verify the container version it is running with the below commands:

    POD_NAME=$(kubectl -n $ROOK_NAMESPACE get pod -o custom-columns=name:.metadata.name --no-headers | grep rook-ceph-mon-a)
    kubectl -n $ROOK_NAMESPACE get pod ${POD_NAME} -o jsonpath='{.spec.containers[0].image}'
    

    The status and container versions for all Rook pods can be collected all at once with the following commands:

    kubectl -n $ROOK_SYSTEM_NAMESPACE get pod -o jsonpath='{range .items[*]}{.metadata.name}{"\n\t"}{.status.phase}{"\t\t"}{.spec.containers[0].image}{"\t"}{.spec.initContainers[0]}{"\n"}{end}' && \
    kubectl -n $ROOK_NAMESPACE get pod -o jsonpath='{range .items[*]}{.metadata.name}{"\n\t"}{.status.phase}{"\t\t"}{.spec.containers[0].image}{"\t"}{.spec.initContainers[0].image}{"\n"}{end}'
    

    The rook-version label exists on Ceph controller resources. For various resource controllers, a summary of the resource controllers can be gained with the commands below. These will report the requested, updated, and currently available replicas for various Rook-Ceph resources in addition to the version of Rook for resources managed by the updated Rook-Ceph operator. Note that the operator and toolbox deployments do not have a rook-version label set.

    kubectl -n $ROOK_NAMESPACE get deployments -o jsonpath='{range .items[*]}{.metadata.name}{"  \treq/upd/avl: "}{.spec.replicas}{"/"}{.status.updatedReplicas}{"/"}{.status.readyReplicas}{"  \trook-version="}{.metadata.labels.rook-version}{"\n"}{end}'
    
    kubectl -n $ROOK_NAMESPACE get jobs -o jsonpath='{range .items[*]}{.metadata.name}{"  \tsucceeded: "}{.status.succeeded}{"      \trook-version="}{.metadata.labels.rook-version}{"\n"}{end}'
    

    Rook Volume Health

    Any pod that is using a Rook volume should also remain healthy:

    • The pod should be in the Running state with few, if any, restarts
    • There should be no errors in its logs
    • The pod should still be able to read and write to the attached Rook volume.

    Rook Operator Upgrade Process

    In the examples given in this guide, we will be upgrading a live Rook cluster running v1.3.9 to the version v1.4.9. This upgrade should work from any official patch release of Rook v1.3 to any official patch release of v1.4.

    Rook release from master are expressly unsupported. It is strongly recommended that you use official releases of Rook. Unreleased versions from the master branch are subject to changes and incompatibilities that will not be supported in the official releases. Builds from the master branch can have functionality changed or removed at any time without compatibility support and without prior notice.

    Let’s get started!

    1. Remove Alpha Snapshots

    As described above in the CSI Snapshots section, if you are using alpha snapshots it is required to remove them before the upgrade since they are not compatible with beta snaphots available with Ceph-CSI v3.0.

    2. Update the RBAC and CRDs

    Automatically updated if you are upgrading via the helm chart

    First apply new resources. This includes modified privileges (RBAC) needed by the Operator and updates to the Custom Resource Definitions (CRDs).

    If you are not using the default rook-ceph namespace, replace the namespace in the following manifest:

    sed -i "s/namespace: rook-ceph/namespace: $ROOK_SYSTEM_NAMESPACE/g" upgrade-from-v1.3-apply.yaml
    

    Now apply the updated privileges:

    kubectl delete -f upgrade-from-v1.3-delete.yaml
    kubectl apply -f upgrade-from-v1.3-apply.yaml -f upgrade-from-v1.3-crds.yaml
    

    3. Update Ceph CSI version to v3.0

    Rook v1.4 will install Ceph-CSI v3.0 use the latest drivers by default. If you have not specified custom CSI images in the Operator deployment this step is unnecessary.

    If you have specified custom CSI images in the Rook-Ceph Operator deployment, it is recommended to update to use the latest Ceph-CSI v3.0 driver. See the section CSI Version for more details.

    4. Update the Rook Operator

    Automatically updated if you are upgrading via the helm chart

    The largest portion of the upgrade is triggered when the operator’s image is updated to v1.4.x. When the operator is updated, it will proceed to update all of the Ceph daemons.

    kubectl -n $ROOK_SYSTEM_NAMESPACE set image deploy/rook-ceph-operator rook-ceph-operator=rook/ceph:v1.4.9
    

    5. Wait for the upgrade to complete

    Watch now in amazement as the Ceph mons, mgrs, OSDs, rbd-mirrors, MDSes and RGWs are terminated and replaced with updated versions in sequence. The cluster may be offline very briefly as mons update, and the Ceph Filesystem may fall offline a few times while the MDSes are upgrading. This is normal.

    The versions of the components can be viewed as they are updated:

    watch --exec kubectl -n $ROOK_NAMESPACE get deployments -l rook_cluster=$ROOK_NAMESPACE -o jsonpath='{range .items[*]}{.metadata.name}{"  \treq/upd/avl: "}{.spec.replicas}{"/"}{.status.updatedReplicas}{"/"}{.status.readyReplicas}{"  \trook-version="}{.metadata.labels.rook-version}{"\n"}{end}'
    

    As an example, this cluster is midway through updating the OSDs from v1.3 to v1.4. When all deployments report 1/1/1 availability and rook-version=v1.4.9, the Ceph cluster’s core components are fully updated.

    Every 2.0s: kubectl -n rook-ceph get deployment -o j...
    
    rook-ceph-mgr-a         req/upd/avl: 1/1/1      rook-version=v1.4.9
    rook-ceph-mon-a         req/upd/avl: 1/1/1      rook-version=v1.4.9
    rook-ceph-mon-b         req/upd/avl: 1/1/1      rook-version=v1.4.9
    rook-ceph-mon-c         req/upd/avl: 1/1/1      rook-version=v1.4.9
    rook-ceph-osd-0         req/upd/avl: 1//        rook-version=v1.4.9
    rook-ceph-osd-1         req/upd/avl: 1/1/1      rook-version=v1.3.9
    rook-ceph-osd-2         req/upd/avl: 1/1/1      rook-version=v1.3.9
    

    An easy check to see if the upgrade is totally finished is to check that there is only one rook-version reported across the cluster.

    # kubectl -n $ROOK_NAMESPACE get deployment -l rook_cluster=$ROOK_NAMESPACE -o jsonpath='{range .items[*]}{"rook-version="}{.metadata.labels.rook-version}{"\n"}{end}' | sort | uniq
    This cluster is not yet finished:
      rook-version=v1.3.9
      rook-version=v1.4.9
    This cluster is finished:
      rook-version=v1.4.9
    

    6. Verify the updated cluster

    At this point, your Rook operator should be running version rook/ceph:v1.4.9.

    Verify the Ceph cluster’s health using the health verification section.

    Ceph Version Upgrades

    Rook v1.4 supports Ceph Nautilus 14.2.5 or newer and Ceph Octopus v15.2.0 or newer. These are the only supported major versions of Ceph.

    IMPORTANT: When an update is requested, the operator will check Ceph’s status, if it is in HEALTH_ERR it will refuse to do the upgrade.

    Rook is cautious when performing upgrades. When an upgrade is requested (the Ceph image has been updated in the CR), Rook will go through all the daemons one by one and will individually perform checks on them. It will make sure a particular daemon can be stopped before performing the upgrade. Once the deployment has been updated, it checks if this is ok to continue. After each daemon is updated we wait for things to settle (monitors to be in a quorum, PGs to be clean for OSDs, up for MDSs, etc.), then only when the condition is met we move to the next daemon. We repeat this process until all the daemons have been updated.

    Ceph images

    Official Ceph container images can be found on Docker Hub. These images are tagged in a few ways:

    • The most explicit form of tags are full-ceph-version-and-build tags (e.g., v15.2.8-20201201). These tags are recommended for production clusters, as there is no possibility for the cluster to be heterogeneous with respect to the version of Ceph running in containers.
    • Ceph major version tags (e.g., v15) are useful for development and test clusters so that the latest version of Ceph is always available.

    Ceph containers other than the official images from the registry above will not be supported.

    Example upgrade to Ceph Octopus

    1. Update the main Ceph daemons

    The majority of the upgrade will be handled by the Rook operator. Begin the upgrade by changing the Ceph image field in the cluster CRD (spec.cephVersion.image).

    NEW_CEPH_IMAGE='ceph/ceph:v15.2.8-20201217'
    CLUSTER_NAME="$ROOK_CLUSTER_NAMESPACE"  # change if your cluster name is not the Rook namespace
    kubectl -n $ROOK_CLUSTER_NAMESPACE patch CephCluster $CLUSTER_NAME --type=merge -p "{\"spec\": {\"cephVersion\": {\"image\": \"$NEW_CEPH_IMAGE\"}}}"
    

    2. Wait for the daemon pod updates to complete

    As with upgrading Rook, you must now wait for the upgrade to complete. Status can be determined in a similar way to the Rook upgrade as well.

    watch --exec kubectl -n $ROOK_NAMESPACE get deployments -l rook_cluster=$ROOK_NAMESPACE -o jsonpath='{range .items[*]}{.metadata.name}{"  \treq/upd/avl: "}{.spec.replicas}{"/"}{.status.updatedReplicas}{"/"}{.status.readyReplicas}{"  \tceph-version="}{.metadata.labels.ceph-version}{"\n"}{end}'
    

    Determining when the Ceph has fully updated is rather simple.

    # kubectl -n $ROOK_NAMESPACE get deployment -l rook_cluster=$ROOK_NAMESPACE -o jsonpath='{range .items[*]}{"ceph-version="}{.metadata.labels.ceph-version}{"\n"}{end}' | sort | uniq
    This cluster is not yet finished:
        ceph-version=14.2.7-0
        ceph-version=15.2.4-0
    This cluster is finished:
        ceph-version=15.2.4-0
    

    3. Verify the updated cluster

    Verify the Ceph cluster’s health using the health verification section.

    CSI Version

    If you have a cluster running with CSI drivers enabled and you want to configure Rook to use non-default CSI images, the following settings will need to be applied for the desired version of CSI.

    The operator configuration variables have recently moved from the operator deployment to the rook-ceph-operator-config ConfigMap. The values in the operator deployment can still be set, but if the ConfigMap settings are applied, they will override the operator deployment settings.

    If the cluster was originally installed prior to v1.3, the configmap may not exist. See the latest operator.yaml for example configmap settings.

    kubectl -n $ROOK_NAMESPACE edit configmap rook-ceph-operator-config
    

    The default upstream images are included below, which you can change to your desired images.

    ROOK_CSI_CEPH_IMAGE: "quay.io/cephcsi/cephcsi:v3.1.1"
    ROOK_CSI_REGISTRAR_IMAGE: "quay.io/k8scsi/csi-node-driver-registrar:v1.2.0"
    ROOK_CSI_PROVISIONER_IMAGE: "quay.io/k8scsi/csi-provisioner:v1.6.0"
    ROOK_CSI_SNAPSHOTTER_IMAGE: "quay.io/k8scsi/csi-snapshotter:v2.1.1"
    ROOK_CSI_ATTACHER_IMAGE: "quay.io/k8scsi/csi-attacher:v2.1.0"
    ROOK_CSI_RESIZER_IMAGE: "quay.io/k8scsi/csi-resizer:v0.4.0"
    

    Use default images

    If you would like Rook to use the inbuilt default upstream images, then you may simply remove all variables matching ROOK_CSI_*_IMAGE from the above ConfigMap and/or the operator deployment.

    Verifying updates

    You can use the below command to see the CSI images currently being used in the cluster.

    # kubectl --namespace rook-ceph get pod -o jsonpath='{range .items[*]}{range .spec.containers[*]}{.image}{"\n"}' -l 'app in (csi-rbdplugin,csi-rbdplugin-provisioner,csi-cephfsplugin,csi-cephfsplugin-provisioner)' | sort | uniq
    quay.io/cephcsi/cephcsi:v3.1.1
    quay.io/k8scsi/csi-attacher:v2.1.0
    quay.io/k8scsi/csi-node-driver-registrar:v1.2.0
    quay.io/k8scsi/csi-provisioner:v1.6.0
    quay.io/k8scsi/csi-resizer:v0.4.0
    quay.io/k8scsi/csi-snapshotter:v2.1.1
    quay.io/k8scsi/csi-resizer:v0.4.0
    

    Replace lvm mode OSDs with raw mode (if you use LV-backed PVC)

    For LV-backed PVC, we recommend replacing lvm mode OSDs with raw mode OSDs. See common issue.