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.

With the release of Rook 1.0, 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. Achieving the level of upgrade automation has been refined by community feedback, and we will always be open to further feedback for improving automation and improving Rook.

We welcome feedback and opening issues!

Supported Versions

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. Only proceed with this guide if you are comfortable with that.
  • 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.

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 v1.2.7 is released, the process of updating from v1.2.0 is as simple as running the following:

  1. kubectl -n rook-ceph set image deploy/rook-ceph-operator rook-ceph-operator=rook/ceph:v1.2.7

If you want to enable the CSI v2.0 driver, additional steps are needed in the latest Rook patch release. There are some new features in 2.0 such as supporting resizing volumes and support for a new csi attacher. If you don’t need these features now, you can skip this step for the patch upgrade.

Enabling the CSI v2.0 driver

If you are deploying the rook-operator with the helm chart you only need to specify the related images in your values.yaml. The RBAC settings will be created for you by the chart.

To enable the CSI v2.0 driver, you will need to apply updated RBAC settings:

  1. kubectl apply -f enable-csi-2.0-rbac.yaml

The following env variables will need to be configured in the Rook-Ceph operator deployment. The suggested upstream images are included below, which you should change to match where your images are located.

  1.   env:
  2.     - name: ROOK_CSI_CEPH_IMAGE
  3.         value: "quay.io/cephcsi/cephcsi:v2.0.0"
  4.     - name: ROOK_CSI_REGISTRAR_IMAGE
  5.         value: "quay.io/k8scsi/csi-node-driver-registrar:v1.2.0"
  6.     - name: ROOK_CSI_PROVISIONER_IMAGE
  7.         value: "quay.io/k8scsi/csi-provisioner:v1.4.0"
  8.     - name: ROOK_CSI_SNAPSHOTTER_IMAGE
  9.         value: "quay.io/k8scsi/csi-snapshotter:v1.2.2"
  10.     - name: ROOK_CSI_ATTACHER_IMAGE
  11.         value: "quay.io/k8scsi/csi-attacher:v2.1.0"
  12.     - name: ROOK_CSI_RESIZER_IMAGE
  13.         value: "quay.io/k8scsi/csi-resizer:v0.4.0"

Upgrading from v1.1 to v1.2

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.

  1. 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.

  1. # Parameterize the environment
  2. export ROOK_SYSTEM_NAMESPACE="rook-ceph"
  3. 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.

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:

  1. kubectl -n $ROOK_SYSTEM_NAMESPACE get pods
  2. 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:

  1. TOOLS_POD=$(kubectl -n $ROOK_NAMESPACE get pod -l "app=rook-ceph-tools" -o jsonpath='{.items[0].metadata.name}')
  2. kubectl -n $ROOK_NAMESPACE exec -it $TOOLS_POD -- ceph status
  1.   cluster:
  2.     id:     a3f4d647-9538-4aff-9fd1-b845873c3fe9
  3.     health: HEALTH_OK
  5.   services:
  6.     mon: 3 daemons, quorum b,c,a
  7.     mgr: a(active)
  8.     mds: myfs-1/1/1 up  {0=myfs-a=up:active}, 1 up:standby-replay
  9.     osd: 6 osds: 6 up, 6 in
  10.     rgw: 1 daemon active
  12.   data:
  13.     pools:   9 pools, 900 pgs
  14.     objects: 67  objects, 11 KiB
  15.     usage:   6.1 GiB used, 54 GiB / 60 GiB avail
  16.     pgs:     900 active+clean
  18.   io:
  19.     client:   7.4 KiB/s rd, 681 B/s wr, 11 op/s rd, 4 op/s wr
  20.     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.

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:

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

All Pods Status and Version

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

  1. 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}' && \
  2. 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.

  1. 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}'
  3. 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 v1.1.9 to the version v1.2.7. This upgrade should work from any official patch release of Rook v1.1 to any official patch release of v1.2. We will further assume that your previous cluster was created using an earlier version of this guide and manifests. If you have created custom manifests, these steps may not work as written.

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. Update the RBAC and CRDs

First update the Ceph Custom Resource Definitions and the privileges (RBAC) needed by the operator. A new CephClient CRD is included in v1.2 and the CSI driver privileges changed slightly.

  1. kubectl apply -f upgrade-from-v1.1-apply.yaml

2. CSI upgrade pre-requisites

In some scenarios there is an issue in the CSI driver that will cause application pods to be disconnected from their mounts when the CSI driver is restarted. Since the upgrade would cause the CSI driver to restart if it is updated, you need to be aware of whether this affects your applications. This issue will happen when using the Ceph fuse client or rbd-nbd:

  • CephFS: If you are provision volumes for CephFS and have a kernel less than version 4.17, The CSI driver will fall back to use the FUSE client.
  • RBD: If you have set the mounter: rbd-nbd option in the RBD storage class, the NBD mounter will have this issue. This setting is not enabled by default.

If you are affected by this issue, you will need to proceed carefully during the upgrade to restart your application pods. The first recommended step is to modify the update strategy of the CSI driver so that you can control when the CSI driver pods are restarted on each node. As seen in the example below, set CSI_CEPHFS_PLUGIN_UPDATE_STRATEGY and CSI_RBD_PLUGIN_UPDATE_STRATEGY values to OnDelete.

To avoid this issue in future upgrades, we recommend that you do not use the fuse client or rbd-nbd. The fuse client can be avoided by setting the following environment variables in the operator now before this upgrade. The side effect of this setting is that the PVC size will not be enforced if the CephFS quotas are not supported in your kernel version. To enable this option and avoid this upgrade issue in the future, set CSI_FORCE_CEPHFS_KERNEL_CLIENT to true.

  1. kubectl -n $ROOK_SYSTEM_NAMESPACE edit deploy rook-ceph-operator
  1.   # If you set this image at the same time as the env variables, you can skip step 3 of this guide and avoid a second operator restart
  2.   image: rook/ceph:v1.2.7
  3.   env:
  4.     # Change the update strategy for the CephFS driver if your cluster is affected
  5.     - name: CSI_CEPHFS_PLUGIN_UPDATE_STRATEGY
  6.       value: "OnDelete"
  7.     # Change the update strategy for the RBD driver if your cluster is affected
  8.     - name: CSI_RBD_PLUGIN_UPDATE_STRATEGY
  9.       value: "OnDelete"
  10.     # To avoid this upgrade issue in the future for CephFS, force enable the kernel client
  11.     - name: CSI_FORCE_CEPHFS_KERNEL_CLIENT
  12.       value: "true"

After the operator and cluster are updated, we will continue with the CSI and application pod restarts in Step 5.

3. Update the Rook Operator

The largest portion of the upgrade is triggered when the operator’s image is updated to v1.2.x. When the operator is updated, it will proceed to update all of the Ceph daemons. (If step 1 was completed, this change has already been applied.)

  1. kubectl -n $ROOK_SYSTEM_NAMESPACE set image deploy/rook-ceph-operator rook-ceph-operator=rook/ceph:v1.2.7

4. 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. Continue on to the next upgrade step while the update is commencing.

Before moving on, the Ceph cluster’s core (RADOS) components (i.e., mons, mgrs, and OSDs) must be fully updated.

  1. 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 v1.1 to v1.2. When all deployments report 1/1/1 availability and rook-version=v1.2.7, the Ceph cluster’s core components are fully updated.

  1. Every 2.0s: kubectl -n rook-ceph get deployment -o j...
  3. rook-ceph-mgr-a         req/upd/avl: 1/1/1      rook-version=v1.2.7
  4. rook-ceph-mon-a         req/upd/avl: 1/1/1      rook-version=v1.2.7
  5. rook-ceph-mon-b         req/upd/avl: 1/1/1      rook-version=v1.2.7
  6. rook-ceph-mon-c         req/upd/avl: 1/1/1      rook-version=v1.2.7
  7. rook-ceph-osd-0         req/upd/avl: 1//        rook-version=v1.2.7
  8. rook-ceph-osd-1         req/upd/avl: 1/1/1      rook-version=v1.1.9
  9. rook-ceph-osd-2         req/upd/avl: 1/1/1      rook-version=v1.1.9

The MDSes and RGWs are the last daemons to update. 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. It is safe to proceed with the next step before the MDSes and RGWs are finished updating.

  1. # 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
  2. This cluster is not yet finished:
  3.   rook-version=v1.1.9
  4.   rook-version=v1.2.7
  5. This cluster is finished:
  6.   rook-version=v1.2.7

5. Verify the updated cluster

At this point, your Rook operator should be running version rook/ceph:v1.2.7.

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

6. CSI Manual Update (optional)

If you determined in step 1 that you were affected by the CSI driver restart issue that disconnects the application pods from their mounts, continue with this section. Otherwise, you can skip to step 7.

Your cluster should now be in a state where Rook has upgraded everything except the CSI driver. The CSI driver pods will not be updated until you delete them manually. This allows you to control when your application pods will be affected by the CSI driver restart.

For each node:

  • Drain your application pods from the node
  • Delete the CSI driver pods on the node
    • The pods to delete will be named with a csi-cephfsplugin or csi-rbdplugin prefix and have a random suffix on each node. However, no need to delete the provisioner pods: csi-cephfsplugin-provisioner-* or csi-rbdplugin-provisioner-*
    • The pod deletion causes the pods to be restarted and updated automatically on the node

7. Update Rook-Ceph custom resource definitions

IMPORTANT: Do not perform this step until ALL existing Rook-Ceph clusters are updated!

After all Rook-Ceph clusters have been updated following the steps above, update the Rook-Ceph Custom Resource Definitions. This will help with creating or modifying Rook-Ceph deployments in the future with the updated schema validation.

  1. kubectl apply -f upgrade-from-v1.1-crds.yaml

Ceph Version Upgrades

Rook v1.2 supports Ceph Mimic v13.2.4 or newer and Ceph Nautilus v14.2.0 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., v14.2.5-20191210). 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., v14) 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 Nautilus

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).

  1. NEW_CEPH_IMAGE='ceph/ceph:v14.2.5-20191210'
  2. CLUSTER_NAME="$ROOK_NAMESPACE"  # change if your cluster name is not the Rook namespace
  3. kubectl -n $ROOK_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.

  1. 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.

  1. # 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
  2. This cluster is not yet finished:
  3.     ceph-version=13.2.6
  4.     ceph-version=14.2.5-0
  5. This cluster is finished:
  6.     ceph-version=14.2.5-0

Note that Ceph version now includes an additional build tag, the -0 suffix in the example above.

3. Verify the updated cluster

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

If you see a health warning about enabling msgr2, please see the section in the Rook v1.0 guide on updating the mon ports.

  1. TOOLS_POD=$(kubectl -n $ROOK_NAMESPACE get pod -l "app=rook-ceph-tools" -o jsonpath='{.items[0].metadata.name}')
  2. kubectl -n $ROOK_NAMESPACE exec -it $TOOLS_POD -- ceph status
  1.   cluster:
  2.     id:     b02807da-986a-40b0-ab7a-fa57582b1e4f
  3.     health: HEALTH_WARN
  4.             3 monitors have not enabled msgr2

Alternatively, this warning can suppressed if a temporary workaround is needed.

  1. TOOLS_POD=$(kubectl -n $ROOK_NAMESPACE get pod -l "app=rook-ceph-tools" -o jsonpath='{.items[0].metadata.name}')
  2. kubectl -n $ROOK_NAMESPACE exec -it $TOOLS_POD -- ceph config set global mon_warn_on_msgr2_not_enabled false

CSI Updates

If you have a v1.1 cluster running with CSI drivers enabled and you have configured the Rook-Ceph operator to use custom CSI images, the environment (env) variables need to be updated periodically. If this is the case, it is easiest to kubectl edit the operator deployment and modify everything needed at once. You can switch between using Rook-Ceph’s default CSI images and custom CSI images as you wish.

Use default images

If you would like Rook to use the inbuilt default upstream images, then you may simply remove all env variables with the ROOK_CSI_ prefix from the Rook-Ceph operator deployment.

Use custom images

OR, if you would like to use images hosted in a different location like a local image registry, then the following env variables will need to be configured in the Rook-Ceph operator deployment. The suggested upstream images are included below, which you should change to match where your images are located.

  1.   env:
  2.     - name: ROOK_CSI_CEPH_IMAGE
  3.         value: "quay.io/cephcsi/cephcsi:v1.2.2"
  4.     - name: ROOK_CSI_REGISTRAR_IMAGE
  5.         value: "quay.io/k8scsi/csi-node-driver-registrar:v1.2.0"
  6.     - name: ROOK_CSI_PROVISIONER_IMAGE
  7.         value: "quay.io/k8scsi/csi-provisioner:v1.4.0"
  8.     - name: ROOK_CSI_SNAPSHOTTER_IMAGE
  9.         value: "quay.io/k8scsi/csi-snapshotter:v1.2.2"
  10.     - name: ROOK_CSI_ATTACHER_IMAGE
  11.         value: "quay.io/k8scsi/csi-attacher:v1.2.0"

Verifying updates

You can use the below command to see the CSI images currently being used in the cluster. Once there is only a single version of each image and the version is the latest one you expect, the CSI pods are updated.

  1. # 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
  2. quay.io/cephcsi/cephcsi:v1.2.2
  3. quay.io/k8scsi/csi-attacher:v1.2.0
  4. quay.io/k8scsi/csi-node-driver-registrar:v1.2.0
  5. quay.io/k8scsi/csi-provisioner:v1.4.0
  6. quay.io/k8scsi/csi-snapshotter:v1.2.2