Skip to content

Latest commit

 

History

History
228 lines (167 loc) · 10 KB

UPGRADE.md

File metadata and controls

228 lines (167 loc) · 10 KB
Raw

General notes

During the upgrade process, provisioning of volumes and attach/detach operations might not work. Existing volumes and volumes already in use by a pod will continue to work without interruption.

Upgrades from v1 to v2

There currently is no documented upgrade path from Operator v1 to v2.

Upgrade from v1.9 to v1.10

If you want to share DRBD configuration directories with the host, please update the CRDs before upgrading using helm:

$ kubectl replace -f ./charts/piraeus/crds

Upgrade from v1.8 to v1.9

If you want to protect metrics endpoints, take a look on guide on enabling rbac-proxy sidecars.

We've also disabled the haController component in our chart. The replacement is available from artifacthub.io, containing much needed improvements in robustness and fail-over speed. If you want to still use the old version, set haController.enabled=true.

Upgrade from v1.7 to v1.8

If you need to set the number of worker threads, or you need to set the log level of LINSTOR components, please update the CRDs before upgrading using helm:

$ kubectl replace -f ./charts/piraeus/crds

In case if you have SSL enabled installation, you need to regenerate your certificates in PEM format. Read the guide on securing the deployment and repeat the described steps.

Upgrade from v1.6 to v1.7

Node labels are now automatically applied to LINSTOR satellites as "Auxiliary Properties". That means you can reuse your existing Kubernetes Topology information (for example topology.kubernetes.io/zone labels) when scheduling volumes using the replicasOnSame and replicasOnDifferent settings. However, this also means that the Operator will delete any existing auxiliary properties that were already applied. To apply auxiliary properties to satellites, you have to apply a label to the Kubernetes node object.

Upgrade from v1.5 to v1.6

The csi-snapshotter subchart was removed from this repository. Users who relied on it for their snapshot support should switch to the seperate charts provided by the Piraeus team on artifacthub.io The new charts also include notes on how to upgrade to newer CRD versions.

To support the new csi.kubeletPath option an update to the LinstorCSIDriver CRD is required:

$ kubectl replace -f ./charts/piraeus/crds

Upgrade from v1.4 to v1.5

To make use of the new monitoringImage value in the LinstorSatelliteSet CRD, you need to replace the existing CRDs before running the upgrade. If the CRDs are not upgraded, the operator will not deploy the monitoring container alongside the satellites.

$ kubectl replace -f ./charts/piraeus/crds

Then run the upgrade:

helm upgrade piraeus-op ./charts/piraeus -f <overrides>

Upgrade from v1.3 to v1.4

No special steps required, unless using the newly introduced additionalEnv and additionalProperties overrides.

If you plan to use the new overrides, replace the CustomResourceDefinitions before running the upgrade

$ kubectl replace -f ./charts/piraeus/crds

Then run the upgrade:

helm upgrade piraeus-op ./charts/piraeus -f <overrides>

Upgrade from v1.2 to v1.3

No special steps required. After checking out the new repo, run a simple helm upgrade:

helm upgrade piraeus-op ./charts/piraeus -f <overrides>

Upgrade from v1.1 to v1.2

Piraeus v1.2 is supported on Kubernetes 1.17+. If you are using an older Kubernetes distribution, you may need to change the default settings (for example the CSI provisioner)

To start the upgrade process, ensure you have a backup of the LINSTOR Controller database. If you are using the etcd deployment included in Piraeus, you can create a backup using:

kubectl exec piraeus-op-etcd-0 -- etcdctl snapshot save /tmp/save.db
kubectl cp piraeus-op-etcd-0:/tmp/save.db save.db

Now you can start the upgrade process. Simply run helm upgrade piraeus-op ./charts/piraeus. If you installed Piraeus with customization, pass the same options you used for helm install to helm upgrade. This will cause the operator pod to be re-created and shortly after all other Piraeus pods.

There is a known issue when updating the CSI components: the pods will not be updated to the newest image and the errors section of the LinstorCSIDrivers resource shows an error updating the DaemonSet. In this case, manually delete deployment/piraeus-op-csi-controller and daemonset/piraeus-op-csi-node. They will be re-created by the operator.

After a short wait, all pods should be running and ready. Check that no errors are listed in the status section of LinstorControllers, LinstorSatelliteSets and LinstorCSIDrivers.

Upgrade from v1.0 to v1.1

  • The LINSTOR controller image given in operator.controller.controllerImage has to have its entrypoint set to k8s-await-election v0.2.0 or newer. All images starting with piraeus-server:v1.8.0 meet this requirement.

    Older images will not work, as the Service will not automatically pick up on the active pod.

    To upgrade, first update the deployed LINSTOR image to a compatible version, then upgrade the operator.

Upgrade to v1.0

Upgrades from v0.* versions to v1.0 are best-effort only. The following guide assumes an upgrade from v0.5.0.

  • CRDs have been updated and set to version v1. You need to replace any existing CRDs with these new ones:

    kubectl replace -f charts/piraeus/crds/piraeus.linbit.com_linstorcontrollers_crd.yaml
kubectl replace -f charts/piraeus/crds/piraeus.linbit.com_linstorcsidrivers_crd.yaml
kubectl replace -f charts/piraeus/crds/piraeus.linbit.com_linstorsatellitesets_crd.yaml

  • Renamed LinstorNodeSet to LinstorSatelliteSet. This brings the operator in line with other LINSTOR resources. Existing LinstorNodeSet resources will automatically be migrated to LinstorSatelliteSet. The old resources will not be deleted. You can verify that migration was successful by running the following command:

    $ kubectl get linstornodesets.piraeus.linbit.com -o=jsonpath='{range .items[*]}{.metadata.name}{"\t"}{.status.ResourceMigrated}{"\t"}{.status.DependantsMigrated}{"\n"}{end}'
piraeus-op-ns true    true

    If both values are true, migration was successful. The old resource can be removed after migration.

  • Renamed LinstorControllerSet to LinstorController. The old name implied the existence of multiple (separate) controllers. Existing LinstorControllerSet resources will automatically be migrated to LinstorController. The old resources will not be deleted. You can verify that migration was successful by running the following command:

    $ kubectl get linstorcontrollersets.piraeus.linbit.com -o=jsonpath='{range .items[*]}{.metadata.name}{"\t"}{.status.ResourceMigrated}{"\t"}{.status.DependantsMigrated}{"\n"}{end}'
piraeus-op-cs true    true

    If both values are true, migration was successful. The old resource can be removed after migration.

  • Along with the CRDs, Helm settings changed too:

    • operator.controllerSet to operator.controller
    • operator.nodeSet to operator.satelliteSet If old settings are used, Helm will return an error.

  • Node scheduling no longer relies on linstor.linbit.com/piraeus-node labels. Instead, all CRDs support setting pod affinity and tolerations. In detail:

    • linstorcsidrivers gained 4 new resource keys, with no change in default behaviour:

      • nodeAffinity affinity passed to the csi nodes
      • nodeTolerations tolerations passed to the csi nodes
      • controllerAffinity affinity passed to the csi controller
      • controllerTolerations tolerations passed to the csi controller

    • linstorcontrollerset gained 2 new resource keys, with no change in default behaviour:

      • affinity affinity passed to the linstor controller pod
      • tolerations tolerations passed to the linstor controller pod

    • linstornodeset gained 2 new resource keys, with change in default behaviour:

      • affinity affinity passed to the linstor controller pod
      • tolerations tolerations passed to the linstor controller pod

      Previously, linstor satellite pods required nodes marked with linstor.linbit.com/piraeus-node=true. The new default value does not place this restriction on nodes. To restore the old behaviour, set the following affinity:

      affinity:
  nodeAffinity:
    requiredDuringSchedulingIgnoredDuringExecution:
      nodeSelectorTerms:
      - matchExpressions:
        - key: linstor.linbit.com/piraeus-node
          operator: In
          values:
          - "true"
  • Other changed helm settings: drbdKernelModuleInjectionMode: renamed to kernelModuleInjectionMode kernelModImage: renamed to kernelModuleInjectionImage If old settings are used, Helm will return an error.

Upgrade between v0.* versions

While the API version was v1alpha1, this project did not maintain stability of or provide conversion of the Custom Resource Definitions.

If you are using the Helm deployment, you may find that upgrades fail with errors similar to the following:

UPGRADE FAILED: cannot patch "piraeus-op-cs" with kind LinstorController: LinstorController.piraeus.linbit.com "piraeus-op-cs" is invalid: spec.etcdURL: Required value

The simplest solution in this case is to manually replace the CRD:

kubectl replace -f charts/piraeus/crds/piraeus.linbit.com_linstorcsidrivers_crd.yaml
kubectl replace -f charts/piraeus/crds/piraeus.linbit.com_linstorsatellitesets_crd.yaml
kubectl replace -f charts/piraeus/crds/piraeus.linbit.com_linstorcontrollers_crd.yaml

Then continue with the Helm upgrade. Values that are lost during the replacement will be set again by Helm.