v0.6.0

Major Themes

Rook v0.6 is a release focused on making progress towards our goal of running Rook everywhere Kubernetes runs. There is a new Rook volume plugin and Rook agent that integrates into Kubernetes to provide on demand block and shared filesystem storage for pods with a streamlined and seamless experience. We are keeping the Alpha status for v0.6 due to these two new components. The next release (v0.7) will be our first beta quality release.

Rook has continued its effort for deeper integration with Kubernetes by defining Custom Resource Definitions (CRDs) for both shared filesystems and object storage as well, allowing management of all storage types natively via kubectl.

Reliability has been further improved with a focus on self-healing functionality to recover the cluster to a healthy state when key components have been detected as unhealthy. Investment in the automated test pipelines has been made to increase automated scenario and environment coverage across multiple versions of Kubernetes and cloud providers.

Finally, the groundwork has been laid for automated software upgrades in future releases by both completing the design and publishing a manual upgrade user guide in this release.

Notable Features

• Rook Flexvolume (@kokhang)
  • Introduces a new Rook plugin based on Flexvolume
  • Integrates with Kubernetes' Volume Controller framework
  • Handles all volume attachment requests such as attaching, mounting and formatting volumes on behalf of Kubernetes
  • Simplifies the deployment by not requiring to have ceph-tools package installed
  • Allows block devices and filesystems to be consumed without any user secret management
  • Improves experience with fencing and volume locking
• Rook-Agents (@kokhang and @jbw976)
  • Configured and deployed via Daemonset by Rook-operator
  • Installs the Rook Flexvolume plugin
  • Handles all storage operations required on the node, such as attaching devices, mounting volumes and formatting filesystem.
  • Performs node cleanups during Rook cluster teardown
• File system (@travisn)
  • File systems are defined by a CRD and handled by the Operator
  • Multiple file systems can be created, although still experimental in Ceph
  • Multiple data pools can be created
  • Multiple MDS instances can be created per file system
  • An MDS is started in standby mode for each active MDS
  • Shared filesystems are now supported by the Rook Flexvolume plugin
    • Improved and streamlined experience, now there are no manual steps to copy monitor information or secrets
    • Multiple shared filesystems can be created and consumed within the cluster
    • More information can be found in the shared filesystems user guides
• Object Store (@travisn)
  • Object Stores are defined by a CRD and handled by the Operator
  • Multiple object stores supported through Ceph realms
• OSDs (@jbw976, @paha, and @phsiao)
  • Bluestore is now the default backend store for OSDs when creating a new Rook cluster.
  • Bluestore can now be used on directories in addition to raw block devices that were already supported.
  • If an OSD loses its metadata and config but still has its data devices, the OSD will automatically regenerate the lost metadata to make the data available again.
  • If an OSD is observed to be in the down status for an extended period of time then it will automatically be restarted.
  • OSDs can now run as lower privileged when devices are not being used.
• Pools (@travisn)
  • The failure domain for the CRUSH map can be specified on pools with the failureDomain property
  • Pools created by file systems or object stores are configurable with all options defined in the pool CRD
• Upgrade (@jbw976)
  • A manual upgrade user guide has been published to help guide users through the recommended upgrade process.
• API (@galexrt)
  • The API pod will continuously refresh its list of monitor endpoints in order to always be able to service requests even after monitor topology has changed
• Test framework (@dangula)
  • Test pipelines cover Kubernetes 1.6-1.8 and cloud providers AWS and GCE
  • Long haul testing pipeline will run for multiple days to simulate longer term workloads
  • Raw block devices in addition to local filesystems (directories) are now tested
  • Helm chart installation is now tested
• HostNetwork (@galexrt)
  • It is now possible to launch a Rook cluster with host networking support. See further details in the Cluster CRD documentation.

Breaking Changes

• Rook Standalone
  • Standalone mode has been disabled and is no longer supported. A Kubernetes environment is required to run Rook.
• Rook-operator now deploys in rook-system namespace
  • If using the example manifest of rook-operator.yaml, the rook-operator deployment is now changed to deploy in the rook-system namespace.
• Rook Flexvolume
  • Persistent volumes from previous releases were created using the RBD plugin. These should be deleted and recreated in order to use the new Flexvolume plugin.
• Pool CRD
  • replication renamed to replicated
  • erasureCode renamed to erasureCoded
• OSDs
  • OSD pods now require RBAC permissions to create/get/update/delete/list config maps.
    An upgraded operator will create the necessary service account, role, and role bindings to enable this.
• API
  • The API pod now uses RBAC permissions that are scoped only to the namespace it is running in.
    An upgraded operator will create the necessary service account, role, and role bindings to enable this.
• Filesystem
  • The Rook Flexvolume uses the mds_namespace option to specify a cephFS. This is only available on Kernel v4.7 or newer. On older kernel, if there are more than one filesystems in the cluster, the mount operation could be inconsistent. See this doc.

Known Issues

• Rook Flexvolume
  • For Kubernetes cluster 1.7.x and older, Kubelet will need to be restarted in order to load the new flexvolume. This has been resolved in K8S 1.8. For more information about the requirements, refer to this doc
  • For Kubernetes cluster 1.6.x, the attacher/detacher controller needs to be disabled in order to load the new Flexvolume. This is caused by a regression on 1.6.x. For more information about the requirements, refer to this doc
  • For CoreOS and Rancher Kubernetes, the Flexvolume plugin dir will need to be specified to be different than the default. Refer to Flexvolume configuration pre-reqs
• OSD pods will not get to the running state for Kubernetes 1.6.4 or lower, due to a regression in Kubernetes that is not compatible with security context settings being applied to the OSD pods. The workaround is to upgrade Kubernetes to 1.6.5 or higher. More details can be found in the tracking issue.

Deprecations

• Rook Standalone
  • As mentioned in the breaking changes section, Standalone mode has been removed.
• Rook API
  • Rook has a goal to integrate natively and deeply with container orchestrators such as Kubernetes and using extension points to manage and access the Rook storage cluster. More information can be found in the tracking issue.
• Rookctl
  • The rookctl client is being deprecated. With the deeper and more native integration of Rook with Kubernetes, kubectl now provides a rich Rook management experience on its own. For direct management of the Ceph storage cluster, the Rook toolbox provides full access to the Ceph tools.