Chapter 1. Introduction to Persistent Storage in OpenStack
OpenStack recognizes two types of storage: ephemeral and persistent. Ephemeral storage is storage that is associated only to a specific Compute instance. Once that instance is terminated, so is its ephemeral storage. This type of storage is useful for basic runtime requirements, such as storing the instance’s operating system.
Persistent storage, on the other hand, is designed to survive ("persist") independent of any running instance. This storage is used for any data that needs to be reused, either by different instances or beyond the life of a specific instance. OpenStack uses the following types of persistent storage:
The OpenStack Block Storage service (openstack-cinder) allows users to access block storage devices through volumes. Users can attach volumes to instances in order to augment their ephemeral storage with general-purpose persistent storage. Volumes can be detached and re-attached to instances at will, and can only be accessed through the instance they are attached to.
Volumes also provide inherent redundancy and disaster recovery through backups and snapshots. In addition, you can also encrypt volumes for added security. For more information about volumes, see Chapter 2, Block Storage and Volumes.Note
Instances can also be configured to use absolutely no ephemeral storage. In such cases, the Block Storage service can write images to a volume; in turn, the volume can be used as a bootable root volume for an instance.
The OpenStack Object Storage service (openstack-swift) provides a fully-distributed storage solution used to store any kind of static data or binary object, such as media files, large datasets, and disk images. The Object Storage service organizes these objects through containers.
While a volume’s contents can only be accessed through instances, the objects inside a container can be accessed through the Object Storage REST API. As such, the Object Storage service can be used as a repository by nearly every service within the cloud. For example, the Data Processing service (openstack-sahara) can manage all of its binaries, data input, data output, and templates directly through the Object Storage service.
- The Shared File System Service (openstack-manila) provides the means to easily provision remote, shareable file systems, or shares. Shares allow tenants within the cloud to openly share storage, and can be consumed by multiple instances simultaneously.
Each storage type is designed to address specific storage requirements. Containers are designed for wide access, and as such feature the highest throughput, access, and fault tolerance among all storage types. Container usage is geared more towards services.
On the other hand, volumes are used primarily for instance consumption. They do not enjoy the same level of access and performance as containers, but they do have a larger feature set and have more native security features than containers. Shares are similar to volumes in this regard, except that they can be consumed by multiple instances.
The following sections discuss each storage type’s architecture and feature set in detail, within the context of specific storage criteria.
1.1. Scalability and Back End
In general, a clustered storage solution provides greater back end scalability. For example, when using Red Hat Ceph as a Block Storage back end, you can scale storage capacity and redundancy by adding more Ceph OSD (Object Storage Daemon) nodes. Both Block Storage and Object Storage services support Red Hat Ceph as a back end.
The Block Storage service can use multiple storage solutions as discrete back ends. At the back end level, you can scale capacity by adding more back ends and restarting the service. The Block Storage service also features a large list of supported back end solutions, some of which feature additional scalability features.
By default, the Object Storage service uses the file system on configured storage nodes, and can use as much space as is available. The Object Storage service supports the XFS and ext4 file systems, and both can be scaled up to consume as much available underlying block storage. You can also scale capacity by adding more storage devices to the storage node.
The Shared File System Service provisions shares backed by storage from a separate storage pool. This pool (which is typically managed by a third-party back end service) provides the share with storage at the file system level. The Shared File System Service can use both NetApp and CephFS, which can be configured to use a storage pool of pre-created volumes which provisioned shares can use for storage. In either deployment, scaling involves adding more volumes to the pool.
1.2. Accessibility and Administration
Volumes are consumed only through instances, and can only be attached to and mounted within one instance at a time. Users can create snapshots of volumes, which can be used for cloning or restoring a volume to a previous state (see Section 1.4, “Redundancy and Disaster Recovery”). The Block Storage service also allows you to create volume types, which aggregate volume settings (for example, size and back end) that can be easily invoked by users when creating new volumes. These types can be further associated with Quality-of-Service specifications, which allow you to create different storage tiers for users.
Like volumes, shares are consumed through instances. However, shares can be directly mounted within an instance, and do not need to be attached through the dashboard or CLI. Shares can also be mounted by multiple instances simultaneously. The Shared File System service also supports share snapshots and cloning; you can also create share types to aggregate settings (similar to volume types).
Objects in a container are accessible via API, and can be made accessible to instances and services within the cloud. This makes them ideal as object repositories for services; for example, the Image service (openstack-glance) can store its images in containers managed by the Object Storage service.
The Block Storage service provides basic data security through volume encryption. With this, you can configure a volume type to be encrypted through a static key; the key will then be used for encrypting all volumes created from the configured volume type. See Section 2.2.6, “Configure Volume Encryption” for more details.
Object and container security, on the other hand, is configured at the service and node level. The Object Storage service provides no native encryption for containers and objects. Rather, the Object Storage service prioritizes accessibility within the cloud, and as such relies solely on the cloud’s network security in order to protect object data.
The Shared File System service can secure shares through access restriction, whether by instance IP, user/group, or TLS certificate. In addition, some Shared File System service deployments can feature a separate share servers to manage the relationship between share networks and shares; some share servers support (or even require) additional network security. For example, a CIFS share server requires the deployment of an LDAP, Active Directory, or Kerberos authentication service.
1.4. Redundancy and Disaster Recovery
The Block Storage service features volume backup and restoration, providing basic disaster recovery for user storage. Backups allow you to protect volume contents. On top of this, the service also supports snapshots; aside from cloning, snapshots are also useful in restoring a volume to a previous state.
In a multi-backend environment, you can also migrate volumes between back ends. This is useful if you need to take a back end offline for maintenance. Backups are typically stored in a storage back end separate from their source volumes to help protect the data. This is not possible, however, with snapshots, as snapshots are dependent on their source volumes.
The Block Storage service also supports the creation of consistency groups, which allow you to group volumes together for simultaneous snapshot creation. This, in turn, allows for a greater level of data consistency across multiple volumes. See Section 2.2.9, “Configure and Use Consistency Groups” for more details.
The Object Storage service provides no built-in backup features. As such, all backups must be performed at the file system or node level. The service, however, features more robust redundancy and fault tolerance; even the most basic deployment of the Object Storage service replicates objects multiple times. You can use failover features like dm-multipath to enhance redundancy.
The Shared File System service provides no built-in backup features for shares, but it does allow you to create snapshots for cloning and restoration.