Storage configuration¶

Atmosphere uses a unified storage configuration variable atmosphere_storage that serves as the single source of truth for all storage backends across OpenStack services. This eliminates the need to configure storage settings in multiple places.

Overview¶

The atmosphere_storage variable organizes storage by purpose:

images: Image storage for Glance. Supports multiple backends.
volumes: Block storage for Cinder. Supports multiple backends.
backup: Backup storage for Cinder backup.
ephemeral: Ephemeral disk storage for Nova and libvirt.

The system automatically derives per-service Helm values from this single variable for Cinder, Glance, Nova, libvirt, and Ceph Provisioners.

Default configuration¶

By default, Atmosphere configures image and volume storage to use the integrated Ceph cluster, with local (qcow2) ephemeral disks for Nova. The atmosphere_storage variable defaults to an empty dictionary ({}), and the built-in defaults provide sensible starting values.

You can override any part of the configuration by setting atmosphere_storage in your inventory. Your overrides merge on top of the built-in defaults using a recursive merge, so you only need to specify the parts you want to change.

The built-in defaults are equivalent to:

images:
  default: rbd1
  backends:
    rbd1:
      type: rbd
      pool: glance.images
      replication: 3
      crush_rule: replicated_rule
      chunk_size: 8
      user: glance
volumes:
  default: rbd1
  backends:
    rbd1:
      type: rbd
      pool: cinder.volumes
      replication: 3
      crush_rule: replicated_rule
      user: cinder
      secret_uuid: 457eb676-33da-42ec-9a8c-9293d545c337
backup:
  type: rbd
  pool: cinder.backups
  replication: 3
  crush_rule: replicated_rule
  user: cinderbackup
ephemeral:
  type: local

The integrated Ceph cluster works without additional configuration.

Using non-Ceph storage¶

The built-in defaults include Ceph-backed rbd1 entries for both images.backends and volumes.backends. Because atmosphere_storage overrides merge recursively, defining a new backend doesn’t remove those defaults.

To replace the default Ceph backends with a non-Ceph layout, explicitly set the inherited rbd1 entries to null. This keeps the storage configuration on the non-Ceph path instead of running with both the default Ceph backend and the new backend.

For a fully non-Ceph deployment:

Set images.default and volumes.default to your non-Ceph backends.
Set images.backends.rbd1: null and volumes.backends.rbd1: null.
Set backup.type: none unless you want to keep Cinder backup on Ceph RBD.
Keep ephemeral.type: local unless you want Nova ephemeral disks on Ceph RBD.

atmosphere_storage:
  images:
    default: cinder_store
    backends:
      rbd1: null
      cinder_store:
        type: cinder
  volumes:
    default: powerstore
    backends:
      rbd1: null
      powerstore:
        type: powerstore
        address: 10.0.0.1
        username: admin
        password: secret
        protocol: iscsi
  backup:
    type: none
  ephemeral:
    type: local

Enabling RBD ephemeral storage¶

To use Ceph RBD for Nova ephemeral disks instead of local storage, set the ephemeral section in your atmosphere_storage override:

atmosphere_storage:
  ephemeral:
    type: rbd
    pool: vms
    replication: 3
    crush_rule: replicated_rule
    user: cinder
    secret_uuid: 457eb676-33da-42ec-9a8c-9293d545c337

Supported backend types¶

Ceph RBD (`rbd`)¶

Standard replicated Ceph RBD backend. Used for images, volumes, backups, and ephemeral storage.

type: rbd
pool: <pool_name>
replication: 3
crush_rule: replicated_rule
user: <username>
secret_uuid: <uuid>       # required for volumes and ephemeral

Ceph RBD erasure-coded (`rbd-ec`)¶

Erasure-coded Ceph RBD backend for improved storage efficiency. Only supported for volume backends. EC pools provide better storage utilization compared to replicated pools, at the cost of slightly higher CPU overhead.

EC pools require two components:

A metadata pool (replicated) that stores metadata
A data pool (EC) that stores the actual volume data

type: rbd-ec
pool: cinder.volumes.ec
erasure_coded:
  k: 4                     # data chunks
  m: 2                     # coding (parity) chunks
  failure_domain: host
  device_class: hdd        # optional
metadata_replication: 3
crush_rule: replicated_rule  # optional: CRUSH rule for metadata pool
user: cinder-ec       # must be unique per EC backend
secret_uuid: <unique-uuid>

The EC pool configuration parameters are:

k: Number of data chunks. Higher values provide better storage efficiency.
m: Number of coding (parity) chunks. Higher values provide better fault tolerance.
device_class: Target specific device types, for example hdd, ssd, or nvme.
failure_domain: Failure domain for data placement, for example host, rack, or room.
crush_rule: CRUSH rule for the replicated metadata pool. Defaults to replicated_rule.

Warning

Each erasure-coded backend requires a dedicated user and a unique secret_uuid. Generate new UUID’s with uuidgen.

Dell PowerStore (`powerstore`)¶

Warning

Make sure that you set up the hosts inside of your storage array. They must not be inside a host group, or individual attachments won’t work.

type: powerstore
address: <management_ip_or_hostname>
username: <username>
password: <password>
protocol: iscsi    # or fc

Pure storage (`pure`)¶

Pure Storage FlashArray integration. When using the NVMe protocol, you can optionally set transport to roce or tcp.

For additional options, see the Cinder Pure Storage documentation.

type: pure
protocol: iscsi    # or fc, nvme
address: <management_ip_or_hostname>
api_token: <token>
transport: roce    # optional, only for nvme protocol

StorPool (`storpool`)¶

StorPool distributed storage backend. The system configures network settings and file system mounts automatically.

type: storpool
template: hybrid-2ssd

HPE Nimble storage (`nimble`)¶

HPE Nimble storage (including Alletra 5000/6000) integration with support for iSCSI and Fibre Channel protocols.

For additional options, see the Cinder HPE Nimble storage documentation.

type: nimble
protocol: iscsi    # or fc
address: <management_ip_or_hostname>
username: <username>
password: <password>
nimble_subnet_label: "*"  # optional

The nimble_subnet_label option defaults to "*".

Cinder (`cinder`)¶

Cinder-backed Glance store. Only used for image backends.

type: cinder

Backup configuration¶

The backup section controls Cinder backup storage. Atmosphere supports these values:

type: rbd: Store Cinder backups in Ceph. This is the built-in default.
type: none: Disable the Cinder backup service and its storage-init job.

For non-Ceph volume backends such as PowerStore, Pure Storage, StorPool, or Nimble, set backup.type: none unless you still plan to use Ceph for backups.

If you keep Ceph backups while using a non-Ceph volume backend, define the backup block explicitly:

atmosphere_storage:
  backup:
    type: rbd
    pool: cinder.backups
    replication: 3
    crush_rule: replicated_rule
    user: cinderbackup

Examples¶

Adding an erasure-coded volume backend¶

To add an EC pool alongside the default replicated backend, add a new entry under atmosphere_storage.volumes.backends:

atmosphere_storage:
  volumes:
    backends:
      rbd_ec:
        type: rbd-ec
        pool: cinder.volumes.ec
        erasure_coded:
          k: 4
          m: 2
          failure_domain: host
        metadata_replication: 3
        user: cinder-ec
        secret_uuid: 808c5658-7c46-4818-8f26-82a217e3a57a

The system automatically configures all required services: Cinder backend configuration, Ceph pool creation, libvirt secret registration, and Ceph client data pool routing.

Using Dell PowerStore¶

atmosphere_storage:
  images:
    default: cinder_store
    backends:
      rbd1: null
      cinder_store:
        type: cinder
  volumes:
    default: powerstore
    backends:
      rbd1: null
      powerstore:
        type: powerstore
        address: 10.0.0.1
        username: admin
        password: secret
        protocol: iscsi
  backup:
    type: none
  ephemeral:
    type: local

Using HPE Nimble storage¶

atmosphere_storage:
  images:
    default: cinder_store
    backends:
      rbd1: null
      cinder_store:
        type: cinder
  volumes:
    default: nimble1
    backends:
      rbd1: null
      nimble1:
        type: nimble
        address: 10.0.0.1
        username: admin
        password: secret
        protocol: iscsi
        nimble_subnet_label: "*"
  backup:
    type: none
  ephemeral:
    type: local

Glance with multiple backends¶

atmosphere_storage:
  images:
    default: rbd1
    backends:
      rbd1:
        type: rbd
        pool: glance.images
        replication: 3
        user: glance
      cinder_store:
        type: cinder

Advanced overrides¶

The atmosphere_storage variable generates base Helm values. You can still use per-service *_helm_values variables (such as cinder_helm_values or glance_helm_values) to override specific settings. The per-service overrides always take precedence.

Migrating from `atmosphere_ceph_enabled`¶

The atmosphere_ceph_enabled variable is no longer supported. Storage configuration now uses the unified atmosphere_storage variable. If your inventory sets atmosphere_ceph_enabled, remove it and configure atmosphere_storage instead.

If you previously set `atmosphere_ceph_enabled: true`¶

No action beyond removing the variable is necessary. The built-in defaults already configure Ceph RBD for images and volumes, which matches the previous behavior.

# Remove this from your inventory:
# atmosphere_ceph_enabled: true

If you previously set `atmosphere_ceph_enabled: false`¶

Remove the variable and configure atmosphere_storage to match your environment. For example, if you use Dell PowerStore for volumes and local ephemeral storage, make sure that you also null out the inherited rbd1 backends:

# Remove this from your inventory:
# atmosphere_ceph_enabled: false

# Add this instead:
atmosphere_storage:
  images:
    default: cinder_store
    backends:
      cinder_store:
        type: cinder
  volumes:
    default: powerstore
    backends:
      powerstore:
        type: powerstore
        address: 10.0.0.1
        username: admin
        password: secret
        protocol: iscsi
  backup:
    type: none
  ephemeral:
    type: local

Storage configuration¶

Overview¶

Default configuration¶

Using non-Ceph storage¶

Enabling RBD ephemeral storage¶

Supported backend types¶

Ceph RBD (rbd)¶

Ceph RBD erasure-coded (rbd-ec)¶

Dell PowerStore (powerstore)¶

Pure storage (pure)¶

StorPool (storpool)¶

HPE Nimble storage (nimble)¶

Cinder (cinder)¶

Backup configuration¶

Examples¶

Adding an erasure-coded volume backend¶

Using Dell PowerStore¶

Using HPE Nimble storage¶

Glance with multiple backends¶

Advanced overrides¶

Migrating from atmosphere_ceph_enabled¶

If you previously set atmosphere_ceph_enabled: true¶

If you previously set atmosphere_ceph_enabled: false¶

Ceph RBD (`rbd`)¶

Ceph RBD erasure-coded (`rbd-ec`)¶

Dell PowerStore (`powerstore`)¶

Pure storage (`pure`)¶

StorPool (`storpool`)¶

HPE Nimble storage (`nimble`)¶

Cinder (`cinder`)¶

Migrating from `atmosphere_ceph_enabled`¶

If you previously set `atmosphere_ceph_enabled: true`¶

If you previously set `atmosphere_ceph_enabled: false`¶