Image-Volume cache

OpenStack Block Storage has an optional Image cache which can dramatically improve the performance of creating a volume from an image. The improvement depends on many factors, primarily how quickly the configured back end can clone a volume.

When a volume is first created from an image, a new cached image-volume will be created that is owned by the Block Storage Internal Tenant. Subsequent requests to create volumes from that image will clone the cached version instead of downloading the image contents and copying data to the volume.

The cache itself is configurable per back end and will contain the most recently used images.

Configure the Internal Tenant

The Image-Volume cache requires that the Internal Tenant be configured for the Block Storage services. This project will own the cached image-volumes so they can be managed like normal users including tools like volume quotas. This protects normal users from having to see the cached image-volumes, but does not make them globally hidden.

To enable the Block Storage services to have access to an Internal Tenant, set the following options in the cinder.conf file:

cinder_internal_tenant_project_id = PROJECT_ID
cinder_internal_tenant_user_id = USER_ID

An example cinder.conf configuration file:

cinder_internal_tenant_project_id = b7455b8974bb4064ad247c8f375eae6c
cinder_internal_tenant_user_id = f46924c112a14c80ab0a24a613d95eef

Note

The actual user and project that are configured for the Internal Tenant do not require any special privileges. They can be the Block Storage service project or can be any normal project and user.

Configure the Image-Volume cache

To enable the Image-Volume cache, set the following configuration option in the cinder.conf file:

image_volume_cache_enabled = True

Note

If you use Ceph as a back end, set the following configuration option in the cinder.conf file:

[ceph]
image_volume_cache_enabled = True

This can be scoped per back end definition or in the default options.

There are optional configuration settings that can limit the size of the cache. These can also be scoped per back end or in the default options in the cinder.conf file:

image_volume_cache_max_size_gb = SIZE_GB
image_volume_cache_max_count = MAX_COUNT

By default they will be set to 0, which means unlimited.

For example, a configuration which would limit the max size to 200 GB and 50 cache entries will be configured as:

image_volume_cache_max_size_gb = 200
image_volume_cache_max_count = 50

Note

As mentioned above, the internal tenant configured as the cache owner does not require any special permissions and is subject to quotas like any other user. Hence, it is possible that the quotas for the internal tenant may need to be adjusted to allow the internal tenant to hold at least image_volume_cache_max_count volumes not exceeding image_volume_cache_max_size_gb total size. Thus, although the default value for these image volume cache settings is 0 (unlimited), in practice, these will be limited by the quotas that apply to the internal tenant.

See Manage Block Storage service quotas for more information.

Notifications

Cache actions will trigger Telemetry messages. There are several that will be sent.

  • image_volume_cache.miss - A volume is being created from an image which was not found in the cache. Typically this will mean a new cache entry would be created for it.

  • image_volume_cache.hit - A volume is being created from an image which was found in the cache and the fast path can be taken.

  • image_volume_cache.evict - A cached image-volume has been deleted from the cache.

Managing cached Image-Volumes

In normal usage there should be no need for manual intervention with the cache. The entries and their backing Image-Volumes are managed automatically.

If needed, you can delete these volumes manually to clear the cache. By using the standard volume deletion APIs, the Block Storage service will clean up correctly.