Per-Tenant Quotas

Glance supports resource consumption quotas on tenants through the use of Keystone’s unified limits functionality. Resource limits are registered in Keystone with suitable default values, and may be overridden on a per-tenant basis. When a resource consumption attempt is made in Glance, the current consumption is computed and compared against the limit set in Keystone; the request is denied if the user is over the specified limit.

Due to the design of Glance, most of the storage-focused quotas in Glance are soft limits. Since Glance allows clients to stream image data of unknown total size during an upload or import operation, it is not possible to determine if quota has been exceeded until after the operation has completed. Thus, a user is permitted to go over their quota for a single operation, and then denied additional stored on subsequent operations. There are object-focused quotas that can help operators limit the damage caused by multiple large competing data streams. Those details are covered below.

Note

Glance also has legacy global resource limits that may be ignored if per-tenant quotas are enabled. Currently the user_storage_quota limit will be ignored if per-tenant quotas are used.

See the Keystone docs for more information on unified limits.

Quota Resource Types

Glance supports quota limits on multiple areas of resource consumption. Limits are enforced at the time in which resource consumption is attempted, so setting an existing user’s quota for any item below the current usage will only prevent them from consuming more data until they free up space.

Total Image Size

The image_size_total limit defines the maximum amount of storage (in MiB) that the tenant may consume across all of their active images. Images with multiple locations contribute to this count according to the number of places the image is stored. Thus, if you have a single 1GiB image stored in four locations, the usage will be considered to be 4GiB.

Total Staging Size

The Interoperable Image Import function uses a two-step upload process, whereby a user first uploads an image into the staging store, and then subsequently imports the image to the final destination(s). The staging store is generally local storage on the API workers themselves, and thus is likely at somewhat of a premium, compared to the bulk shared storage allocated for general images. The image_stage_total limit defines the total amount of staging space that may be used. This should be set to a value sufficient to allow a user to import one or more images at the same time, according to your desired level of parallelism. It may be appropriate to provide the user with a very generous image_size_total quota, but a relatively restrictive image_stage_total allocation, effectively limiting them to one image being imported at any given point.

Keep in mind that images being imported using the web-download method will need to fit within this allocation as well, as those are first downloaded to the staging store before being imported to the final destination(s). Images being copied from one store to another using the copy-image method are similarly affected. Note that the conventional image upload method does not stage the image, and thus is not impacted by this limit.

Total Number of Images

The image_count_total limit controls the maximum number of image objects that the user may have, regardless of the individual or collective sizes or impact to storage. This limit may be useful if you wish to prevent users from taking thousands of small server snapshots without ever deleting them.

Total Number of In-Progress Uploads

Because Glance can not enforce storage-focused quotas until after a stream is finished, it may be useful to limit the number of parallel upload operations that can be in-progress at any single point. The image_count_uploading limit provides this control, and affects conventional image upload, pre-import stage (including web-download and glance-direct), as well as any copy-image operations that may be pending. It may be desirable to limit untrusted users to a single in-progress image upload, which will limit the amount of damage a malicious user may be able to inflict on your image storage if they initiate multiple simultaneous unbounded upload streams.

Quota Strategies

Below are a couple of use-case example strategies for different types of deployments. In all cases, it makes sense for image_size_total and image_stage_total to be set to at least the size of the largest image you expect a user to use. The global limit on a single image (see configuration item image_size_cap) may be relevant as well. Users with an image_count_total of zero will be unable to create any images, and with an image_count_uploading of zero will be able to upload data to any images.

  1. Public cloud, users are billed per-byte: In this case, it probably makes sense to set fairly high default quota limits for each of the above resource classes, allowing users to consume as much as they are willing to pay for. It still may be desirable to set image_stage_total to something modest to prevent overrunning limited staging space, if you have import enabled.

  2. Private cloud, trusted users are billed by quota: In this case, each user pays for the amount of resource they are allowed to consume, instead of what they are consuming. Generally this involves billing total space, so image_size_total is set to their allotment, potentially with some upper bound on total images via image_count_total. If they are somewhat trusted or low-impact customers, limiting the staging usage and upload count is probably not necessary, and can be left unbounded or set to some high upper bound.

  3. Private cloud, semi-trusted third party users: This case may be similar to either of the above in terms of paying for allotment or strict usage. However, the lack of full trust may suggest limiting the total number of image uploads to something like 10% of their compute quota (to allow for snapshots) and limiting staging usage to enough for one or two image imports at a time.

Configuring Glance for Per-Tenant Quotas

  1. Register quota limits (optional):

    If you decide to use per-tenant quotas in Glance, you must register the limits in Keystone first:

    $ openstack --os-cloud devstack-system-admin registered limit create \
      --service glance --default-limit 1000 --region RegionOne image_size_total
    
    +---------------+----------------------------------+
    | Field         | Value                            |
    +---------------+----------------------------------+
    | default_limit | 1000                             |
    | description   | None                             |
    | id            | 9cedfc5de80345a9b13ed00c2b5460f2 |
    | region_id     | RegionOne                        |
    | resource_name | image_size_total                 |
    | service_id    | e38c84a2487f49fd9864193bdc8a3174 |
    +---------------+----------------------------------+
    
    $ openstack --os-cloud devstack-system-admin registered limit create \
      --service glance --default-limit 1000 --region RegionOne image_stage_total
    
    +---------------+----------------------------------+
    | Field         | Value                            |
    +---------------+----------------------------------+
    | default_limit | 1000                             |
    | description   | None                             |
    | id            | 5a68712b6ba6496d823d0c66e5e860b9 |
    | region_id     | RegionOne                        |
    | resource_name | image_stage_total                |
    | service_id    | e38c84a2487f49fd9864193bdc8a3174 |
    +---------------+----------------------------------+
    
    $ openstack --os-cloud devstack-system-admin registered limit create \
      --service glance --default-limit 100 --region RegionOne image_count_total
    
    +---------------+----------------------------------+
    | Field         | Value                            |
    +---------------+----------------------------------+
    | default_limit | 100                              |
    | description   | None                             |
    | id            | beb91b043296499f8e6268f29d8b2749 |
    | region_id     | RegionOne                        |
    | resource_name | image_count_total                |
    | service_id    | e38c84a2487f49fd9864193bdc8a3174 |
    +---------------+----------------------------------+
    
    $ openstack --os-cloud devstack-system-admin registered limit create \
      --service glance --default-limit 100 --region RegionOne image_count_uploading
    
    +---------------+----------------------------------+
    | Field         | Value                            |
    +---------------+----------------------------------+
    | default_limit | 100                              |
    | description   | None                             |
    | id            | fc29649c047a45bf9bc03ec4a7bcb8af |
    | region_id     | RegionOne                        |
    | resource_name | image_count_uploading            |
    | service_id    | e38c84a2487f49fd9864193bdc8a3174 |
    +---------------+----------------------------------+
    

    Be sure to also set use_keystone_limits=True in your glance-api.conf file.

  2. Tell Glance to use Keystone quotas

  • In the [oslo_limit] section, configure access to keystone:

    [oslo_limit]
    auth_url = http://controller:5000
    auth_type = password
    user_domain_id = default
    username = MY_SERVICE
    system_scope = all
    password = MY_PASSWORD
    endpoint_id = ENDPOINT_ID
    region_name = RegionOne
    

    Make sure that the MY_SERVICE account has reader access to system-scope resources (like limits):

    $ openstack role add --user MY_SERVICE --user-domain Default --system all reader
    

    See the oslo_limit docs for more information about configuring the unified limits client.

  • In the [DEFAULT] section, optionally enable per-tenant quotas:

    [DEFAULT]
    use_keystone_limits = True
    

    Note that you must have created the registered limits as described above if this is enabled.