glance/doc/source/admin/multistores.rst

Multi Store Support

Note

The Multi Store feature was introduced as EXPERIMENTAL in Rocky and is now fully supported in the Train release.

Scope of this document

This page describes how to enable multiple stores in glance.

Prerequisites

• Glance version 17.0.0 or Later
• Glance Store Library 0.25.0 or Later
• Glance not using the Glance Registry
• Available backends

Procedure

In this section, we discuss what configuration options are available to operators to enable multiple stores support.

• in the [DEFAULT] options group:

  • enabled_backends must be set as a key:value pair where key represents the identifier for the store and value will be the type of the store. Valid values are one of file, http, rbd, swift, cinder or vmware. In order to have multiple stores operator can specify multiple key:value separated by comma.

    Warning

    The store identifier prefix os_glance_ is reserved. If you define a store identifier with this prefix, the glance service will refuse to start.

    The http store type is always treated by Glance as a read-only store. This is indicated in the response to the /v2/info/stores call, where an http type store will have the attribute read-only: True in addition to the usual id and description fields.

    [DEFAULT]
    enabled_backends = fast:rbd, cheap:rbd, shared:file, reliable:file

• in the [glance_store] options group:

  • default_backend must be set to one of the identifier which are defined using enabled_backends option. If default_backend is not set or if it is not representing one of the valid store drivers then it will prevent glance api service from starting.

    [glance_store]
    default_backend = fast

• For each of the store identifier defined in enabled_backends section operator needs to add a new config group which will define config options related to that particular store.

  [shared]
  filesystem_store_datadir = /opt/stack/data/glance/shared_images/
  store_description = "Shared filesystem store"
  
  [reliable]
  filesystem_store_datadir = /opt/stack/data/glance/reliable
  store_description = "Reliable filesystem backend"
  
  [fast]
  store_description = "Fast rbd backend"
  rbd_store_chunk_size = 8
  rbd_store_pool = images
  rbd_store_user = admin
  rbd_store_ceph_conf = /etc/ceph/ceph.conf
  rados_connect_timeout = 0
  
  [cheap]
  store_description = "Cheap rbd backend"
  rbd_store_chunk_size = 8
  rbd_store_pool = images
  rbd_store_user = admin
  rbd_store_ceph_conf = /etc/ceph/ceph1.conf
  rados_connect_timeout = 0

Backend Defaults

Glance offers a way to minimize the configuration when configuring a multi-store deployment with the help of [backend_defaults] section. We can add all the common configuration options in the [backend_defaults] group which can reduce the configuration effort and make the backend sections more readable.

Suppose we have two ceph clusters configured as two different backends in glance.

[fast]
store_description = "Fast rbd backend"
rbd_store_chunk_size = 8
rbd_store_pool = images
rbd_store_user = admin
rbd_store_ceph_conf = /etc/ceph/ceph.conf
rados_connect_timeout = 0

[cheap]
store_description = "Cheap rbd backend"
rbd_store_chunk_size = 8
rbd_store_pool = images
rbd_store_user = admin
rbd_store_ceph_conf = /etc/ceph/ceph1.conf
rados_connect_timeout = 0

We can notice that most of the configuration options are repeated in both the backend groups and hence are redundant. By using the [backend_defaults] section, we can minimize the configuration file like following:

[backend_defaults]
rbd_store_chunk_size = 8
rbd_store_pool = images
rbd_store_user = admin
rados_connect_timeout = 0

[fast]
store_description = "Fast rbd backend"
rbd_store_ceph_conf = /etc/ceph/ceph.conf

[cheap]
store_description = "Cheap rbd backend"
rbd_store_ceph_conf = /etc/ceph/ceph1.conf

We can notice that apart from making the configuration file more compact, the [backend_defaults] section also makes backend groups look cleaner to make the operator focus on just the relevant information for that particular backend.

Store Configuration Issues

Please keep the following points in mind.

• Due to the special read only nature and characteristics of the http store type, configuring multiple instances of the http type store is not supported. (This constraint is not currently enforced in the code.)
• Each instance of the filesystem store must have a different value for the filesystem_store_datadir. (This constraint is not currently enforced in the code.)

Reserved Stores

With the Train release, Glance is beginning a transition from its former reliance upon local directories for temporary data storage to the ability to use backend stores accessed via the glance_store library.

In the Train release, the use of backend stores for this purpose is optional unless you are using the multi store support feature. Since you are reading this document, this situation most likely applies to you.

Note

Currently, only the filesystem store type is supported as a Glance reserved store.

The reserved stores are not intended to be exposed to end users. Thus they will not appear in the response to the store discovery call, GET /v2/info/stores, or as values in the OpenStack-image-store-ids response header of the image-create call.

You do not get to select the name of a reserved store; these are defined by Glance and begin with the prefix os_glance_. In the Train release, you do not get to select the store type: all reserved stores must be of type filesystem.

Currently, there are two reserved stores:

os_glance_tasks_store

This store is used for the tasks engine. It replaces the use of the DEPRECATED configuration option [task]/work_dir.

os_glance_staging_store

This store is used for the staging area for the interoperable image import process. It replaces the use of the DEPRECATED configuration option [DEFAULT]/node_staging_uri.

Note

If end user wants to retrieve all the available stores using CONF.enabled_backends then they need to remove reserved stores from that list explicitly.

Configuration

As mentioned above, you do not get to select the name or the type of a reserved store (though we anticipate that you will be able configure the store type in a future release).

The reserved stores must be of type filesystem. Hence, you must provide configuration for them in your glance-api.conf file. You do this by introducing a section in glance-api.conf for each reserved store as follows:

[os_glance_tasks_store]
filesystem_store_datadir = /var/lib/glance/tasks_work_dir

[os_glance_staging_store]
filesystem_store_datadir = /var/lib/glance/staging

Since these are both filesystem stores (remember, you do not get a choice) the only option you must configure for each is the filesystem_store_datadir. Please keep the following points in mind:

• The path for filesystem_store_datadir used for the reserved stores must be different from the path you are using for any filesystem store you have listed in enabled_backends. Using the same data directory for multiple filesystem stores is unsupported and may lead to data loss.
• The identifiers for reserved stores, that is, os_glance_tasks_store and os_glance_staging_store, must not be included in the enabled_backends list.
• The reserved stores will not appear in the store discovery response or as values in the OpenStack-image-store-ids response header of the image-create call.
• The reserved stores will not be accepted as the value of the X-Image-Meta-Store header on the image-data-upload call or the image-import call.