Provide the means to store the configuration options that would normally stored in domain-specific configuration files in an SQL table instead.
Domain specific configuration files are used to specify the unique options for a given domain with regard to its identity backend (for example [ldap] url to be used for that domain). This use of a separate configuration file for each domain leads to some complexity in the choreography of on-boarding a new domain, e.g.:
The two different mediums of where information about a domain is stored (i.e. identity driver information in the configuration file as well as domain name, description etc. in SQL) also makes it rather difficult to easily see what options are defined for a specific domain.
This proposal will provide the ability to store the configuration options in an SQL backend, indexed by domain ID. A main configuration option will dictate whether Keystone uses this extension to obtain domain specific configuration data or the existing domain specific configuration files approach. It will be an all or nothing option - there will be no mixing of some options in the extension and some in configuration files. If the extension is specified as the store of configuration data, then any domain-specific configuration files will be ignored. Keystone will go through the logic of building the domain configurations in the same way as it does today, either reading the configurations options from the extension or from the set files.
Whenever any of the domain-specific configuration options are updated, the identity backend for a given domain will be re-loaded (using whatever new configuration data there may be), further assisting the on-boarding of a domain.
This new capability can only be used to replace options that would have appeared in a domain specific file - it will not allow options to be specified that would normally only appear in the main Keystone configuration file. Since the options are available via REST, the openstack client and Horizon can provide the ability to view and set this information.
Although the configuration options will be stored in SQL, this does not imply anything about the medium of the identity backends themselves (which could be all LDAP if required).
One concern over this ability would be that some configuration options may be considered highly sensitive and would want to be stored separately and while can be written via the API, they cannot be read (for example password of the user required for doing LDAP searches, in the case when anonymous query is not supported). To address this concern, it is proposed that internally we enforce an explicit whitelist of config options that can be read. All others will be stored in a separate backend table and not returned on read. The current proposal is that all config options except password will be included in the whitelist. This means that the LDAP url will be returned on read (since that will be a common option cloud administrators might want to check). There may be situations where the url is required to encode a non-whitelisted item (e.g. password). To avoid these then being visible on read, we will support the substitution commands within any config strings that allow any non-whitelisted items to be referenced, by specifying within the option:
The configuration option specified for substitution, which must exist in the same group as the option being defined (in this case “url”), will be substituted when Keystone uses the configuration option internally.
This new capability will initially be classed as experimental in-tree functionality, with the intention of migrating to stable as soon as possible, with a stretch goal of this occurring before the release of Kilo.
Use of federation to access IdPs would (effectively) move this problem to creating mapping rules.
Changes to the data model will be restricted to a new tables for storing the configuration information.
The exact API specification will be defined as part of a review of changes to the Identity API.
This functionality exposes a new API to the backend configuration data for a domain. Like any other v3 API, it will be subject to the standard RBAC permissions model.
None, other than this functionality will subscribe to domain deletion events.
For cloud providers who have already deployed a domain-specific installation, a one-shot option to keystone-manage will be provided that will copy the contents of the domain-specific configuration files into the SQL store.
The work for supporting this API in Horizon will be proposed separately.
Beyond the regular unit testing, there will be testing for the migration options.
Changes to the Identity API and configuration.rst.