Some Thoughts on Associations in 4.9

Last modified 05 Mar 2024 10:29 +01:00
Table of Contents

Use Cases

The following use cases should be supported both in ConnId and in midPoint.

Structured Source Information

Consider HR system that models persons and their contracts to the organization, connected to organizational unit plus an independent structure of cost centers:

HR structure
Figure 1. HR structure

There are many more conceivable types of objects, like addresses of a person: permanent, temporary, correspondence, and so on.

Addresses in HR
Figure 2. Addresses in HR

Rich Membership

Consider a target resource that does not model accesses using traditional membership relation between users and groups, but using more elaborate structures. For example, let’s have a document management system (DMS) that has users, document stores, and access rights.

Document management system structure
Figure 3. Document management system structure

Plain Membership

We still want to support "plain" membership information, i.e., one without any additional data.

Like in Active Directory:

AD structure
Figure 4. AD structure
This is an entity-relationship diagram but not for a database, so it’s somehow tweaked. "Sec" means the secondary identifier there. Also, the two relations are, in fact, two manifestations of a single conceptual many-to-many relation: group membership of a user.

Required ConnId Support for Querying and Updating

At the level of ConnId, we need to query and update individual data elements.

Plain Membership

We need to support these simple relations bidirectionally, that is, they should be queryable and updatable potentially from both sides:

  • when reading user-side attribute (e.g., memberOf), we would like to get all the groups the user is member of;

  • when updating this attribute (e.g., adding or removing DNs from memberOf), we would like to modify the group membership of this user.

We need to be aware that:

  1. There are resources that provide no user-side information at all.

  2. There are resources that provide user-side information only for querying, not for updating.

  3. There may be some limitations, like in AD, where the memberOf attribute provides DNs of all the groups the user is member of, except for the user’s primary group.

    That’s why we probably should not reuse memberOf for the purpose of this connector-managed query-and-update-membership feature, but provide a different name.

Structured Source Information & Rich Membership

Again, also here we would like to query, and probably update the associated objects (contract in the HR case, accessRight in the DMS case) right on the owning object: person in HR, user in DMS.

Proposed ConnId Design

There are two new concepts there:

  1. Associated objects

  2. Associations

Associated Objects

Structured Source Information & Rich Membership Cases

  1. The contract, address, accessRight will be fully functional connector objects.

  2. To distinguish them from the ordinary objects, let us call them associated objects. They have their object class definitions, with a flag that indicates that the object class is an associated object class.

  3. The ordinary objects that are connected to the associated objects are called owning objects, or owners.

  4. These associated objects will be accessible as part of their owners' search operations. (Not necessarily from all the sides, though.)

    For example, when retrieving a person, the client could request returning also the values of (multivalued) contract item: either identifiers (contractId values), or the whole contract objects. The client would request which behavior (ignore, return identifiers, return full values) it desires. The full objects could be returned in the form of ConnectorObject instances, embedded in the item values being returned.

    It is not clear if the item containing reference or references to the associated objects (e.g., the contract item) should be an attribute or something different, like an association, or a reference. It could be an attribute of a specific non-primitive type: ObjectAssociation or maybe ObjectReference. There are reasons for both approaches. For the time being, let us talk about them as association attributes or reference attributes.

  5. In theory, the client could request also deeper resolution, like providing also resolved orgUnit objects in the contract objects returned. But in the near future this probably won’t be necessary.

  6. When updating, there are three cases:

    1. Adding associated objects. This is most naturally done by providing the association/reference attribute values with the embedded objects (in object create or modify operation).

    2. Deleting associated objects. This can be done by providing identifiers of the embedded objects to be dis-associated from the owning object. (The question if the associated objects should be deleted or not. It depends on whether relation between subject and the associated object is mandatory at the associated object’s side. If so - and this is perhaps the most usual case - the associated object will need to be deleted.)

    3. Modifying associated objects. This will have to be a modification operation against the associated object class. But the primary identifier (Uid in the ConnId language) won’t be necessarily a single attribute. For associated objects like accessRight in the DMS example, it will be a composite identifier, consisting of uid, storeId, and accessLevel attributes.

Plain Membership Case

There won’t be any extra object classes.

Associations / References

What is needed, though, is a new concept of an association / reference [attribute]. It is a kind of link between objects that can be queried and updated in a flexible way. For example, a link between AD user and AD group is an association/reference.

An association/reference always has two ends. It can be one-to-one, one-to-many, or many-to-many. If it’s one-to-X, the "one" end may be optional or mandatory.

The association/reference will have a name that is potentially different on its ends.

Plain Membership Case

For example, the association/reference implementing the group membership in AD can be named member or members on the group side and group or groups on the user (member) side. Both these sides will be multivalued.

Structured Source Information & Rich Membership Case

Here, there will be associations/references between the ordinary objects (person, orgUnit, costCenter in HR, and user, documentStore in DMS) and the associated objects (contract in HR, accessRight in DMS).

For example, a person will have an association/reference named contract that will point to its associated contract objects. There may or may not be a manifestation of this association/reference at the contract side. (It is not necessary mainly because it probably won’t be addressed, as it is not updated from the outside.)

The contract will have associations/references named orgUnit and costCenter pointing to the orgUnit and costCenter objects, respectively. Again, there need not be manifestations of these associations/references at the target sides, for the same reason as above.

For the DMS case, the user will have an association/reference named accessRight pointing to its associated accessRight objects. As the accessRight objects will be probably queried and/or updated separately from the user objects, this reference will be represented also in accessRight object, probably as user item. The same is true for the documentStore side of accessRight.

Summary

For ConnId, we need two more or less independent new concepts:

  1. association/reference,

  2. associated object class.

The association/reference can exist without an associated object class, like in plain membership cases. But associated object class requires associations/references to bind the associated objects to their owners.

Proposed MidPoint Design

How should all of this be represented in midPoint?

Sources with Associated Object Classes

For strictly source resources, like HR, this situation is quite simple. For each associated object class, we can define object type (or types), and create the binding to midPoint data.

For simplicity, we can assume that the associated objects will be retrieved together with the owner object. (This means that there won’t be any import tasks dealing with contract object class. But that’s OK.)

Each associated object type will be mapped to a specific focus item. For example,

  • The contract will be mapped to assignment, with the subtype (in the future, archetype) of contract.

  • The permanentAddress (if specified as an association) will be mapped to extension/primaryAddress of type ext:AddressType.

  • The correspondenceAddress (again, if specified as an association) will be mapped to extension/correspondenceAddress of type ext:AddressType.

Notes:

  1. There may be items of the associated object that will get mapped directly into the user’s properties. For example, we may want to fill-in organizationUnit property with the values of orgUnit attribute in contract object. This can be configured simply by specifying $focus/organizationalUnit as the target in the corresponding mapping. (Because the default will be $focus no longer! It will be the assignment value into which we are mapping the respective contract.)

  2. We assume that the target item values (e.g., assignment values) will be created and deleted along with the appearance and disappearance of the source values. (This is very similar to the behavior of addFocus and deleteFocus synchronization reactions at the ordinary resource object level.) So, we only need to know what are the keys for the target item, i.e. when there is a change in the contract object - assuming that it has no externally visible ID - whether we have to create a new assignment, or just update the existing one.

Sources with Plain Associations/Relations

TODO

Targets

There are a lot of open questions here.

TODO

Was this page helpful?
YES NO
Thanks for your feedback