Skip to end of metadata
Go to start of metadata

The Policy subsystem of ONAP maintains, distributes, and operates on the set of rules that underlie ONAP’s control, orchestration, and management functions. Policy provides a centralized environment for the creation and management of easily-updatable conditional rules. It enables users to validate policies and rules, identify and resolve overlaps and conflicts, and derive additional policies where needed. Policies can support infrastructure, products and services, operation automation, and security. Users, who can be a variety of stakeholders such as network and service designers, operations engineers, and security experts, can easily create, change, and manage policy rules from the  Policy Manager in the ONAP Portal.

Policies are used to control, to influence, and to help ensure compliance with goals. A policy can be defined at a high level to create a condition, requirement, constraint, or need that must be provided, maintained, and enforced. A policy can also be defined at a lower or functional level, such as a machine-readable rule or software condition/assertion which enables actions to be taken based on a trigger or request, specific to particular selected conditions in effect at that time.  Some examples of types of policies are

  • VNF placement — rules governing where VNFs should be placed, including affinity rules
  • Data and feed management — what data to collect and when, retention periods, and when to send alarms about issues
  • Access control — who (or what) can have access to which data
  • Trigger conditions and actions — what conditions are actionable, and what to do under those conditions
  • Interactions — how interactions between change management and fault/performance management are handled (for example, should closed loops be disabled during maintenance?)

ONAP supports XACML policies and Drools rules.  Several configuration and operational policies are pre-loaded.

Once validated and corrected for any conflicts, the policies are placed in an appropriate repository, and are made available to the other subsystems and components that might make use of them. In addition, the decisions and actions taken by the policy are distributed. Policies are distributed either in conjunction with installation packages (if for example, the policy is related to service instantiation) or independently, if the policy is unrelated to a particular service.  With this methodology, policies will already be available when needed by a component, minimizing real-time requests to a central policy engine or PDP (Policy Decision Point). This improves scalability and reduces latency.

Figure 1 depicts the Policy architecture.


Figure 1. Policy high-level architecture

System Architecture

ONAP Policy is composed of several subcomponents: the Policy Administration Point (PAP), which offers interfaces for policy creation, and two types of Policy Decision Point (PDP), each based on a specific rules technology.  PDP-X is based on XACML technology  and PDP-D is based on Drools technology.  PDP-X is stateless and can be implemented using a resource pool of PDP-X servers.  The number of servers can be grown to increase both capacity (horizontal scalability) and to increase availability.  The PDP-D is stateful, as it utilizes Drools in its native, stateful way and transactions persist so long as the PDP-D is active.  Persistent Drools sessions, state management, local and geo-redundancy have been deactivated for the initial release of ONAP Policy and will be turned on in a future release.

As illustrated in Figure 2, the Policy components are supported by a number of interfaces and subsystems.  The ONAP Portal provides a human interface for the creation, management and deployment of policies.  It is a web-based system that utilizes internal APIs in the PAP.


Figure 2. Policy subsystem system architecture 

The PAP provides interfaces and management of policy definitions.  It utilizes the XACML database to store policy definitions, which are then distributed to the PDPs.

The XACML and Drools databases are hosted in a MariaDB cluster.  The XACML database is used to persist policy definitions and provide a point for PDPs to retrieve policy definitions.  The XACML database also has tables used for node state management, detection of node failure and failover. As indicated above, the state management tables will only include entries for the PAP and PDP-X as the testing is not yet complete for the PDP-D.

The PDP-X receives deployed policies and has interfaces to handle XACML policy transactions.  These transactions are stateless and once complete, they are removed from memory.  If a policy that is deployed to the PDP-X is of an operational nature it will contain Drools rules and Java executables.  These artifacts are processed into Maven artifacts and pushed to the Maven repository.  The PDP-D is then notified a new policy has been deployed.

When the PDP-D is notified a new policy has been deployed, it downloads it from the Maven repository and assigns it to an internal controller.  This controller provides the external Closed Loop interfaces to the DMaaP message bus over which events and messages are exchanged with external systems.  As events or messages arrive at the PDP-D, they are assigned to the appropriate controller and a Drools session is either created or retrieved from memory.  The events, messages or facts are passed to the Drools session and the rule engine is fired,  resulting in a change of internal session state and possibly actions taken in response to the rule processing. Response messages and requests are passed by the controller back over the DMaaP message bus to the appropriate system.  The Drools session can also have timers and autonomous events. In a future release the PDP-D will enable the node state management and session persistence in the Drools DB. 

Policy Creation

The Policy Creation component of the Policy subsystem enables creation of new policies and modification of existing polices, both online and offline.


Policy Distribution

After a policy has been initially created or an existing policy has been modified, the Policy Distribution Framework sends the policy from the repository to its points of use, known as Policy Distribution Points, before it is actually needed.

Separate notifications or events communicate the link or URL for a policy to the components that need it.  Then, when a component needs the policy, it uses the link to fetch it. Components in some cases might also publish events indicating that they need new policies, eliciting a response with updated links or URLs. Also, in some cases, policies can indicate to components that they should subscribe to one or more policies, so that they receive automatic updates to those policies as they become available.

Policy Decision and Enforcement

Run-time policy enforcement is performed by ONAP subsystems that are called from the policy. For example, policy rules for data collection are enforced by the data collection functionality of DCAE. Analytic policy rules, identification of anomalous or abnormal conditions, and publication of events signaling detection of such conditions are enforced by DCAE analytic applications. Policy rules for associated remedial actions, or for further diagnostics, are enforced by the correct component in a control loop such as the MSO, a Controller, or DCAE.


Policy Unification and Organization

Because the policy framework is expandable and multipurpose, it might contain many types of policies, which users might want to organize according to some useful dimensions.  Users can define attributes that specify the scope of policies, and these attributes can be extended to the policy-enabled functions and components. Useful policy organizing dimensions might include:

  • Policy type or category (taxonomical)
  • Policy ownership or administrative domain
  • Geographic area or location, 
  • Technology type  
  • Policy language and version 
  • Security level or other security-related values, specifiers, or limiters

Attributes can be specified for each dimension. In addition to being defined for individual policies themselves, these attributes can be used to define the scope of these additional additional policy-related functions:

  • Policy events or requests/triggers 
  • Policy decision, enforcement, or other functions 
  • Virtual functions of any type 

Policy-writers can define attributes so that policy events or requests self-indicate their scope. The scope is then examined by a suitable function and subsequently acted upon accordingly. Policy decisions and enforcement functions can self-indicate their scope of decision-making, enforcement, or other capabilities. Virtual functions can be automatically attached to the appropriate Policy Framework and distribution mechanisms.





















  • No labels

3 Comments

  1. I had seen in some of the architecture diagrams earlier that the policies are also available at VNF level. These policies can direct VNF what to report, frequency etc. Is my understanding correct and this kind of capability is supported in the current release? Since it also puts the requirement on VNF supplier to provide an interface and support such policies.

  2. Editing the line-

    VNF placement — rules governing where VNFs should be placed, including affinity rules

    Earlier it was , VM placement. 
    reference - https://www.youtube.com/watch?v=Hldl2f2nWeI&list=PLxgUkHTvXNoY0JxgkO26y70swCGOdngu7&index=1   (4.47)

  3. I am suggesting to mention, flow diagram on config, action and decision policy. And please also mention difference among these policies with examples.