...
Code Block | ||||||
---|---|---|---|---|---|---|
| ||||||
tosca_definitions_version: tosca_simple_yaml_1_0_0 policy_types: onap.policies.controlloop.operational.TriggerStylePolicyType: derived_from: tosca.policies.Root version: 1.0.0 description: Operational Policy for Control Loop execution using Triggers metadata: properties: controlLoopName: type: string description: The unique control loop id required: true timeout: type: integer description: Overall timeout for executing all the operations in the policy. required: true targets: [] # optional must be Node Types or Group Types applied to triggers: # ? topology_template: policies: - firewall_restart: type: onap.policies.controlloop.operational.TriggerStylePolicyType description: vFirewall use case - APPC restart metadata: properties: controlLoopName: ControlLoop-vFirewall-d0a1dfc6-94f5-4fd4-a5b5-4630b438850a timeout: 1200s targets: [] # optional must be Node Templates or Groups applied to triggers: - on_onset: description: When the ONSET event is received event_type: ControlLoopEvent schedule: start_time: T06:00:00Z end_time: T23:59:59Z target_filter: node: requirement: capability: condition: action: # scripts? primary: # The optional implementation artifact (i.e., the primary script file within a TOSCA CSAR file). dependencies: # The optional ordered list of one or more dependent or secondary implementation artifacts which are referenced by the primary implementation artifact timeout: # timeout in seconds operation_host: # node executed on constraint: # seems the same as condition above period: 300s evaluations: 3 # is this retries? The optional number of evaluations that must be performed over the period to assert the condition exists. method: # optional statistical method name to use to perform evaluation of condition |
2.3 Flat Structure - v1
Code Block | ||||||
---|---|---|---|---|---|---|
| ||||||
tosca_definitions_version: tosca_simple_yaml_1_0_0 policy_types: onap.policies.controlloop.operational.Common: derived_from: tosca.policies.Root version: 2.0.0 description: Operational Policy for Control Loop execution properties: id: type: String description: The unique control loop id required: true timeout: type: Integer description: Overall timeout for executing all the operations. required: true abatement: type: Boolean description: Whether an abatement event message will be expected for the control loop. required: true default: false operation: description: List of operations to be performed when Control Loop is triggered. required: true entry_schema: type: onap.datatype.controlloop.Operation trigger: type: String description: Initial operation to execute upon receiving an Onset event message for the Control Loop. required: true onap.policies.controlloop.operational.Apex: derived_from: onap.policies.controlloop.operational.Common version: 1.0.0 description: Operational policies for Apex PDP properties: onap.policies.controlloop.operational.Drools: derived_from: onap.policies.controlloop.operational.Common version: 1.0.0 description: Operational policies for Drools PDP properties: controller: description: Drools controller properties required: true entry_schema: type: onap.datatype:controlloop.controller.Drools data_types: onap.datatype.controlloop.controller.Drools: derived_from: tosca.datatypes.Root version: 1.0.0 description: Definition of the target type for operations during control loops properties: onap.datatype.controlloop.Operation: derived_from: tosca.datatypes.Root version: 1.0.0 description: An operation supported by an actor properties: id: type: String description: Unique identifier for the operation required: true description: type: String description: A user-friendly description of the intent for the operation required: false actor: type: String description: The actor that will perform the operation required: true operation: type: String description: The operation the actor should perform on the target required: true target: type: String description: The target the operation is to be performed on. required: true payload: type: Map description: Optional payload that is passed from the policy to the actor for operation execution. required: false timeout: type: Integer description: The amount of time for the actor to perform the operation. required: true retries: type: Integer description: The number of retries the actor should attempt to perform the operation. required: true default: 0 success: type: String description: Points to the operation to invoke on success. A value of "final_success" indicates and end to the operation. required: false default: final_success failure: type: String description: Points to the operation to invoke on Actor operation failure. required: false default: final_failure failure_timeout: description: Points to the operation to invoke when the time out for the operation occurs. required: false default: final_failure_timeout failure_retries: description: Points to the operation to invoke when the current operation has exceeded its max retries. required: false default: final_failure_retries failure_exception: description: Points to the operation to invoke when the current operation causes an exception. required: false default: final_failure_exception failure_guard: description: Points to the operation to invoke when the current operation is blocked due to guard policy enforcement. required: false default: final_failure_guard |
Issue: The "policies" property can be confusing with the TOSCA Topology Template value "policies".
- Pamela Dragosh - use "operations" instead since it really is a set of operations that are really being enforced.
Issue: How to ensure that CLAMP does not have to hard-code the meaning of properties. Specifically, success, failure, failure_timeout etc. - need to point to other actor/recipe's to be triggered/executed.
Issue: How to ensure that CLAMP does not have to hard-code where CSAR details need to be populated in the policy. Eg. Target Type
- Need to use defaults, descriptions, constraints, required fields
- Pam and Liam will work on Operational policies to look like Seb's JSON schema so Seb can test it out. Then identify next problem to solve.
Seb's JSON schema from El Alto:
Issue: TOSCA Constraints would be nice to have something with suggested values, so the policy isn't locked into a set of values. Example is "actors" - ONAP has a set of actors, but it would be nice to be able to extend to other internal actors companies support.
2.3 Flat Structure - v2
Flat Structure vs - use of metadata and actor as a datatype
Code Block | ||||
---|---|---|---|---|
| ||||
tosca_definitions_version: tosca_simple_yaml_1_0_0
policy_types:
onap.policies.controlloop.operational.Common:
derived_from: tosca.policies.Root
version: 2.0.0 | ||||
Code Block | ||||
# REST of policy ommitted for brevity onap.datatype.controlloop.operationalpolicies: derived_from: tosca.datatypes.Rootdescription: Operational Policy for Control Loop execution versionproperties: 1.0.0 descriptionid: Policy definitions for taking actions during a Control Loop event type: String properties:description: The unique control loop id # other properties ommitted forrequired: brevitytrue actortimeout: type: StringInteger description: TheOverall actortimeout thatfor willexecuting performall the operationoperations. required: true constraints: possible_values: [SO, SDNC, APPC, VFC, SNDR, CDS] |
Issue: TOSCA does NOT have a way to relate properties together. For example, if the actor property is selected as APPC, then the contraints (eg possible values) for the operation property SHOULD be "restart", "reboot", "modifyConfig". How can that be specified?
Issue: Can/Should we consolidate onap.policies.Operational into a common policy type and then allow Drools and Apex (or be extendable to companies ingesting ONAP) to be derived_from that policy.
Question: Should the actor be a capability? If its a datatype, then you cannot pre-define them. A capability type would allow us to predefine actors as well as allow companies that ingest ONAP to create their own.
Question: How does CLAMP call CDS API for blueprint information? to be filled in
TODO: Pam add examples.
Question: If CSAR is prepopulated with datatypes and policytypes, how would one design a PNF control loop? Do we need a PNF service distributed to it to "kick off"?
- Possible to have re-usable Operations?
- Jorge Hernandez will investigate re-usable operations
- Issue: Flat structure vs Nested structure
- Jorge Hernandez will investigate flat policies (old way)
- Pamela Dragosh will investigate nested policies
- Issue: We should define the incoming event that triggers the Control Loop Pamela Dragosh will add this definition
- Current ControlLoopEvent structure
- Future possible VES Events
- Possible that status of a policy execution may need to be an event for reusable operations
- Issue: Should it be capability_type or relationship for the Actor operations??
- undefined undefined will investigate
- Liam Fallon and Ram Krishna Verma will look into Apex Operational Policy.
- Issue: Can we specify the guard policies into the operational policies? Do we make guard optional in policy operational directive executive? How to make it available to all PDP engines?
- Issue: Other policies needed such as is-closed-loop-disable=true and prov_status=in-maint:
Jira server ONAP JIRA serverId 425b2b0a-557c-3c0c-b515-579789cceedb key POLICY-1071 - Issue: A&AI how to specify which VNF to take directive on. What about PNF?
- Could switch to where a selection of the operation constrains the actor.
- Instead of specify BOTH actor and operation, select an operation and then the actor supported by the operation.
Issue: Policies and Reusable Operations Action Item Follow up - Jorge Hernandez
A nested policy approach has the following issues:
- Number of hierarchical layers, which must be bounded. The CLAMP GUI driven by an open ended "recursive" schema would have to be bounded. One could argue that after N hierarchical layers, the information would be hard to be tracked by a human operator.
- Associated with the number of hierarchical layers, and the most important issue is the duplication of operations, as duplicated operations would be repeated at multiple levels, and branches.
- Associated with (2), there are issues of conciseness and human readability of policies.
- A smaller issue relates to the storage space in the policy repository as it would likely require more space due to dups operations across levels.
- A smaller issue relates to come up with canonical representation of policies and operations, policies will have to be flattened out and put in a canonical representation to support static analysis on the policy repo, for example for conflict detection.
The main benefits for nested operations over a flat structure are:
- Clearly drive the GUI display and the user input (may be more intuitive), and
- if the operations are fairly simple (policies mono-operational) which is the high runner case at this time (minimum duplication).
...
abatement:
type: Boolean
description: Whether an abatement event message will be expected for the control loop.
required: true
default: false
operation:
description: List of operations to be performed when Control Loop is triggered.
required: true
entry_schema:
type: onap.datatype.controlloop.Operation
trigger:
type: String
description: Initial operation to execute upon receiving an Onset event message for the Control Loop.
required: true
onap.policies.controlloop.operational.Apex:
derived_from: onap.policies.controlloop.operational.Common
version: 1.0.0
description: Operational policies for Apex PDP
properties:
onap.policies.controlloop.operational.Drools:
derived_from: onap.policies.controlloop.operational.Common
version: 1.0.0
description: Operational policies for Drools PDP
properties:
controller:
description: Drools controller properties
required: true
entry_schema:
type: onap.datatype:controlloop.controller.Drools
data_types:
onap.datatype.controlloop.controller.Drools:
derived_from: tosca.datatypes.Root
version: 1.0.0
description: Definition of the target type for operations during control loops
properties:
onap.datatype.controlloop.Actor:
derived_from: tosca.datatypes.Root
version: 1.0.0
description: An actor/operation/target definition
properties:
actor:
type: String
required: true
operation:
type: String
required: true
target:
type: String
description: The resource the operation should be performed on
required: true
metadata:
clamp_csar_lookup: csar_plugin
payload:
type: Map
description: Name/value pairs of payload information passed by Policy to the actor
required: false
onap.datatype.controlloop.Operation:
derived_from: tosca.datatypes.Root
version: 1.0.0
description: An operation supported by an actor
properties:
id:
type: String
description: Unique identifier for the operation
required: true
description:
type: String
description: A user-friendly description of the intent for the operation
required: false
operation:
description: The actor that will perform the operation
required: true
metadata:
clamp_catalog: get_operations
entry_schema:
type: onap.datatype.controlloop.Actor
timeout:
type: Integer
description: The amount of time for the actor to perform the operation.
required: true
retries:
type: Integer
description: The number of retries the actor should attempt to perform the operation.
required: true
default: 0
success:
type: String
description: Points to the operation to invoke on success. A value of "final_success" indicates and end to the operation.
required: false
default: final_success
failure:
type: String
description: Points to the operation to invoke on Actor operation failure.
required: false
default: final_failure
failure_timeout:
description: Points to the operation to invoke when the time out for the operation occurs.
required: false
default: final_failure_timeout
failure_retries:
description: Points to the operation to invoke when the current operation has exceeded its max retries.
required: false
default: final_failure_retries
failure_exception:
description: Points to the operation to invoke when the current operation causes an exception.
required: false
default: final_failure_exception
failure_guard:
description: Points to the operation to invoke when the current operation is blocked due to guard policy enforcement.
required: false
default: final_failure_guard
|
Issue: The "policies" property can be confusing with the TOSCA Topology Template value "policies".
- Pamela Dragosh - use "operations" instead since it really is a set of operations that are really being enforced.
Issue: How to ensure that CLAMP does not have to hard-code the meaning of properties. Specifically, success, failure, failure_timeout etc. - need to point to other actor/recipe's to be triggered/executed.
Issue: How to ensure that CLAMP does not have to hard-code where CSAR details need to be populated in the policy. Eg. Target Type
- Need to use defaults, descriptions, constraints, required fields
- Pam and Liam will work on Operational policies to look like Seb's JSON schema so Seb can test it out. Then identify next problem to solve.
Seb's JSON schema from El Alto:
Issue: TOSCA Constraints would be nice to have something with suggested values, so the policy isn't locked into a set of values. Example is "actors" - ONAP has a set of actors, but it would be nice to be able to extend to other internal actors companies support.
Code Block |
---|
# REST of policy ommitted for brevity
onap.datatype.controlloop.operationalpolicies:
derived_from: tosca.datatypes.Root
version: 1.0.0
description: Policy definitions for taking actions during a Control Loop event
properties:
# other properties ommitted for brevity
actor:
type: String
description: The actor that will perform the operation
required: true
constraints:
possible_values: [SO, SDNC, APPC, VFC, SNDR, CDS]
|
Issue: TOSCA does NOT have a way to relate properties together. For example, if the actor property is selected as APPC, then the contraints (eg possible values) for the operation property SHOULD be "restart", "reboot", "modifyConfig". How can that be specified?
Issue: Can/Should we consolidate onap.policies.Operational into a common policy type and then allow Drools and Apex (or be extendable to companies ingesting ONAP) to be derived_from that policy.
Question: Should the actor be a capability? If its a datatype, then you cannot pre-define them. A capability type would allow us to predefine actors as well as allow companies that ingest ONAP to create their own.
Question: How does CLAMP call CDS API for blueprint information? to be filled in
TODO: Pam add examples.
Question: If CSAR is prepopulated with datatypes and policytypes, how would one design a PNF control loop? Do we need a PNF service distributed to it to "kick off"?
- Possible to have re-usable Operations?
- Jorge Hernandez will investigate re-usable operations
- Issue: Flat structure vs Nested structure
- Jorge Hernandez will investigate flat policies (old way)
- Pamela Dragosh will investigate nested policies
- Issue: We should define the incoming event that triggers the Control Loop Pamela Dragosh will add this definition
- Current ControlLoopEvent structure
- Future possible VES Events
- Possible that status of a policy execution may need to be an event for reusable operations
- Issue: Should it be capability_type or relationship for the Actor operations??
- undefined undefined will investigate
- Liam Fallon and Ram Krishna Verma will look into Apex Operational Policy.
- Issue: Can we specify the guard policies into the operational policies? Do we make guard optional in policy operational directive executive? How to make it available to all PDP engines?
- Issue: Other policies needed such as is-closed-loop-disable=true and prov_status=in-maint:
Jira server ONAP JIRA serverId 425b2b0a-557c-3c0c-b515-579789cceedb key POLICY-1071 - Issue: A&AI how to specify which VNF to take directive on. What about PNF?
- Could switch to where a selection of the operation constrains the actor.
- Instead of specify BOTH actor and operation, select an operation and then the actor supported by the operation.
Issue: Policies and Reusable Operations Action Item Follow up - Jorge Hernandez
A nested policy approach has the following issues:
- Number of hierarchical layers, which must be bounded. The CLAMP GUI driven by an open ended "recursive" schema would have to be bounded. One could argue that after N hierarchical layers, the information would be hard to be tracked by a human operator.
- Associated with the number of hierarchical layers, and the most important issue is the duplication of operations, as duplicated operations would be repeated at multiple levels, and branches.
- Associated with (2), there are issues of conciseness and human readability of policies.
- A smaller issue relates to the storage space in the policy repository as it would likely require more space due to dups operations across levels.
- A smaller issue relates to come up with canonical representation of policies and operations, policies will have to be flattened out and put in a canonical representation to support static analysis on the policy repo, for example for conflict detection.
The main benefits for nested operations over a flat structure are:
- Clearly drive the GUI display and the user input (may be more intuitive), and
- if the operations are fairly simple (policies mono-operational) which is the high runner case at this time (minimum duplication).
- Recommendation is to stay with the flat structure previously used vs nested. Not ideal but use description field as a way of instructing the user on what values to put in the fields.
- Determined that we need a catalog for actors/operations and what target they work on
Actor Catalog needs to capture this information:
Code Block |
---|
catalog:
appc:
ModifyConfig: [VNF]
appclcm:
ModifyConfig: [VNF]
Restart: [VNF]
Rebuild: [VNF]
Migrate: [VNF]
so:
VF Module Create: [VFModule]
VF Module Delete: [VFModule]
cds:
ModifyConfig: [VNF, PNF]
sdnc:
reroute: [VNF]
BandwidthOnDemand: [VNF]
sdnr:
ModifyConfig: [PNF]
vfc:
Restart: [VNF]
|
Proposal is to use "metadata" field in order to instruct CLAMP on where to lookup actor information and how to apply logic around looking up information in the CSAR for actor/operation.
...