CPS-3194: Cloud Infrastructure Support

CPS-3194: Cloud Infrastructure Support

References

Goal

Support E//-managed Cloud Infrastructure (where Cloud RAN is deployed) in CPS-NCMP.

This adds 4 new top-level node types: NetworkElement, CloudInfrastructure, CloudInfrastructureManager, OCloud — in addition to the existing MeContext, SubNetwork, ManagedElement. Discovery, registration and the CM Handle inventory model are unchanged; the new nodes register as ordinary CM Handles.

"Top-level" means: when one of these types is present, it is the last element in the alternate-id (FDN). They behave exactly like the existing specific/terminal types (MeContext/ManagedElement), including the contains(id, ...) search and longest-match resolution. Container segments (e.g. SubNetwork) may precede them.

FDN examples for NetworkElement (the other 3 types follow the same rule):

/NetworkElement=NE1 /SubNetwork=SN1/NetworkElement=NE1 /SubNetwork=SN1/SubNetwork=SN2/NetworkElement=NE1 /SubNetwork=SN1/MeContext=ME1/NetworkElement=NE1 /SubNetwork=SN1/SubNetwork=SN2/MeContext=ME1/NetworkElement=NE1

Code analysis — affected areas & changes needed

A codebase-wide search was done for every consumer of alternate-id resolution (all callers of AlternateIdMatcher) and every user of JexParser. The node-type names are hardcoded as a matching rule in only two production spots, both on the data-job read path. Everywhere else the FDN/reference is treated as an opaque string and passed straight to the (type-agnostic) matcher, or built by the generic FDN-prefix parser — so those paths already handle the new types with no change.

Changes needed

File

What / why

Change

File

What / why

Change

1

JexParser.java — CONTAINS_PATTERN

Extracts the search term from a contains(id, …) selector (used by extractSearchTerm). Only matched ManagedElement|MeContext.

Add the 4 new types to the alternation

2

ReadRequestExaminer.java — SPECIFIC_NODE_ID_PATTERN

isBroadcast() uses this to decide specific-node vs broadcast. Without the new types, a /NetworkElement[id=…] selector was wrongly treated as broadcast (fanned to all DMIs) instead of resolved to one CM Handle.

Add the 4 new types to the alternation

3

JexParserSpec.groovy

Unit coverage (100% rule)

Extend extractFdnPrefix + extractSearchTerm tables with the 4 types and the NetworkElement FDN examples

4

ReadRequestExaminerSpec.groovy

Unit coverage (100% rule)

Add a data-driven test asserting the 4 types classify as specific nodes (routed to DMI, not broadcast)

The generic FDN-prefix builder JexParser.extractFdnPrefix matches any Name[id="…"] segment, so it already rebuilds all the FDN examples above correctly. Only the two search/specific-node patterns are type-gated.

Verified NOT affected (no change needed)

Consumer

How it resolves the FDN

Why no change

Consumer

How it resolves the FDN

Why no change

1

WriteRequestExaminer (data job write)

Uses writeOperation.path() directly → matcher

No JexParser, no type regex — segment-agnostic

2

NetworkCmProxyController (legacy /ncmp)

Passes decoded reference string → matcher

Opaque string; no type gate

3

DmiDataOperations (legacy passthrough)

Passes cmHandleReference → matcher

Opaque string; no type gate

4

ProvMnSCmHandleRetriever / ProvMnS controllers

Passes DN/fdn → matcher

Path/DN-based; no type gate

5

CmSubscriptionHandlerImpl (subscriptions)

Uses generic extractFdnPrefix

Already type-agnostic

6

CmResourceAddress, NcmpCachedResourceRequestHandler

getCmHandleId(reference)

Opaque reference; no type gate

7

AlternateIdMatcher (the resolver itself)

Longest-match on the FDN string

Segment/type-agnostic by design

Issues & Decisions

Issue

Notes

Decision

Issue

Notes

Decision

1

Which cloud node types must be supported?

NetworkElement

CloudInfrastructure

CloudInfrastructureManager

OCloud

Agreed Csaba Kocsis

2

Do the 4 new types require YANG models?

No change for NCMP discovery & registration - register/discover via the existing flow.

Agreed Csaba Kocsis

3

What identifies a cloud-infra CM Handle (type field, module set tag, property)?

None needed. Discovery/registration unchanged, so the nodes register as ordinary CM Handles; alternate-id resolution keys off FDN segment names, not a type field.

Agreed Csaba Kocsis

4

Performance / KPI impact

100k cm handles with use of optimised model

 

5

DCM read impact

Can be seperated

"DCM" = the test-only test-dcm-rest harness (DataJobControllerForTest, /do-not-use/dataJobs) over the data-job read path (DataJobService → ReadRequestExaminer). That path is exactly what Story 1 updates; no separate production impact.

Agreed Csaba Kocsis

Out of scope

Legacy batch and model request handling. Assumes optimised model enabled and ModuleSetTag reused as-is (as IaS).