CPS-1301: Insert Module Prefixes

1

Sample (wanted) output of: /module1:A/B/module2:C/D

See also decision #3

2

Prefix can be inserted before each container but this is optional

See RFC...

1

Performance issue with current solution

Current solution used inefficient query to populate (first) prefix property for ALL fragments although only used for top level 

Will be handled by separate Jira already in progress, see CPS-1326: Creation of DataNodeBuilder with module name prefix is very slow Closed

2

Top level container is not always defined in first module

This might not be required as the first short-term solution is delivered first

Will be handled by separate JiraCPS-1344: Top level container (prefix) is not always the first moduleClosed

3

Add prefix for container in same module

This is allowed but not done normally. Team concerned about performance issues and impact on URLs when using this

@Ahila P Oct 21, 2022 : Prefix is not required for the child modules or nodes, if it carries the same prefix as parent

4

Add prefix to first container when this is not the top level for the module it is defined

retrieve B or C directly (without parent), in that case should the prefix stil be used ie. Module1:/B

Team Oct 21, 2022 Prefix required for top level nodes in output

5

/module1:A/B/module2:C/D 
Which module reference should be inserted in the cpaths for 'Module1' and 'Module2'. Prefix or Name, or last token of namespace?!

See sample from https://github.com/YangModels/yang/blob/main/vendor/cisco/xr/534/ietf-inet-types.yang

ietf-inet-types.yang

(Unfortunately in above example module name and last token of namespace are the same, not sure if this just convention or mandatory
Use 'ietf-inet-types' or 'inet' to insert 

CPS Team (made-up) example

bookstore.yang

@Ahila P Oct 21, 2022  It refers to the 'prefix' in the yang model  (highlighted in the sample yang model).

module ietf-inet-types {  

namespace "urn:ietf:params:xml:ns:yang:ietf-inet-types";  

prefix "inet";

Prefix for (additional) leaves that are defined in a separate module then the container itself

Required, not sure how common this is. DB Solution for this might be quite difficult. Toine's POC should handle it (to be tested)

How to handle lists (instead of container)

Possibly ignore [@key=value] part. Small neglectable change of same model defined both under same parent... Toine's Poc can be modified to handle this with help from establish regular-expressions from other use cases

Performance Considerations

SchemaContext generation is expensive. 

1. Consider distributed cache (already used for other things in CPS) to store resolved paths 

2. Consider normal cache for schemaContext object has they are not serializable)

3. Consider generating matcher for all possible paths upon SChemaContext creation (would be limited number) and store in (distributed) cache (with a reasonable time to live)

Use Hazelcast to store 'simplified Model' as defined in Poc Toine

9

Is 'prefix' attribute mandatory for any yang module

@Ahila P checked Documentation not clear btu in practice creating a yang module without prefix gives a 'compilation' liek error in PiYang

Oct 25, 2022 base in Ahila's feedback we assume it is mandatory.

Oct 25, 2022 

@Toine Siebelink @arne.chrestin Lena Peuker (DT) @Ahila P @Priyank Maheshwari @Joe K 

• Open Issues, Roadmap, resouring, Wow

Nov 28, 2022 

@Ahila P @arne.chrestin @Lena Peuker @Toine Siebelink @Andras Zoltan Kiss @Joe K @Levente Csanyi @Luke Gleeson (Unlicensed) @Seán Beirne @Tony Finnerty 

• @Ahila P demonstrated her current solution based on storing prefix (module-name) upfront i.e as part of data in fragment table.

• Pros

  • seems to work for all output use-cases including edge cases such as leaves etc.

  • module name prefix gets inserted in the first level of the output which might not be the top level element in a module. 

  • compatible with SDNC behavior (@arne.chrestin confirmed too)

• Cons

  • Client needs to specify prefix when using xPath requests

  • Backward incompatible

  • cpsPath (search) capabilities broken

  • Impacts on performance (unclear)

  • modulename prefix not per Yang RFX(see decision #5) above (TBC)

• Actions

  • @Toine Siebelink to check prefix v. modulename RFC guideline

  • @Ahila P store 'xPath-with-prefixes' in a  new field in fragment table  to maintain backward compatibility

  • @Ahila P break down required functionality in smaller user stories including

    • make xPath queries check both possible paths (with and with prefixes) if needed

    • use of prefixes with cpsPatrh functionality 

    • performance impact of new solution

DB Solution, Part 1 (containers and list)

1. Detailed Analysis incl. PoC: 2 man-weeks

2. DB Updates to store prefix using Liquibase: 4-6 Manweeks

3. Save Data update (using some of the PoC Code): 1-2 Weeks

4. Insert Prefix as needed on dat retrieval  (using some of the PoC Code): 1 Week

5. Develop demo model with multiple modules and update CSIT: 1-2 weeks

6. Performance Test and Tuning if needed: 1 week

14 Man weeks

Excludes ramping-up in CPS

7 people in parallel weeks, still seems optimistic

DB Solution, Part 2 (leaves)

@Toine Siebelink 

PTL, Manage Jira's, Generaal support

@Ahila P 

Analysis & Implementation

Analysis details are updated at CPS-1380: Node Prefix Issue AnalysisClosed

@arne.chrestin 

PO & Devolvement

Available after next internal product build

@Lena Peuker 

PO & Devolvement

Available after next internal product build