validation only required in Service-Layer. REst layer wil be handled using exception handling. Persistence Layer will depend on validation depend beforehand.
agreed that xPath stored in DB is for 'internal use' only and not exposed to end-user. We can normalize storage for better compatibility and adherence to xpath conventions when retrieving it. Lathishbabu Ganesan will work out prosed solution and do some PoC for this.
829
Priyank Maheshwari and Sourabh Sourabh are working together on this and with help of Bruno Sakoto they can now test using Kafka docker image approach which is good enough for now. Any (future) Strimzi Kafka integration is Integration Team responsibility (Fiachra Corcoran) and onto for CPS team.
agreed that xPath stored in DB is for 'internal use' only and not exposed to end-user. We can normalize storage for better compatibility and adherence to xpath conventions when retrieving it. Lathishbabu Ganesan will work out prosed solution and do some PoC for this.
aditya puthuparambil CPS-847 no need for 3 variations of operations in CSIT test. Issue with deployment of CPS_Temporal related to bguf Bruno Sakoto is working on. Aditya and Bruno will work together to resolve. Team test strategy needs to consider 'component/black-box testing' Toine Siebelink will call meeting to follow that up.
Exact structure of response, team prefer format with status for each node so successful updates could (later) be included too. Toine Siebelink wil check with stakeholders
Renu Kumari observed logging management require no authorization but could impacts system! Only accessible from inside cluster which is good enough for now.
The logging demonstrated is only to be turned on for specific functional tets to determine time spent during functional execution. Load test wil be affected too much by this.
Logging (configuration) guidelines will be documented as part of
Jira
server
ONAP Jira
serverId
425b2b0a-557c-3c0c-b515-579789cceedb
key
CPS-604
The logging introduced here probably needs re-start of service as it is enable using an annotation. We'll await client-feedback to see if this suffices.
'restconf' responses are somewhat smaller then equivalent 'rests' responses. This needs some further investigation. CSIT test using the new version of SDN-C will determine if there are significant issues but this is not expected
Some small changes were seen comparing legacy (SDNC 1.8) output with current output on the 'rests' interface. But these could be related to small model and data instance updates that have been part of the new SDN-C Deployment
Follow up E2E Slicing use-case impacts with Niranjana Y (now limited to
CPS team will work on datastores:operational cache during Jakarta
Toine Siebelink has added a new work item (#15 on R10 release plan) to allow update of cache when SDN-R updates corresponding device data through a messaging based solution)
E2E Slicing team will continue to write using SND-R during Jakarta and work with CSP team to fully cached solution in next release (Kohn)
CPS team wil meet again with Ahila P to follow up and clarify Jakarta scope
This will have some impacts on TBDMT process of data.
suggested to update template to handle required transformation to original data format (get child from parent)0 CPS Team can help out with these impacts to speed up changes. Ahila P will consider and revert to Toine Siebelink and Luke Gleeson
For 'passtrough' (bypassing cache) CPS-E-05 can already support XML (using JSON wrapping)
Cache not yet enabled in NCMP (planned for Jakarta). CPS-Core can be used to store any data and 'cache' it but currently only support json payload for data
Dell would like CPS to also support XML payload on CPS-E-05, CPS-E-02 and CPS-E-03 interfaces (probably does not apply to CPS-E-01 (TBC)
CPS-E-02 could easily be made to 'accept' XML format too
CPS-E-03 would required an additional parameter? to request XML output
DMI interface might be affect to, needs further investigation
DB storage is de-coupled from REST input format (interfaces are using java map objects instead) Postgress solution uses json format internally but this is not related to/affected by payload format
Conclusion; CPS-NCMP can scale horizontally without code changes
Lee Anjella Macabuhay will add some information on how to increase the DB connection pool size for 'vertical scaling of a singel instance. as it was observed the pool size became a bottleneck when more then 5 simultaneous requests were made.
as part of discussion future possibilities for domain based (manual) setup of groups of CPS instances were discussed. Sharding could be used to distribute cm-handles.
kieran mccarthy confirm spike goal was achieved and wil contact Peter Turcsany for follow up load testing.
Michael Pruss (Dell) explained his team is interested in using CPS but with support for XML data. He offered that his team can contribute to CPS to realize this in Jakarta. We agrred to start mail exchange next week and probably a first meeting in januari to understand the requirements more clearly...
Holiday Schedule: no more weekly meeting until Jan 7th!
Team agreed to NOT submit (backward incompatible) DMI-plugin updates until CPS-NCMP changes are approved too. This mean CSIT test can only be ran successful locally until the first merge.
Delete List / List Element / DataNode discussion with Dylan Byrne
Team agreed to use the same refactored code to delete List(element) and/or Container. For backward compatibility the rest endpoint that deletes lists (element) only it will thrown an exception if the target is a normal container
Renu Kumari pointed out that larger file used for Yang Data (but less frequent) Luke will add to study
Team confirmed preference for Jackson parser mainly because of GSON numeric issue and performance impact not significant enough to mix both throughout our code.
Passthrough 'root' support. Rahul Tyagi documented his findings in
Jira
server
ONAP Jira
serverId
425b2b0a-557c-3c0c-b515-579789cceedb
key
CPS-745
. This will be taken as input for the root-function support items planned for Jakarta. While Toine Siebelink wil start conversation with SDN-C team (PTL Dan Timoney ) about possible fixes in SDN-C and/or ODL. Tony Finnerty suggested a future proof workaround where the system will only try the restconf interface on SDN-C if the original request using the rests interface failed and the resource identifier equates to root.
that an easy work-around is available. The ensuing discussing made clear that the stakeholders might need to consider a cm-handle sync-state & watchdog in NCMP to ensure model and dat sync of registered nodes (with attributes to enable/disable this functionality)
Renu Kumari wil investigate option. Stakeholder want clear separation between 'client' and 'internal' interfaces. Although both need to be available for publication, just different target audiences. Preferred Internal interface pre-fix is /ncmpDmi
Discuss adding CM-Handle for SDN-R (network Slicing) with Ahila P
Toine Siebelink clarified that instead of Async flow steps 2&3 a 'temporary' sync option is available right now on DMI-Plugin. Toine wil mail specification links to Ahila
Steps 4 & 5 are also already developed and test in Jakarta using SDN-C. These might need to be tested in SND-R environment.
CPS teams prefers NOT to have a third option for adding CM Handles. Instead we agreed on the following possible evolution path:
SDN-R use synchronous method delivered in release I, mentioned above (can be done now)
SDN-R publishes event of the same schema and same topic as PNF. This release CPS is delivering the relevant event based API in J which has already been investigated , see
Jira
server
ONAP Jira
serverId
425b2b0a-557c-3c0c-b515-579789cceedb
key
CPS-393
eventually, post Jakarta, both SDN-R wil use the current A&AI event based interfaces
At any stage Topic and or Event bus provide could be made configurable as long as te event schema stays the same
Ahila P would like to discuss a proposal from E2E Slicing regarding the registration of SDN-R handled nodes with DMI possibly using a (new) dMaaP Topic. She will schedule a meeting to follow this up.
team agreed not to cerate I branch until we need to deliver Jakarta specific functionality or affect stability i.e. interface split
CSIT Improvements
aditya puthuparambil started discussion re. CSIT testing with Morgan Richomme to include in CPS CSIT in OOM build. Not yet clear how exactly this will work (in Jakarta) Aditya will follow up
Team wil continue to deliver CSIT test as planned for Istanbul using current process/framework
passthrough-running read uses-case are completed bar a demo (incl. config/non-config filter options)
Toine Siebelink will create new user stories for passtrough-running write uses case: Replace, Delete & Patch. On story for each component NCMP / DMI Plugin
aditya puthuparambil CPS-847 no need for 3 variations of operations in CSIT test. Issue with deployment of CPS_Temporal related to bguf Bruno Sakoto is working on. Aditya and Bruno will work together to resolve. Team test strategy needs to consider 'component/black-box testing' Toine Siebelink will call meeting to follow that up.
Exact structure of response, team prefer format with status for each node so successful updates could (later) be included too. Toine Siebelink wil check with stakeholders
Renu Kumari observed logging management require no authorization but could impacts system! Only accessible from inside cluster which is good enough for now.
The logging demonstrated is only to be turned on for specific functional tets to determine time spent during functional execution. Load test wil be affected too much by this.
Logging (configuration) guidelines will be documented as part of
Jira
server
ONAP Jira
serverId
425b2b0a-557c-3c0c-b515-579789cceedb
key
CPS-604
The logging introduced here probably needs re-start of service as it is enable using an annotation. We'll await client-feedback to see if this suffices.
'restconf' responses are somewhat smaller then equivalent 'rests' responses. This needs some further investigation. CSIT test using the new version of SDN-C will determine if there are significant issues but this is not expected
Some small changes were seen comparing legacy (SDNC 1.8) output with current output on the 'rests' interface. But these could be related to small model and data instance updates that have been part of the new SDN-C Deployment
Follow up E2E Slicing use-case impacts with Niranjana Y (now limited to
CPS team will work on datastores:operational cache during Jakarta
Toine Siebelink has added a new work item (#15 on R10 release plan) to allow update of cache when SDN-R updates corresponding device data through a messaging based solution)
E2E Slicing team will continue to write using SND-R during Jakarta and work with CSP team to fully cached solution in next release (Kohn)
CPS team wil meet again with Ahila P to follow up and clarify Jakarta scope
This will have some impacts on TBDMT process of data.
suggested to update template to handle required transformation to original data format (get child from parent)0 CPS Team can help out with these impacts to speed up changes. Ahila P will consider and revert to Toine Siebelink and Luke Gleeson
For 'passtrough' (bypassing cache) CPS-E-05 can already support XML (using JSON wrapping)
Cache not yet enabled in NCMP (planned for Jakarta). CPS-Core can be used to store any data and 'cache' it but currently only support json payload for data
Dell would like CPS to also support XML payload on CPS-E-05, CPS-E-02 and CPS-E-03 interfaces (probably does not apply to CPS-E-01 (TBC)
CPS-E-02 could easily be made to 'accept' XML format too
CPS-E-03 would required an additional parameter? to request XML output
DMI interface might be affect to, needs further investigation
DB storage is de-coupled from REST input format (interfaces are using java map objects instead) Postgress solution uses json format internally but this is not related to/affected by payload format
Conclusion; CPS-NCMP can scale horizontally without code changes
Lee Anjella Macabuhay will add some information on how to increase the DB connection pool size for 'vertical scaling of a singel instance. as it was observed the pool size became a bottleneck when more then 5 simultaneous requests were made.
as part of discussion future possibilities for domain based (manual) setup of groups of CPS instances were discussed. Sharding could be used to distribute cm-handles.
kieran mccarthy confirm spike goal was achieved and wil contact Peter Turcsany for follow up load testing.
Michael Pruss (Dell) explained his team is interested in using CPS but with support for XML data. He offered that his team can contribute to CPS to realize this in Jakarta. We agrred to start mail exchange next week and probably a first meeting in januari to understand the requirements more clearly...
Holiday Schedule: no more weekly meeting until Jan 7th!
Team agreed to NOT submit (backward incompatible) DMI-plugin updates until CPS-NCMP changes are approved too. This mean CSIT test can only be ran successful locally until the first merge.
Delete List / List Element / DataNode discussion with Dylan Byrne
Team agreed to use the same refactored code to delete List(element) and/or Container. For backward compatibility the rest endpoint that deletes lists (element) only it will thrown an exception if the target is a normal container
Renu Kumari pointed out that larger file used for Yang Data (but less frequent) Luke will add to study
Team confirmed preference for Jackson parser mainly because of GSON numeric issue and performance impact not significant enough to mix both throughout our code.
Passthrough 'root' support. Rahul Tyagi documented his findings in
Jira
server
ONAP Jira
serverId
425b2b0a-557c-3c0c-b515-579789cceedb
key
CPS-745
. This will be taken as input for the root-function support items planned for Jakarta. While Toine Siebelink wil start conversation with SDN-C team (PTL Dan Timoney ) about possible fixes in SDN-C and/or ODL. Tony Finnerty suggested a future proof workaround where the system will only try the restconf interface on SDN-C if the original request using the rests interface failed and the resource identifier equates to root.
that an easy work-around is available. The ensuing discussing made clear that the stakeholders might need to consider a cm-handle sync-state & watchdog in NCMP to ensure model and dat sync of registered nodes (with attributes to enable/disable this functionality)
Renu Kumari wil investigate option. Stakeholder want clear separation between 'client' and 'internal' interfaces. Although both need to be available for publication, just different target audiences. Preferred Internal interface pre-fix is /ncmpDmi
Discuss adding CM-Handle for SDN-R (network Slicing) with Ahila P
Toine Siebelink clarified that instead of Async flow steps 2&3 a 'temporary' sync option is available right now on DMI-Plugin. Toine wil mail specification links to Ahila
Steps 4 & 5 are also already developed and test in Jakarta using SDN-C. These might need to be tested in SND-R environment.
CPS teams prefers NOT to have a third option for adding CM Handles. Instead we agreed on the following possible evolution path:
SDN-R use synchronous method delivered in release I, mentioned above (can be done now)
SDN-R publishes event of the same schema and same topic as PNF. This release CPS is delivering the relevant event based API in J which has already been investigated , see
Jira
server
ONAP Jira
serverId
425b2b0a-557c-3c0c-b515-579789cceedb
key
CPS-393
eventually, post Jakarta, both SDN-R wil use the current A&AI event based interfaces
At any stage Topic and or Event bus provide could be made configurable as long as te event schema stays the same
Ahila P would like to discuss a proposal from E2E Slicing regarding the registration of SDN-R handled nodes with DMI possibly using a (new) dMaaP Topic. She will schedule a meeting to follow this up.
team agreed not to cerate I branch until we need to deliver Jakarta specific functionality or affect stability i.e. interface split
CSIT Improvements
aditya puthuparambil started discussion re. CSIT testing with Morgan Richomme to include in CPS CSIT in OOM build. Not yet clear how exactly this will work (in Jakarta) Aditya will follow up
Team wil continue to deliver CSIT test as planned for Istanbul using current process/framework
passthrough-running read uses-case are completed bar a demo (incl. config/non-config filter options)
Toine Siebelink will create new user stories for passtrough-running write uses case: Replace, Delete & Patch. On story for each component NCMP / DMI Plugin
