#acumos-meeting: Architecture Committee

Meeting summary

1. Marting Welss Ai4EU C++ Onboarding architecture (farheen_cefalu, 14:24:07)
   1. https://wiki.acumos.org/display/MOB/Acumos-c-client+design (farheen_cefalu, 14:24:35)
   2. zoom meeting is being recorded for the proposal. (farheen_cefalu, 14:25:08)
   3. Introspection? Yes, One api for R Python java and another api for ONNX and PFX. Normally we look at the meta data.json for toolkit type. Based on that we do the onboarding. As long as the meta data is provided by the client there would be a code change but a UI changes. It will have to showcased in the UI that will be needed. (farheen_cefalu, 14:35:37)
   4. we need a new client to do the packaging. Martin - it is quicker to implement using python than c or c++. (farheen_cefalu, 14:37:00)
   5. Manoop - We do have some aspects of exposing Swagger apis from a python model. This will be from the model runner part to support it. We do support it. There is something we can keep in mind and not implement right away. When we deploy a model it automatically comes with a swagger ui (services, test inputs, data types). So it takes a run (farheen_cefalu, 14:40:47)
   6. Manoop - Do you have any blockers related to C++. Phillippe - No blockers for the moment. Guy - from onboarding perspective we have a plan for that. The people we have are sufficient. We are not going to be able to do type inference. We don't need to do data serialization. The client packaging code in Python is simpler here where you do (farheen_cefalu, 14:43:41)
   
   
2. Deployment Parichay Gupta (TechM) (farheen_cefalu, 14:44:20)
   1. Template approach with the jenkins service. We have daily calls with the CMLP team. The plan to hand this over to deployment. Check the wiki OA&M /deployment wikis. (farheen_cefalu, 14:45:43)
   2. Parichay is going to deploy in a stand alone vm. A service exposed to the K8 VM. A new component deployment client. Deploy to a K8 target. That will use java Apis. We're going to use a seed (jenkins job template from CMLP). By the end of sprint 1 we're going to have a jenkins server, a java client and a demo model to deploy to k8. That (farheen_cefalu, 14:47:50)
   3. Parichay is concerned about onboarding models and deployments. (farheen_cefalu, 14:49:47)
   4. https://wiki.acumos.org/pages/viewpage.action?pageId=20547265 (farheen_cefalu, 14:49:51)
   
   
3. next meeting agenda (farheen_cefalu, 14:59:24)

Action items

1. (none)

People present (lines said)

1. farheen_cefalu (24)
2. collabot` (3)