Background:

Analytics application can contain multiple components - Some components are collection services, some are action dispatchers and some are spark applications.  Collection services and action dispatcher services are containers and hence dockerhub (public or private) can be used to store images. In case of spark components, the image is typically a tar file or war file.  At run time, this is expected to be submitted to the spark framework. Hence, we need spark application image management.  Image management consists of two items - Management service at the ONAP-Central and image synchronizer in various cloud-regions (part of framework) to synchronize images to local file system.

Same is the case with ML/DL models. These models would be uploaded and used later on by inferencing applications. Hence, ML/DL model management is also required.

Tasks:

Creation of application image management service (that sits in ONAP-Central)

• Define APIs for application image management (CRUD operations)
  • Create/Delete/Modify/Get application image record (with application image name, description, version and other metadata)
  • Upload/Delete/Get application image to the application image record.
  • Add cloud-region to record (Multiple operations to add multiple cloud regions).  Get/Delete operations for completeness.
• Implement APIs
  • Store information in DB (Mango DB as being used by other projects)
• Synchronization:
  • Notify cloud-regions synchronizer (that are known to this module as part of adding cloud region to application image records).
• Helm charts to bring this service when OOM brings DCAE

Creation of application image syncrhonizer (This service sits in the analytics framework that sits in the cloud-region and hence this should be made as one of the components of analytics framework).

• API
  • POST notification (This notification URL is used by application image service) to notify whenever there is a configuration change.
• Processing 
  • Syncrhonize by making set of GET calls to the application image service in ONAP-Central.
  • If appliation image record is not present, then it issues GET and download the image.
  • If the version is different, it deletes the local copy and gets latest one.
  • Internally can use PNDA deployment manager to upload the image to HDFS and other storage mechanisms (But study if it can be avoided by taking just the logic, not the entire micro service)

Above explanation is valid for ML/DL images too.

• Have ML/DL model management service (in ONAP-Central)
• Have a ML/DL model synchronizer service (part of analytics framework)