Replacement of pipeline_run.publish_pipeline() in SDK V2

Question

In SDK V1 we are able to publish Azure ML Pipeline through pipeline_run.publish_pipeline(). This pipeline ID is then allowed to use in ML Ops build pipeline to automate the build & then release. Our existing architecture is based on the same concept, where when a new file is published in bloblstorage by ADF pipeline, a logic App pipeline triggers the ML Ops build pipeline. Build pipeline triggers Azure ML Pipeline. Once Azure ML pipeline completes, the release pipeline gets triggered to deploy the model to production environment. I went through the documentation of V2, but couldn't find any replacement for publish_pipeline(). We can't enable CI/CD with Azure ML without publishing the ML pipeline. Please help me getting similar package name in V2 which will resolve this issue.

Answer 1

@Dash, Shishir With SDK v1 the pipelines can be published but with SDK v2 a pipeline component needs to be deployed or submitted.

The following table shows a comparison of each of the concepts from v1 to v2:

See this example notebook for using pipeline component.

See this page for details about pipeline component deployment(preview).

For CI/CD deployment scenarios see this page for intro of azure pipelines with CI/CD scenarios. With v2 I think there will be changes on how your current scenarios are handled and might need to be rebuilt scenario by scenario based on the changes of v2 architecture of Azure ML SDK.

I hope this helps!!

If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.