How to reuse an existing Endpoint in pipeline?

[MMichael-S]

Hi Experts

Expected Behavior

Hope to reuse an existing endpoint in pipeline.

Also see that google_cloud_pipeline_components.v1.endpoint has only EndpointCreateOp method.
https://google-cloud-pipeline-components.readthedocs.io/en/google-cloud-pipeline-components-1.0.0/google_cloud_pipeline_components.v1.endpoint.html

Is there a way to reuse an existing Endpoint? And how to reuse it? This will saves costs.

Actual Behavior

Have to create a new Endpoint in the pipeline.
See that the EndpointCreateOp step is included in all the sample code.

 endpoint_op = EndpointCreateOp(
     project=project,
     location=region,
     display_name="train-automl-flowers",
 )

 ModelDeployOp(
     model=training_run_task.outputs["model"],
     endpoint=endpoint_op.outputs["endpoint"],
     automatic_resources_min_replica_count=1,
     automatic_resources_max_replica_count=1,
 )

Steps to Reproduce the Problem

N/A

Specifications

• Version:
  google-cloud-pipeline-components-1.0.0
• Platform:
  Google Cloud Vertex AI

[andrewferlitsch]

This does appear to be a limitation the prebuilt components. You could then write your own python component, something like:

@component()
def get_endpoint(resource_name: str,
endpoint: Output[Artifact]
)
endpoint = aiplatform.Endpoint(endpoint_name=resource_name)

[MMichael-S]

@andrewferlitsch
Thank you very much.
We have done in the same way by using customized components to import the existed endpoint.

Code as following, for someone who need the reference:

from google_cloud_pipeline_components.types import artifact_types
from kfp.v2.components import importer_node


endpoint = importer_node.importer(
    artifact_uri="https://us-central1-aiplatform.googleapis.com/v1/projects/xxxxxxxx/locations/us-central1/endpoints/xxxxxx",
    artifact_class=artifact_types.VertexEndpoint,
    metadata={
        "resourceName": "projects/xxxx/locations/us-central1/endpoints/xxxxx"
    }
)
...
ModelDeployOp(
        model=training_run_task.outputs["model"],
        endpoint=endpoint.outputs['artifact'],
        automatic_resources_min_replica_count=1,
        automatic_resources_max_replica_count=1,
    )

[andrewferlitsch]

Thanks @MMichael-S for posting example using the importer_node method