issue with google_cloud_pipeline_components 0.1.9 ModelUploadOp ("invalid value")

[amygdala]

with the v0.1.9 of ModelUploadOp I'm seeing the following error. I confirmed that things work fine in 0.1.7.

PORT is set as follows:

PORT = 8080

Then, in the pipeline definition, ModelUploadOp is configured as follows:

    model_upload_op = gcc_aip.ModelUploadOp(
        project=project,
        display_name=model_display_name,
        serving_container_image_uri=build_image_task.outputs['serving_container_uri'],
        serving_container_predict_route="/predictions/{}".format(MAR_MODEL_NAME),
        serving_container_health_route="/ping",
        serving_container_ports=[PORT]

This is the error I see (again, I do NOT see this problem with v0.1.7):

2021-11-01 13:01:17.976 PDTRuntimeError: Failed to create the resource. Error: {'code': 400, 'message': "Invalid value at 'model.container_spec.ports[0]' (type.googleapis.com/google.cloud.aiplatform.v1.Port), 8080", 'status': 'INVALID_ARGUMENT', 'details': [{'@type': 'type.googleapis.com/google.rpc.BadRequest', 'fieldViolations': [{'field': 'model.container_spec.ports[0]', 'description': "Invalid value at 'model.container_spec.ports[0]' (type.googleapis.com/google.cloud.aiplatform.v1.Port), 8080"}]}]}      

I wondered if something has changed about the expected format of this arg, but from the documentation (serving_container_ports (Optional[Sequence[int]]=None) it doesn't seem so. Again, this code runs fine with 0.1.7.

[amygdala]

Here's what the set of input args looks like in the console: https://screenshot.googleplex.com/HdYMJLiz8oVhxjB
/cc @SinaChavoshi @IronPan

[jagadeeshi2i]

@amygdala try configuring port as below - serving_container_ports=[{"containerPort" : PORT}]. The component follows k8s v1/core spec. Worked for me with above config.

[amygdala]

Interesting -- the api spec says: serving_container_ports (Optional[Sequence[int]]=None): but maybe the docs need to be updated.

[LeonardoEssence]

We spent hours trying to find an answer to this. We need better alignment between the documentation in here https://google-cloud-pipeline-components.readthedocs.io/en/google-cloud-pipeline-components-0.2.0/google_cloud_pipeline_components.aiplatform.html and here https://cloud.google.com/vertex-ai/docs/predictions/use-custom-container

[andodet]

Stumbled upon this issue after more time I'd like to publicly admit. Passing the port argument as @jagadeeshi2i mentioned did the trick for me.

I am now struggling to set serving_container_environment_variables as I think it might be suffering the same problem. I am currently passing env vars to the container in the following way (docs lists it as a Optional[Dict[str, str]]):

model_upload_op = gcc_aip.ModelUploadOp(
    project="and-reporting",
    location="us-west1",
    display_name="session_model",
    serving_container_image_uri="gcr.io/and-reporting/pred:latest",
    # The following is creating troubles...
    serving_container_environment_variables={"MODEL_BUCKET": "ml_session_model/model},
    serving_container_ports=[{"containerPort": 5000}],
    serving_container_predict_route="/predict",
    serving_container_health_route="/health",
)

Which produces the following error:

RuntimeError: Failed to create the resource. Error: {'code': 400, 'message': 'Invalid JSON payload received. Unknown name "MODEL_BUCKET" at \'model.container_spec.env[0]\': Cannot find field.', 'status': 'INVALID_ARGUMENT', 'details': [{'@type': 'type.googleapis.com/google.rpc.BadRequest', 'fieldViolations': [{'field': 'model.container_spec.env[0]', 'description': 'Invalid JSON payload received. Unknown name "MODEL_BUCKET" at \'model.container_spec.env[0]\': Cannot find field.'}]}]}

I think ModelUploadOp might have been left out from the refresh in #5481.

Solution

Can confirm serving_container_environment_variables documentation is outdated. Setting it accordingly to the kubernetes docs solved it for me:

serving_container_environment_variables=[
    {"name": "MODEL_BUCKET", "value": "ml_session_model/model"}
],