Deploying custom container with arguments pointing at cloud storage location (file in the bucket)

[gcpbegginer]

Dear all,

I would like to deploy a custom container with arguments that are basically specific file locations in my bucket on cloud storage. I am talking about the filed below. When starting the custom training job the path format accepted therein looks like this --myfile_location_folder=/gcs/mybucket/input/.

Is it the same here? Each time I want to pass an argument pointing at a file in a bucket I get an error that the location does not exists. Once the files are within the container all works beautiful though. I just can't figure out how to make that container see the given bucket

I am aware that this might be also a service account issue and I tried different settings. However I still suspect that I might be providing the wrong path format. .Should it be this /gcs/mybucket/input/.,  gs://mybucket/input/ or maybe other?

And about the service accounts. This is also very confusing. What exact privileges should the endpoint service account have in order to make buckets accessible to custom prediction container  ?

Thank you all!

[lsolatorio]

Hi @gcpbegginer, 

Welcome and appreciate you reaching out to our community.

I get that you are having challenges with your storage bucket and service account permission.

Please check the following as it might be helpful in your situation.

• Make sure that the bucket and file are existing.
• The path should be in the format gs://<bucket-name>/<file-path>.
• Make sure that the container has permission to access the bucket and file.
• Run the container in the same region as the bucket.

Here are some other helpful resources for your reference.

• Custom container requirements for prediction
• Vertex AI IAM permissions
• Roles for service account authentication