Blob client 'from_string' method regex does not allow bucket root

[mcompen]

Environment details

• All OS types and Python versions

Steps to reproduce

Create a Blob client with the from_string method at bucket root.

Code example

from google.cloud import storage

project = "project"
credentials = "credentials"
storage_client = storage.Client(project=project, credentials=credentials)
storage.Blob.from_string(
        uri="gs://bucket_name",
        client=storage_client,
    )

Stack trace

"/Users/manu.compen/Library/Caches/pypoetry/virtualenvs/bla-py3.9/lib/python3.9/site-packages/google/cloud/aiplatform/pipeline_jobs.py", line 404, in submit
    gcs_utils.create_gcs_bucket_for_pipeline_artifacts_if_it_does_not_exist(
  File "/Users/manu.compen/Library/Caches/pypoetry/virtualenvs/bla-py3.9/lib/python3.9/site-packages/google/cloud/aiplatform/utils/gcs_utils.py", line 237, in create_gcs_bucket_for_pipeline_artifacts_if_it_does_not_exist
    pipelines_bucket = storage.Blob.from_string(
  File "/Users/manu.compen/Library/Caches/pypoetry/virtualenvs/bla-py3.9/lib/python3.9/site-packages/google/cloud/storage/blob.py", line 410, in from_string
    raise ValueError("URI scheme must be gs")
ValueError: URI scheme must be gs

I came across this (as you can see in the stacktrace) using the aiplatform package to submit a Vertex Pipeline job where you have to specify a bucket root, but I think the bug is on this side of the fence.

[m-rath]

Same problem here. Using aiplatform.init(staging_bucket="gs://my_staging_bucket").
Or just storage.Blob.from_string("gs://my_staging_bucket"). Same Value error. Not sure how to circumvent. Thanks

[cojenco]

Thanks for filing. Blob.from_string is intened to get a constructor for blob object with a blob URI. The blob URI should include an object name such as gs://bucket_name/object_name

For a bucket root, it'd make more sense to use the Bucket class method Bucket.from_string. I believe this is a bug in the aiplatform where there is a misusage of Blob.from_string here. Changing to use Bucket.from_string should fix the issue. I'll go ahead and file an issue in the aiplatform repository.

[cojenco]

I see that googleapis/python-aiplatform#2936 is already filed along with a fix googleapis/python-aiplatform#2948
Thank you @mcompen

[mcompen]

Closing this as the issue is with the aiplatform package and I opened an issue there.

[Ark-kun]

Blob.from_string is intened to get a constructor for blob object with a blob URI. The blob URI should include an object name such as gs://bucket_name/object_name

What is the prescribed way to get a Bucket object for a given gs:// URI that may or may not contain path?

[yan-hic]

What is the prescribed way to get a Bucket object for a given gs:// URI that may or may not contain path?

See here - the same method exists for Bucket