GCS - Python download blobs with directory structure

Below is the working solution. I ended up stripping away the path from the object name and creating the directory structure on the fly. A better way might be as @Brandon Yarbrough suggested using 'prefix + response['prefixes'][0]' but I couldn't quite figure that out. Hope this helps others out.

#!/usr/local/bin/python3

from google.cloud import storage
from googleapiclient import discovery
from oauth2client.client import GoogleCredentials
import json
import os
import pathlib

bucket_name = 'test-bucket'
restore_epoch = '1519189202'
restore_location = '/Users/admin/data/'

credentials = GoogleCredentials.get_application_default()
service = discovery.build('storage', 'v1', credentials=credentials)

storage_client = storage.Client()
source_bucket = storage_client.get_bucket(bucket_name)


def listall_objects():
    request = service.objects().list(
        bucket=bucket_name,
        versions=True
    )
    response = request.execute()
    print(json.dumps(response, indent=2))


def listname_objects():
    request = service.objects().list(
        bucket=bucket_name,
        versions=True
    )
    response = request.execute()

    for item in response['items']:
        print(item['name'] + ' Uploaded on: ' + item['updated'] +
              ' Epoch: ' + item['metadata']['epoch'])


def downloadepoch_objects():
    request = service.objects().list(
        bucket=bucket_name,
        versions=True
    )
    response = request.execute()

    try:
        for item in response['items']:
            if item['metadata']['epoch'] == restore_epoch:
                print('Downloading ' + item['name'] + ' from ' +
                      item['bucket'] + '; Epoch= ' + item['metadata']['epoch'])
                print('Saving to: ' + restore_location)
                blob = source_bucket.blob(item['name'])
                path = pathlib.Path(restore_location + r'{}'.format(item['name'])).parent
                if os.path.isdir(path):
                    blob.download_to_filename(restore_location + '{}'.format(item['name']))
                    print('Download complete')
                else:
                    os.mkdir(path)
                    blob.download_to_filename(restore_location + '{}'.format(item['name']))
                    print('Download complete')
    except Exception:
        pass


# listall_objects()
# listname_objects()
downloadepoch_objects()

GCS does not have a notion of "directories," although tools like gsutil do a good job of pretending for convenience. If you want all of the objects under the "nfs/media/docs/" path, you can specify that as a prefix, like so:

request = service.objects.list(
    bucket=bucket_name,
    versions=True,
    prefix='nfs/media/docs/',  # Only show objects beginning like this
    delimiter='/'  # Consider this character a directory marker.
)
response = request.execute()
subdirectories = response['prefixes']
objects = response['items']

Because of the prefix parameter, only objects that begin with 'nfs/media/docs' will be returned in response['items']. Because of the delimiter parameter, "subdirectories" will be returned in response['prefixes']. You can get more details in the Python documentation of the objects.list method.

If you were to use the newer google-cloud Python library, which I'd recommended for new code, the same call would look pretty similar:

from google.cloud import storage

client = storage.Client()
bucket = client.bucket(bucket_name)
iterator = bucket.list_blobs(
    versions=True,
    prefix='nfs/media/docs/',
    delimiter='/'
)
subdirectories = iterator.prefixes
objects = list(iterator)

Following solution worked for me. I am recursively downloading all blobs from a path prefix to a model directory at the project root, while maintaining the folder structure. Multiple blobs are being downloaded concurrently.

GCS client version google-cloud-storage==1.41.1

import os
from datetime import datetime
from google.cloud import storage
from concurrent.futures import ThreadPoolExecutor

BUCKET_NAME = "ml-model"

def timer(func):
    def time_wrapper(*arg, **kwargs):
        start = datetime.now()
        func(*arg, **kwargs)
        diff = datetime.now() - start
        logger.info(f"{func.__name__} took {diff.seconds} s and {diff.microseconds//1000} ms")
    return time_wrapper

def fetch_environment() -> str:
    env = os.environ.get("environment", "staging")
    return env


def create_custom_folder(dir_name: str):
    if not os.path.exists(dir_name):
        os.makedirs(dir_name)


def fetch_gcs_credential_file_path():
    return os.environ.get("GCS_CREDENTIAL_FILE_PATH")


class GCS:
    def __init__(self):
        cred_file_path = fetch_gcs_credential_file_path()
        self.client = storage.Client.from_service_account_json(cred_file_path)
        self.bucket = self.client.bucket(BUCKET_NAME)

    def download_blob(self, blob):
        filename = blob.name.replace(self.path_prefix, '')
        delimiter_based_splits = filename.split('/')
        if len(delimiter_based_splits) > 1:
            dir_name = "model/" + "/".join(delimiter_based_splits[: len(delimiter_based_splits)-1])
            create_custom_folder(dir_name)
            blob.download_to_filename(f"{dir_name}/{delimiter_based_splits[-1]}")
        else:
            blob.download_to_filename(f"model/" + filename)
    @timer
    def download_blobs_multithreaded(self, prefix: str):
        '''
        CREATE FOLDER IF NOT EXISTS
        '''
        create_custom_folder("model")

        blobs = self.bucket.list_blobs(prefix=prefix)

        self.path_prefix = prefix
        with ThreadPoolExecutor() as executor:
            executor.map(self.download_blob, blobs


def download_model():
    env = fetch_environment()
    folder_path_prefix = f"ml/{env}/{ML_MODEL_NAME}/v1/tf-saved-model/"
    gcs = GCS()
    gcs.download_blobs_multithreaded(folder_path_prefix)

if __name__ == '__main__':
    download_model()