Python BigQuery allowLargeResults with pandas.io.gbq

Question

I want to use the Pandas library to read BigQuery data. How do I allow large results?
For non-Pandas BigQuery interactions, this can be achieved like this.

Curr

Answer 1

EDIT: I've posted the proper way to do this with in my other answer; by dropping off the data in google storage first. This way you'll never have data that is too large.

---

Ok, I didn't find a direct way to do it with pandas, so I had to write a little extra with the normal API. Here is my fix (also most of the work to do it natively without Pandas):

sProjectID = "project-id"
sQuery = '''
    SELECT 
        column1, column2
    FROM [dataset_name.tablename]
'''

df = create_dataframe(sQuery, sProjectID, bLargeResults=True)


#*******Functions to make above work*********



def create_dataframe(sQuery, sProjectID, bLargeResults=False):
    "takes a BigQuery sql query and returns a Pandas dataframe"

    if bLargeResults:
        oService = create_service()
        dDestinationTable = run_query(sQuery, oService, sProjectID)
        df = pandas_get_table(dDestinationTable)
    else:
        df = pandas_query(sQuery, sProjectID)

    return df



def pandas_query(sQuery, sProjectID):
    "go into bigquery and get the table with sql query and return dataframe"
    from pandas.io import gbq
    df = gbq.read_gbq(sQuery, sProjectID)

    return df 



def pandas_get_table(dTable):
    "fetch a table and return dataframe"
    from pandas.io import gbq

    sProjectID = dTable['projectId']
    sDatasetID = dTable['datasetId']
    sTableID = dTable['tableId']
    sQuery = "SELECT * FROM [{}.{}]".format(sDatasetID, sTableID)

    df = gbq.read_gbq(sQuery, sProjectID)

    return df 




def create_service():
    "create google service"
    from oauth2client.client import GoogleCredentials
    from apiclient.discovery import build
    credentials = GoogleCredentials.get_application_default()
    oService = build('bigquery', 'v2', credentials=credentials)
    return oService



def run_query(sQuery, oService, sProjectID):
    "runs the bigquery query"

    dQuery = {
        'configuration': {
            'query': {
                'writeDisposition': 'OVERWRITE',
                'useQueryCache': False,
                'allowLargeResults': True,
                'query': sQuery,
                'destinationTable': {
                    'projectId': sProjectID,
                    'datasetId': 'sandbox',
                    'tableId': 'api_large_result_dropoff',
                },
            }
        }
    }

    job = oService.jobs().insert(projectId=sProjectID, body=dQuery).execute()


    return job['configuration']['query']['destinationTable']

Answer 2

Decided to post the proper way to do this via the python3 google.cloud API. Looking at my previous answer I see that it would fail like yosemite_k said.

Large results really need to follow BigQuery -> Storage -> local -> dataframe pattern.

BigQuery resources:

• https://cloud.google.com/bigquery/docs/reference/libraries
• https://googlecloudplatform.github.io/google-cloud-python/stable/bigquery-client.html
• http://google-cloud-python.readthedocs.io/en/latest/bigquery-usage.html

Storage resources:

• https://googlecloudplatform.github.io/google-cloud-python/stable/storage-client.html
• https://googlecloudplatform.github.io/google-cloud-python/stable/storage-blobs.html

Pandas Resources:

• http://pandas.pydata.org/pandas-docs/stable/generated/pandas.read_csv.html

Installation:

pip install pandas
pip install google-cloud-storage
pip install google-cloud-bigquery

Full implementation (bigquery_to_dataframe.py):

"""
We require python 3 for the google cloud python API
    mkvirtualenv --python `which python3` env3
And our dependencies:
    pip install pandas
    pip install google-cloud-bigquery
    pip install google-cloud-storage
"""
import os
import time
import uuid

from google.cloud import bigquery
from google.cloud import storage
import pandas as pd


def bq_to_df(project_id, dataset_id, table_id, storage_uri, local_data_path):
    """Pipeline to get data from BigQuery into a local pandas dataframe.

    :param project_id: Google project ID we are working in.
    :type project_id: str
    :param dataset_id: BigQuery dataset id.
    :type dataset_id: str
    :param table_id: BigQuery table id.
    :type table_id: str
    :param storage_uri: Google Storage uri where data gets dropped off.
    :type storage_uri: str
    :param local_data_path: Path where data should end up.
    :type local_data_path: str
    :return: Pandas dataframe from BigQuery table.
    :rtype: pd.DataFrame
    """
    bq_to_storage(project_id, dataset_id, table_id, storage_uri)

    storage_to_local(project_id, storage_uri, local_data_path)

    data_dir = os.path.join(local_data_path, "test_data")
    df = local_to_df(data_dir)

    return df


def bq_to_storage(project_id, dataset_id, table_id, target_uri):
    """Export a BigQuery table to Google Storage.

    :param project_id: Google project ID we are working in.
    :type project_id: str
    :param dataset_id: BigQuery dataset name where source data resides.
    :type dataset_id: str
    :param table_id: BigQuery table name where source data resides.
    :type table_id: str
    :param target_uri: Google Storage location where table gets saved.
    :type target_uri: str
    :return: The random ID generated to identify the job.
    :rtype: str
    """
    client = bigquery.Client(project=project_id)

    dataset = client.dataset(dataset_name=dataset_id)
    table = dataset.table(name=table_id)

    job = client.extract_table_to_storage(
        str(uuid.uuid4()),  # id we assign to be the job name
        table,
        target_uri
    )
    job.destination_format = 'CSV'
    job.write_disposition = 'WRITE_TRUNCATE'

    job.begin()  # async execution

    if job.errors:
        print(job.errors)

    while job.state != 'DONE':
        time.sleep(5)
        print("exporting '{}.{}' to '{}':  {}".format(
            dataset_id, table_id, target_uri, job.state
        ))
        job.reload()

    print(job.state)

    return job.name


def storage_to_local(project_id, source_uri, target_dir):
    """Save a file or folder from google storage to a local directory.

    :param project_id: Google project ID we are working in.
    :type project_id: str
    :param source_uri: Google Storage location where file comes form.
    :type source_uri: str
    :param target_dir: Local file location where files are to be stored.
    :type target_dir: str
    :return: None
    :rtype: None
    """
    client = storage.Client(project=project_id)

    bucket_name = source_uri.split("gs://")[1].split("/")[0]
    file_path = "/".join(source_uri.split("gs://")[1].split("/")[1::])
    bucket = client.lookup_bucket(bucket_name)

    folder_name = "/".join(file_path.split("/")[0:-1]) + "/"
    blobs = [o for o in bucket.list_blobs() if o.name.startswith(folder_name)]

    # get files if we wanted just files
    blob_name = file_path.split("/")[-1]
    if blob_name != "*":
        print("Getting just the file '{}'".format(file_path))
        our_blobs = [o for o in blobs if o.name.endswith(blob_name)]
    else:
        print("Getting all files in '{}'".format(folder_name))
        our_blobs = blobs

    print([o.name for o in our_blobs])

    for blob in our_blobs:
        filename = os.path.join(target_dir, blob.name)

        # create a complex folder structure if necessary
        if not os.path.isdir(os.path.dirname(filename)):
            os.makedirs(os.path.dirname(filename))

        with open(filename, 'wb') as f:
            blob.download_to_file(f)


def local_to_df(data_path):
    """Import local data files into a single pandas dataframe.

    :param data_path: File or folder path where csv data are located.
    :type data_path: str
    :return: Pandas dataframe containing data from data_path.
    :rtype: pd.DataFrame
    """
    # if data_dir is a file, then just load it into pandas
    if os.path.isfile(data_path):
        print("Loading '{}' into a dataframe".format(data_path))
        df = pd.read_csv(data_path, header=1)
    elif os.path.isdir(data_path):
        files = [os.path.join(data_path, fi) for fi in os.listdir(data_path)]
        print("Loading {} into a single dataframe".format(files))
        df = pd.concat((pd.read_csv(s) for s in files))
    else:
        raise ValueError(
            "Please enter a valid path.  {} does not exist.".format(data_path)
        )

    return df


if __name__ == '__main__':
    PROJECT_ID = "my-project"
    DATASET_ID = "bq_dataset"
    TABLE_ID = "bq_table"
    STORAGE_URI = "gs://my-bucket/path/for/dropoff/*"
    LOCAL_DATA_PATH = "/path/to/save/"

    bq_to_df(PROJECT_ID, DATASET_ID, TABLE_ID, STORAGE_URI, LOCAL_DATA_PATH)

Answer 3

You can do it by changing the default dialect from legacy to standard in pd.read_gbq function.

pd.read_gbq(query, 'my-super-project', dialect='standard')

Indeed, you can read in Big Query documentation for the parameter AllowLargeResults:

AllowLargeResults: For standard SQL queries, this flag is ignored and large results are always allowed.