Fetch results from BigQueryOperator in airflow

问题

I am trying to fetch results from BigQueryOperator using airflow but I could not find a way to do it. I tried calling the next() method in the bq_cursor member (available in 1.10) however it returns None. This is how I tried to do it

import datetime
import logging

from airflow import models
from airflow.contrib.operators import bigquery_operator
from airflow.operators import python_operator


yesterday = datetime.datetime.combine(
    datetime.datetime.today() - datetime.timedelta(1),
    datetime.datetime.min.time()
)

def MyChequer(**kwargs):
    big_query_count = bigquery_operator.BigQueryOperator(
        task_id='my_bq_query',
        sql='select count(*) from mydataset.mytable'
    )

    big_query_count.execute(context=kwargs)

    logging.info(big_query_count)
    logging.info(big_query_count.__dict__)
    logging.info(big_query_count.bq_cursor.next())

default_dag_args = {
    'start_date': yesterday,
    'email_on_failure': False,
    'email_on_retry': False,
    'project_id': 'myproject'
}

with models.DAG(
        'bigquery_results_execution',
        # Continue to run DAG once per day
        schedule_interval=datetime.timedelta(days=1),
        default_args=default_dag_args) as dag:

    myoperator = python_operator.PythonOperator(
        task_id='threshold_operator',
        provide_context=True,
        python_callable=MyChequer
    )

    # Define DAG
    myoperator

Taking a look to bigquery_hook.py and bigquery_operator.py it seems to be the only available way to fetch the results.

回答1:

I create my own operator using the BigQuery hook whenever I need to get the data from a BigQuery query and use it for something.I usually call this a BigQueryToXOperator and we have a bunch of these for sending BigQuery data to other internal systems.

For example, I have a BigQueryToPubSub operator that you might find useful as an example for how to query BigQuery and then handle the results on a row by row basis, sending them to Google PubSub. Consider the following generalized sample code for how to do this on your own:

class BigQueryToXOperator(BaseOperator):
    template_fields = ['sql']
    ui_color = '#000000'

    @apply_defaults
    def __init__(
            self,
            sql,
            keys,
            bigquery_conn_id='bigquery_default',
            delegate_to=None,
            *args,
            **kwargs):
        super(BigQueryToXOperator, self).__init__(*args, **kwargs)
        self.sql = sql
        self.keys = keys # A list of keys for the columns in the result set of sql
        self.bigquery_conn_id = bigquery_conn_id
        self.delegate_to = delegate_to


    def execute(self, context):
        """
        Run query and handle results row by row.
        """
        cursor = self._query_bigquery()
        for row in cursor.fetchall():
            # Zip keys and row together because the cursor returns a list of list (not list of dicts)
            row_dict = dumps(dict(zip(self.keys,row))).encode('utf-8')

            # Do what you want with the row...
            handle_row(row_dict)


    def _query_bigquery(self):
        """
        Queries BigQuery and returns a cursor to the results.
        """
        bq = BigQueryHook(bigquery_conn_id=self.bigquery_conn_id,
                          use_legacy_sql=False)
        conn = bq.get_conn()
        cursor = conn.cursor()
        cursor.execute(self.sql)
        return cursor

回答2:

You can use BigQueryOperator to save results in a temporary destination table and then use BigQueryGetDataOperator to fetch the results as below and then use BigQueryTableDeleteOperator to delete the table:

get_data = BigQueryGetDataOperator(
    task_id='get_data_from_bq',
    dataset_id='test_dataset',
    table_id='Transaction_partitions',
    max_results='100',
    selected_fields='DATE',
    bigquery_conn_id='airflow-service-account'
)

Docs:

BigQueryGetDataOperator: https://airflow.readthedocs.io/en/1.10.0/integration.html#bigquerygetdataoperator
BigQueryTableDeleteOperator: https://airflow.readthedocs.io/en/1.10.0/integration.html#bigquerytabledeleteoperator

回答3:

Thanks to @kaxil and @Mike for their answers. I found the problem. There is a kind of bug (in my mind) in the BigQueryCursor. As part of the run_with_configuration, the running_job_id is being returned but never assigned to job_id which is used to pull the results in the next method. A workaround (not really elegant but good if you do not want to re-implement everything), is assign the job_id based on the running_job_id in the hook like this

big_query_count.execute(context=kwargs)
#workaround
big_query_count.bq_cursor.job_id = big_query_count.bq_cursor.running_job_id
logging.info(big_query_count.bq_cursor.next())

One the problem get fixed on the run_with_configuration assigning the correct job_id at the end of the process, the row after workaround can be removed

来源：https://stackoverflow.com/questions/53565834/fetch-results-from-bigqueryoperator-in-airflow

标签

python

google-bigquery

airflow