时间戳

首页 > 解决方案 > 使用 Apache Beam 以 CSV 格式将 BigQuery 结果写入 GCS

python - 使用 Apache Beam 以 CSV 格式将 BigQuery 结果写入 GCS

问题描述

我是 Apache Beam 的新手，我正在尝试编写一个管道来从 Google BigQuery 中提取数据并使用 Python 以 CSV 格式将数据写入 GCS。

使用beam.io.read(beam.io.BigQuerySource())我能够从 BigQuery 读取数据，但不确定如何以 CSV 格式将其写入 GCS。

是否有自定义功能可以实现相同的功能，您能帮帮我吗？

import logging

import apache_beam as beam
from apache_beam.io.BigQueryDisposition import CREATE_IF_NEEDED
from apache_beam.io.BigQueryDisposition import WRITE_TRUNCATE

PROJECT='project_id'
BUCKET='project_bucket'


def run():
    argv = [
        '--project={0}'.format(PROJECT),
        '--job_name=readwritebq',
        '--save_main_session',
        '--staging_location=gs://{0}/staging/'.format(BUCKET),
        '--temp_location=gs://{0}/staging/'.format(BUCKET),
        '--runner=DataflowRunner'
    ]

    with beam.Pipeline(argv=argv) as p:

        # Execute the SQL in big query and store the result data set into given Destination big query table.
        BQ_SQL_TO_TABLE = p | 'read_bq_view' >> beam.io.Read(
            beam.io.BigQuerySource(query =  'Select * from `dataset.table`', use_standard_sql=True))
        # Extract data from Bigquery to GCS in CSV format.
        # This is where I need your help

        BQ_SQL_TO_TABLE | 'Write_bq_table' >> beam.io.WriteToBigQuery(
                table='tablename',
                dataset='datasetname',
                project='project_id',
                schema='name:string,gender:string,count:integer',
                create_disposition=CREATE_IF_NEEDED,
                write_disposition=WRITE_TRUNCATE)

if __name__ == '__main__':
   logging.getLogger().setLevel(logging.INFO)
   run()

标签： pythongoogle-bigquerygoogle-cloud-dataflowapache-beam

解决方案

您可以使用WriteToText添加.csv后缀和headers. 考虑到您需要将查询结果解析为 CSV 格式。例如，我使用了莎士比亚公共数据集和以下查询：

SELECT word, word_count, corpus FROM `bigquery-public-data.samples.shakespeare` WHERE CHAR_LENGTH(word) > 3 ORDER BY word_count DESC LIMIT 10

我们现在读取查询结果：

BQ_DATA = p | 'read_bq_view' >> beam.io.Read(
    beam.io.BigQuerySource(query=query, use_standard_sql=True))

BQ_DATA现在包含键值对：

{u'corpus': u'hamlet', u'word': u'HAMLET', u'word_count': 407}
{u'corpus': u'kingrichardiii', u'word': u'that', u'word_count': 319}
{u'corpus': u'othello', u'word': u'OTHELLO', u'word_count': 313}

我们可以应用一个beam.Map函数来只产生值：

BQ_VALUES = BQ_DATA | 'read values' >> beam.Map(lambda x: x.values())

摘录BQ_VALUES：

[u'hamlet', u'HAMLET', 407]
[u'kingrichardiii', u'that', 319]
[u'othello', u'OTHELLO', 313]

最后再次映射以使所有列值用逗号而不是列表分隔（考虑到如果双引号可以出现在字段中，则需要转义它们）：

BQ_CSV = BQ_VALUES | 'CSV format' >> beam.Map(
    lambda row: ', '.join(['"'+ str(column) +'"' for column in row]))

现在我们使用后缀和标题将结果写入 GCS：

BQ_CSV | 'Write_to_GCS' >> beam.io.WriteToText(
    'gs://{0}/results/output'.format(BUCKET), file_name_suffix='.csv', header='word, word count, corpus')

书面结果：

$ gsutil cat gs://$BUCKET/results/output-00000-of-00001.csv
word, word count, corpus
"hamlet", "HAMLET", "407"
"kingrichardiii", "that", "319"
"othello", "OTHELLO", "313"
"merrywivesofwindsor", "MISTRESS", "310"
"othello", "IAGO", "299"
"antonyandcleopatra", "ANTONY", "284"
"asyoulikeit", "that", "281"
"antonyandcleopatra", "CLEOPATRA", "274"
"measureforemeasure", "your", "274"
"romeoandjuliet", "that", "270"

推荐阅读