如何使用 python 在 Dataflow 中将字典写入 Bigquery

Question

我正在尝试从 GCP 存储中读取 csv，将其转换为字典，然后写入 Bigquery table，如下所示：

p | ReadFromText("gs://bucket/file.csv") 
  | (beam.ParDo(BuildAdsRecordFn()))
  | WriteToBigQuery('ads_table',dataset='dds',project='doubleclick-2',schema=ads_schema)

其中：'doubleclick-2'和'dds'是现有的项目和数据集，ads_schema定义如下：

ads_schema='Advertiser_ID:INTEGER,Campaign_ID:INTEGER,Ad_ID:INTEGER,Ad_Name:STRING,Click_through_URL:STRING,Ad_Type:STRING'

BuildAdsRecordFn() 定义如下：

class AdsRecord:
  dict = {}

  def __init__(self, line):
    record = line.split(",")
    self.dict['Advertiser_ID'] = record[0]
    self.dict['Campaign_ID'] = record[1]
    self.dict['Ad_ID'] = record[2]
    self.dict['Ad_Name'] = record[3]
    self.dict['Click_through_URL'] = record[4]
    self.dict['Ad_Type'] = record[5]


class BuildAdsRecordFn(beam.DoFn):
  def __init__(self):
    super(BuildAdsRecordFn, self).__init__()

  def process(self, element):
    text_line = element.strip()
    ads_record = AdsRecord(text_line).dict
    return ads_record

但是，当我运行管道时，我得到了以下错误：

"dataflow_job_18146703755411620105-B" failed., (6c011965a92e74fa): BigQuery job "dataflow_job_18146703755411620105-B" in project "doubleclick-2" finished with error(s): errorResult: JSON table encountered too many errors, giving up. Rows: 1; errors: 1., error: JSON table encountered too many errors, giving up. Rows: 1; errors: 1., error: JSON parsing error in row starting at position 0: Value encountered without start of object

这里是我使用的样本测试数据：

100001,1000011,10000111,ut,https://bloomberg.com/aliquam/lacus/morbi.xml,Brand-neutral
100001,1000011,10000112,eu,http://weebly.com/sed/vel/enim/sit.jsp,Dynamic Click

我是 Dataflow 和 python 的新手，所以无法弄清楚上面的代码可能有什么问题。非常感谢任何帮助！

Answer 1

我刚刚实现了您的代码，但效果不佳，但我收到了一条不同的消息错误（类似于 "you can't return a dict as the result of a ParDo"）。

这段代码对我来说正常工作，请注意，不仅我没有使用 class 属性 dict，而且现在返回了一个列表：

ads_schema='Advertiser_ID:INTEGER,Campaign_ID:INTEGER,Ad_ID:INTEGER,Ad_Name:STRING,Click_through_URL:STRING,Ad_Type:STRING'

class BuildAdsRecordFn(beam.DoFn):
    def __init__(self):
      super(BuildAdsRecordFn, self).__init__()

    def process(self, element):
      text_line = element.strip()
      ads_record = self.process_row(element)      
      return ads_record

    def process_row(self, row):
        dict_ = {}

        record = row.split(",")
        dict_['Advertiser_ID'] = int(record[0]) if record[0] else None
        dict_['Campaign_ID'] = int(record[1]) if record[1] else None
        dict_['Ad_ID'] = int(record[2]) if record[2] else None
        dict_['Ad_Name'] = record[3]
        dict_['Click_through_URL'] = record[4]
        dict_['Ad_Type'] = record[5]
        return [dict_]

with beam.Pipeline() as p:

    (p | ReadFromText("gs://bucket/file.csv")
       | beam.Filter(lambda x: x[0] != 'A')
       | (beam.ParDo(BuildAdsRecordFn()))
       | WriteToBigQuery('ads_table', dataset='dds',
           project='doubleclick-2', schema=ads_schema))
      #| WriteToText('test.csv'))

这是我模拟的数据：

Advertiser_ID,Campaign_ID,Ad_ID,Ad_Name,Click_through_URL,Ad_Type
1,1,1,name of ad,www.url.com,sales
1,1,2,name of ad2,www.url2.com,sales with sales

我还过滤掉了我在我的文件中创建的 header 行（在 Filter 操作中），如果你没有 header 那么这不是必需的

如何使用 python 在 Dataflow 中将字典写入 Bigquery

How to write dictionaries to Bigquery in Dataflow using python

python

dataflow

google-bigquery

gcp