流之前BigQuery表被截断不起作用

Question

我们正在使用BigQuery Python API进行一些分析。为此，我们创建了以下适配器：

def stream_data(self, table, data, schema, how=None):
    r = self.connector.tables().list(projectId=self._project_id,
                                     datasetId='lbanor').execute()
    table_exists = [row['tableReference']['tableId'] for row in
                    r['tables'] if
                    row['tableReference']['tableId'] == table]
    if table_exists:
        if how == 'WRITE_TRUNCATE':
            self.connector.tables().delete(projectId=self._project_id,
                                           datasetId='lbanor',
                                           tableId=table).execute()
            body = {
                'tableReference': {
                    'tableId': table,
                    'projectId': self._project_id,
                    'datasetId': 'lbanor'
                },
                'schema': schema
            }
            self.connector.tables().insert(projectId=(
                                           self._project_id),
                                           datasetId='lbanor',
                                           body=body).execute()
    else:
        body = {
            'tableReference': {
                'tableId': table,
                'projectId': self._project_id,
                'datasetId': 'lbanor'
            },
            'schema': schema
        }
        self.connector.tables().insert(projectId=(
                                       self._project_id),
                                       datasetId='lbanor',
                                       body=body).execute()

    body = {
        'rows': [
            {
                'json': data,
                'insertId': str(uuid.uuid4())
            }
        ]
    }
    self.connector.tabledata().insertAll(projectId=(
                                         self._project_id),
                                         datasetId='lbanor',
                                         tableId=table,
                                               body=body).execute(num_retries=5)

哪里connector是构建对象。

其主要目的是将数据流式传输到给定的表。如果该表已经存在并且“如何”输入作为“
WRITE_TRUNCATE”传递，则该表首先被删除并再次创建。之后，继续进行数据流。

当不一次又一次删除表时，一切工作正常。

举例来说，这是结果，当我们没有模拟写截断选项（一运行该脚本for循环不断给你打电话stream_data与how=None）：

[
  {
    "date": "2016-04-25",
    "unix_date": "1461606664981207",
    "init_cv_date": "2016-03-12",
    "end_cv_date": "2016-03-25",
    "days_trained": "56",
    "days_validated": "14",
    "navigated_score": "1",
    "carted_score": "3",
    "purchased_score": "10",
    "description": "First trial of top seller alg. No filter nor any condition is applied. Skus not present in train count as rank=0.5",
    "metric": "rank",
    "result": "0.31729249914663893"
  },
  {
    "date": "2016-04-25",
    "unix_date": "1461606599745107",
    "init_cv_date": "2016-03-06",
    "end_cv_date": "2016-03-25",
    "days_trained": "80",
    "days_validated": "20",
    "navigated_score": "1",
    "carted_score": "3",
    "purchased_score": "10",
    "description": "First trial of top seller alg. No filter nor any condition is applied. Skus not present in train count as rank=0.5",
    "metric": "rank",
    "result": "0.32677143128667446"
  },
  {
    "date": "2016-04-25",
    "unix_date": "1461606688950415",
    "init_cv_date": "2016-03-14",
    "end_cv_date": "2016-03-25",
    "days_trained": "48",
    "days_validated": "12",
    "navigated_score": "1",
    "carted_score": "3",
    "purchased_score": "10",
    "description": "First trial of top seller alg. No filter nor any condition is applied. Skus not present in train count as rank=0.5",
    "metric": "rank",
    "result": "0.3129267723358932"
  },
  {
    "date": "2016-04-25",
    "unix_date": "1461606707195122",
    "init_cv_date": "2016-03-16",
    "end_cv_date": "2016-03-25",
    "days_trained": "40",
    "days_validated": "10",
    "navigated_score": "1",
    "carted_score": "3",
    "purchased_score": "10",
    "description": "First trial of top seller alg. No filter nor any condition is applied. Skus not present in train count as rank=0.5",
    "metric": "rank",
    "result": "0.310620987663015"
  },
  {
    "date": "2016-04-25",
    "unix_date": "1461606622432947",
    "init_cv_date": "2016-03-08",
    "end_cv_date": "2016-03-25",
    "days_trained": "72",
    "days_validated": "18",
    "navigated_score": "1",
    "carted_score": "3",
    "purchased_score": "10",
    "description": "First trial of top seller alg. No filter nor any condition is applied. Skus not present in train count as rank=0.5",
    "metric": "rank",
    "result": "0.32395802949369296"
  }
]

但是，当我们在输入how =“ WRITE_TRUNCATE”中使用相同的适配器时，其行为发生了变化，并且变得不可预测。

有时它可以正常工作，并且数据会保存到表中。但是有时，即使没有出现错误，也没有数据保存到表中。

尝试查询表时，不返回任何数据。它只返回“查询返回零结果”。

删除表，再次创建表并传输数据时出了点问题。我们在犯一些错误吗？

如果您需要更多信息，请告诉我。提前致谢！

Answer 1

参见Jordan Jordan的答案和SeanChen对的评论（均为BigQuery工程师）。

摘要是：

• 重新创建或截断表时，“您需要在流式传输前等待2分钟以上，以避免数据丢失。

这样就可以解释为什么您会得到这种不确定的行为。