我在气流中创建了一个流程,我需要每10分钟从SQL Server数据库中导出一个新文件并播放到BigQuery!生成的文件是一个csv,该文件自动包含处理日期为YYYYMMDDHHMMSS格式的文件名。
当我从步骤1(导出)转到步骤2(插入BigQuery)时,气流中继再次每个脚本都会更改文件名变量名,并且处理日期与步骤1不同!
示例: 步骤1:test_20190624113656.csv 步骤2:test_20190624113705.csv
在这种情况下,我想在第一步中保留文件名。
nm_arquivo = 'test_' + datetime.today().strftime('%Y%m%d%H%M%S') + '.csv'
def insert_bigquery(ds, **kwargs):
bigquery_client = bigquery.Client(project="project_name")
dataset_ref = bigquery_client.dataset('test_dataset')
job_config = bigquery.LoadJobConfig()
job_config.schema = [
bigquery.SchemaField('id','INTEGER',mode='REQUIRED'),
bigquery.SchemaField('sigla','STRING',mode='REQUIRED'),
bigquery.SchemaField('nome_en','STRING',mode='REQUIRED'),
bigquery.SchemaField('nome_pt','STRING',mode='REQUIRED'),
]
job_config.source_format = bigquery.SourceFormat.CSV
time_partitioning = bigquery.table.TimePartitioning()
job_config.time_partitioning = time_partitioning
job_config.clustering_fields = ["id", "sigla"]
uri = "gs://bucket_name/"+nm_arquivo
load_job = bigquery_client.load_table_from_uri(
uri,
dataset_ref.table('bdb'),
job_config=job_config
)
print('Starting job {}'.format(load_job.job_id))
load_job.result()
print('Job finished.')
#step1
import_orders_op = MsSqlToGoogleCloudStorageOperator(
task_id='import_orders',
mssql_conn_id='mssql_conn',
google_cloud_storage_conn_id='gcp_conn',
sql="""select * from bdb""",
bucket='bucket_name',
filename=nm_arquivo,
dag=dag)
#step2
run_this = PythonOperator(
task_id='insert_bigquery',
provide_context=True,
python_callable=insert_bigquery,
dag=dag,
)
run_this.set_upstream(import_orders_op)
答案 0 :(得分:2)
您应该使用DAG的执行时间。
您可以使用{{ ts_nodash }}
Airflow宏。它格式化execution_date.isoformat()
(例如:2018-01-01T00:00:00+00:00
)以删除-
和:
,例如:20180101T000000
。可以在任何模板化参数中使用此宏。
有关更多信息和所有其他可用变量的列表:
答案 1 :(得分:0)
您可以使用文件存储文件名:
import pickle
nm_arquivo = 'test_' + datetime.today().strftime('%Y%m%d%H%M%S') + '.csv'
#step 1
with open('filename.pickle', 'wb') as handle:
pickle.dump(nm_arquivo, handle)
#step 2
with open('filename.pickle', 'rb') as handle:
nm_arquivo = pickle.load(handle)