来自作曲家错误的 gcp 触发数据流作业

Question

我正在尝试运行使用以下代码从 Composer Airflow DAG 进行数据流作业。

我收到 2 种类型的错误消息，具体取决于代码。

请提出解决方法。

a) 错误 1：当服务帐户电子邮件被评论时 (#)

# "serviceAccountEmail": "service-7276363xxxxx@cloudcomposer-accounts.iam.gserviceaccount.com",

错误：

Error: Required 'compute.subnetworks.get' permission for 'projects/vpc-host/regions/us-central1/subnetworks/sbn-dataflow'

b) 错误 2：使用服务帐户电子邮件时

"serviceAccountEmail": "service-7276363xxxxx@cloudcomposer-accounts.iam.gserviceaccount.com",

错误：

Current user cannot act as service account service-7276363xxxxx@cloudcomposer-accounts.iam.gserviceaccount.com

代码：

import datetime

from airflow import models
from airflow.contrib.operators.dataflow_operator import DataflowTemplateOperator
from airflow.utils.dates import days_ago

bucket_path = models.Variable.get("bucket_path")
project_id = models.Variable.get("project_id")
gce_zone = models.Variable.get("gce_zone")


default_args = {
    "owner": "Airflow",
    "start_date": days_ago(1),
    "depends_on_past": False,
    "dataflow_default_options": {
        "project": project_id,
        "zone": gce_zone,
        "serviceAccountEmail": "service-7276363xxxxx@cloudcomposer-accounts.iam.gserviceaccount.com",
        "subnetwork": "https://www.googleapis.com/compute/v1/projects/vpc-host/regions/us-central1/subnetworks/sbn-dataflow",
        "tempLocation": bucket_path + "/tmp/",
    }
}


with models.DAG(
    dag_id="composer_dataflow_dag",
    default_args=default_args,
    schedule_interval=datetime.timedelta(days=1)
) as dag:
    dataflow_template_job = DataflowTemplateOperator(
        task_id="dataflow_csv_to_bq",
        template="gs://dataflow-templates/latest/GCS_Text_to_BigQuery",
        parameters={
            "javascriptTextTransformFunctionName": "transformCSVtoJSON",
            "javascriptTextTransformGcsPath": bucket_path + "/SCORE_STG.js",
            "JSONPath": bucket_path + "/SCORE_STG.json",
            "inputFilePattern": bucket_path + "/stg_data.csv",
            "outputTable": project_id + ":gcp_stage.SCORE_STG",
            "bigQueryLoadingTemporaryDirectory": bucket_path + "/tmp/",
        },
        dag=dag,
    )

Answer 1

您必须使用不同的服务 accounts.Remember 它必须能够访问资源。这应该可以解决这两个问题。

您可以按照 Role Assignment 中的说明创建一个服务帐户作为工作人员。即：一名工人和一名管理员。

除此之外，我没有发现任何异常情况。连参数都正确传递了。供其他用户参考：

来自作曲家错误的 gcp 触发数据流作业

gcp trigger dataflow job from composer error

google-cloud-platform

google-cloud-dataflow

airflow

google-cloud-composer