0

我来到这里是因为我在 Google Cloud 中的计划作业遇到问题。在 Vertex AI Workbench 中，我使用 Python 3 创建了一个笔记本，它创建了一个管道，该管道使用来自公共信用卡数据集的数据来训练 AutoML。如果我在创建结束时运行该作业，则一切正常。但是，如果我按照Job Cloud Scheduler 中的描述安排作业运行，则管道已启用但运行失败。

这是我拥有的代码：

import os
# import sys
import google.cloud.aiplatform as aip
import kfp
# from kfp.v2.dsl import component
from google_cloud_pipeline_components import aiplatform as gcc_aip
from kfp.v2 import compiler
# from kfp.v2.google.client import AIPlatformClient

PROJECT_ID = "fraud-detection-project-329506"
REGION = "us-central1"

credential_path = r"C:\Users\...\fraud-detection-project-329506-4d16889a494a.json"
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = credential_path
      
BUCKET_NAME = "gs://..."
SERVICE_ACCOUNT = "...@fraud-detection-project-329506.iam.gserviceaccount.com"

API_ENDPOINT = "{}-aiplatform.googleapis.com".format(REGION)
PIPELINE_ROOT = "{}/dataset".format(BUCKET_NAME)

aip.init(project=PROJECT_ID, staging_bucket=BUCKET_NAME)

# file names
TRAIN_FILE_NAME = "creditcard_train.csv"
TEST_FILE_NAME = "creditcard_test.csv"

# path for train and test dataset 
gcs_csv_path_train = f"{PIPELINE_ROOT}/{TRAIN_FILE_NAME}"
gcs_csv_path_test = f"{PIPELINE_ROOT}/{TEST_FILE_NAME}"

#gcs location where the output is to be written to
gcs_destination_prefix = "{}/output".format(BUCKET_NAME)

@kfp.dsl.pipeline(name="automl-tab-training-v2")
def pipeline(project: str = PROJECT_ID):
    
    # create tabular dataset
    dataset_create_op = gcc_aip.TabularDatasetCreateOp(
        project=project, display_name="creditcard", gcs_source=gcs_csv_path_train
    )
    
  
    # Training with AutoML
    training_op = gcc_aip.AutoMLTabularTrainingJobRunOp(
        project=project,
        display_name="train-automl-fraud-detection",
        optimization_prediction_type="classification",
        column_transformations=[
            {"numeric": {"column_name": "Time"}},
            {"numeric": {"column_name": "V1"}},
            {"numeric": {"column_name": "V2"}},
            {"numeric": {"column_name": "V3"}},
            {"numeric": {"column_name": "V4"}},
            {"numeric": {"column_name": "V5"}},
            {"numeric": {"column_name": "V6"}},
            {"numeric": {"column_name": "V7"}},
            {"numeric": {"column_name": "V8"}},
            {"numeric": {"column_name": "V9"}},
            {"numeric": {"column_name": "V10"}},
            {"numeric": {"column_name": "V11"}},
            {"numeric": {"column_name": "V12"}},
            {"numeric": {"column_name": "V13"}},
            {"numeric": {"column_name": "V14"}},
            {"numeric": {"column_name": "V15"}},
            {"numeric": {"column_name": "V16"}},
            {"numeric": {"column_name": "V17"}},
            {"numeric": {"column_name": "V18"}},
            {"numeric": {"column_name": "V19"}},
            {"numeric": {"column_name": "V20"}},
            {"numeric": {"column_name": "V21"}},
            {"numeric": {"column_name": "V22"}},
            {"numeric": {"column_name": "V23"}},
            {"numeric": {"column_name": "V24"}},
            {"numeric": {"column_name": "V25"}},
            {"numeric": {"column_name": "V26"}},
            {"numeric": {"column_name": "V27"}},
            {"numeric": {"column_name": "V28"}},
            {"numeric": {"column_name": "Amount"}},
        ],
        dataset=dataset_create_op.outputs["dataset"],#dataset_with_FeatEng,
        target_column="Class",
        budget_milli_node_hours=1000,
    )
    
    # batch prediction after training
    batchprediction_op = gcc_aip.ModelBatchPredictOp(
        model=training_op.outputs["model"],
        job_display_name='prediction1',
        gcs_source=gcs_csv_path_test,
        project=project,
        machine_type="n1-standard-2",
        gcs_destination_prefix=gcs_destination_prefix,
    )
    

COMPILED_PIPELINE_PATH = r"C:\Users\...\tabular_classification_pipeline.json"
SCHEDULE = "5 5 * * *"
DISPLAY_NAME = 'fraud_detection'

# compile pipeline
compiler.Compiler().compile(
    pipeline_func=pipeline,
    package_path=COMPILED_PIPELINE_PATH,
)

# job run after its creation
job = aip.PipelineJob(
    display_name=DISPLAY_NAME,
    template_path=COMPILED_PIPELINE_PATH,
    pipeline_root=PIPELINE_ROOT,
)
job.run()

# api_client = AIPlatformClient(project_id=PROJECT_ID, region=REGION)

# schedule training/prediction every day at a certain hour
# api_client.create_schedule_from_job_spec(
#    job_spec_path=COMPILED_PIPELINE_PATH,
#    pipeline_root=PIPELINE_ROOT,
#    schedule=SCHEDULE,
# )

查看错误日志，我发现：

{
httpRequest: {
status: 404
}
insertId: "13yj575g2rylrz9"
jsonPayload: {
@type: "type.googleapis.com/google.cloud.scheduler.logging.AttemptFinished"
jobName: "projects/fraud-detection-project-329506/locations/us-central1/jobs/pipeline_pipeline_179e648c_0-11-a-a-a"
status: "NOT_FOUND"
targetType: "HTTP"
url: "https://us-central1-fraud-detection-project-329506.cloudfunctions.net/templated_http_request-v1"
}
logName: "projects/fraud-detection-project-329506/logs/cloudscheduler.googleapis.com%2Fexecutions"
receiveTimestamp: "2021-10-19T18:00:00.309225533Z"
resource: {
labels: {
job_id: "pipeline_pipeline_179e648c_0-11-a-a-a"
location: "us-central1"
project_id: "fraud-detection-project-329506"
}
type: "cloud_scheduler_job"
}
severity: "ERROR"
timestamp: "2021-10-19T18:00:00.309225533Z"
}

这是否意味着我必须在运行笔记本之前创建 URL？我不知道该怎么做。先感谢您。

4

1 回答 1

1

从您分享的错误来看，显然Cloud Function无法创建作业。

status: "NOT_FOUND" 
targetType: "HTTP" 
url: "https://us-central1-fraud-detection-project-329506.cloudfunctions.net/templated_http_request-v1"

Cloud Function 方面的一个可能原因可能是之前您的项目中未使用Cloud Build API或者它已被禁用。您可以检查它是否已启用并重试吗？如果您最近启用了此 API，请等待几分钟，让操作传播到系统并重试。

于 2021-11-09T07:41:34.643 回答