How do I keep ETL job script from executing when there is no data

Question

I have a Glue ETL script which I will eventually have execute via Lambda with S3 event trigger.  But for now I am just running regularly.  If there is no new data, then the script runs and produces errors since my transformations have empty DataFrames.

I am using Bookmarks.  My ingest and export sections are using transformation contexts.  But my regular transformation code is not wrapped inside any sort of if-clause or anything like that.

So my code looks like this:

```
import sys
from awsglue.transforms import *
from awsglue.utils import getResolvedOptions
from pyspark.context import SparkContext
from awsglue.context import GlueContext
from awsglue.job import Job
from pyspark.sql.functions import to_timestamp, to_date, date_format
from awsglue.dynamicframe import DynamicFrame
import pyspark.sql.functions as f

## @params: [JOB_NAME]
args = getResolvedOptions(sys.argv, ['JOB_NAME'])

sc = SparkContext()
glueContext = GlueContext(sc)
spark = glueContext.spark_session
job = Job(glueContext)
job.init(args['JOB_NAME'], args)

# Data Catalog: database and table name
db_name = "my-database"
tbl_name = "my-table"

# Script generated for node S3 bucket
S3bucket_node1 = glueContext.create_dynamic_frame.from_options(
    format_options={"quoteChar": '"', "withHeader": True, "separator": ","},
    connection_type="s3",
    format="csv",
    connection_options={"paths": ["s3://my-bucket/my-folder"], "recurse": True},
    transformation_ctx="S3bucket_node1",
)

df = S3bucket_node1.toDF()
df = df.drop("col6")

df = df.withColumn("time", f.unix_timestamp("time", 'dd-MM-yyyy HH:mm:ss') * 1000)
df = df.withColumn("timegmt", f.unix_timestamp("timegmt", 'dd-MM-yyyy HH:mm:ss') * 1000)
df = df.withColumn("value", df.VALUE.cast('int'))
df = df.withColumn("filename", f.split(f.input_file_name(), '/')[4])
df = df.withColumnRenamed('VID', 'vid')
df = df.withColumnRenamed('ALTID', 'altid')
df = df.withColumnRenamed('VTYPE', 'vtype')

## SPARK DF -> DYNAMIC DF
dynamic_df = DynamicFrame.fromDF(
    df,
    glueContext,
    "convert_ctx"
)

# Data Catalog WRITE
DataCatalogtable_node2 = glueContext.write_dynamic_frame.from_catalog(
    frame = dynamic_df,
    database = db_name,
    table_name = tbl_name,
    transformation_ctx = "DataCatalogtable_node2",
)

job.commit()
```

so I am trying to put logic so that it doesn't try to execute all those transformations if there is in fact no new data (i.e. empty DataFrame).

Accepted Answer

I don't love this method because it causes overhead but I'm not sure there is a better way: before you convert back to the dynamic frame add something that checks to see if the dataframe is empty. Another common way is to count the number of rows in the dataframe and if it is not zero then proceed, however, the take method is more performant as it evaluates less partitions/rows.

```
if df.take(1):

job.commit()
```

How do I keep ETL job script from executing when there is no data

相關內容