PySpark read csv from zip file in s3 with two different file types

Question

I have a zip file with a CSV and a json mapping file in it. I would like to read the csv into a spark data frame and the json mapping file into a dictionary. I have done the latter part doing this:

import boto3

obj = s3.get_object(Bucket='bucket', Key='key')

z = zipfile.ZipFile(io.BytesIO(obj["Body"].read()))

csvjson = json.loads(z.open(files[1]).read().decode('utf-8'))

In general, I would like to do the following to get the df from the csv file:

dfRaw = spark.read \
    .format("text") \
    .option("multiLine","true") \
    .option("inferSchema","false") \
    .option("header","true") \
    .option("ignoreLeadingWhiteSpace","true") \
    .option("ignoreTrailingWhiteSpace","true") \
    .load(z.open(files[0]).read().decode('utf-8'))

However this obviously doesn't work because load() expects a file path, and not the lines itself. How can I read this file from the zip file into a spark data frame?

Answer 1

Since you're manually "unzip" CSV file and get the output as String, you can use parallelize as follow

z = zipfile.ZipFile(io.BytesIO(obj["Body"].read()))
csv = [l.decode('utf-8').replace('\n', '') for l in z.open(files[0]).readlines()]

(spark
    .sparkContext
    .parallelize(csv)
    .toDF(T.StringType())
    .withColumn('value', F.from_csv('value', 'ID int, Trxn_Date string')) # your schema goes here
    .select('value.*')
    .show(10, False)
)

# Output
+----+----------+
|ID  |Trxn_Date |
+----+----------+
|null|Trxn_Date |
|100 |2021-03-24|
|133 |2021-01-22|
+----+----------+