Here is a scenario. There is a huge CSV file on Amazon S3. We need to write a Python function that downloads, reads, and prints the value in a specific column on the standard output (stdout).

Simple Googling will lead us to the answer to this assignment in Stack Overflow. The code should look like something like the following:

import codecs
import csv

import boto3


client = boto3.client("s3")

def read_csv_from_s3(bucket_name, key, column):
    data = client.get_object(Bucket=bucket_name, Key=key)

    for row in csv.DictReader(codecs.getreader("utf-8")(data["Body"])):
        print(row[column])

We will explore the solution above in detail in this article. Imagine this like a rubber duck programming and you are the rubber duck in this case.

Downloading File from S3

Let's get started. First, we need to figure out how to download a file from S3 in Python. The official AWS SDK for Python is known as Boto3. According to the documentation, we can create the client instance for S3 by calling boto3.client("s3"). Then we call the get_object() method on the client with bucket name and key as input arguments to download a specific file.

Now the thing that we are interested in is the return value of the get_object() method call. The return value is a Python dictionary. In the Body key of the dictionary, we can find the content of the file downloaded from S3. The body data["Body"] is a botocore.response.StreamingBody. Hold that thought.

Reading CSV File

Let's switch our focus to handling CSV files. We want to access the value of a specific column one by one. csv.DictReader from the standard library seems to be an excellent candidate for this job. It returns an iterator (the class implements the iterator methods __iter__() and __next__()) that we can use to access each row in a for-loop: row[column]. But what should we pass into X as an argument? According to the documentation, we should refer to the reader instance.

All other optional or keyword arguments are passed to the underlying reader instance.

There we can see that the first argument csvfile

can be any object which supports the iterator protocol and returns a string each time its next() method is called

botocore.response.StreamingBody supports the iterator protocol 🎉.

Unfortunately, it's __next__() method does not return a string but bytes instead.

_csv.Error: iterator should return strings, not bytes (did you open the file in text mode?)

Reading CSV file from S3

So how do we bridge the gap between botocore.response.StreamingBody type and the type required by the cvs module? We want to "convert" the bytes to string in this case. Therefore, the codecs module of Python's standard library seems to be a place to start.

Most standard codecs are text encodings, which encode text to bytes

Since we are doing the opposite, we are looking for a "decoder," specifically a decoder that can handle stream data: codecs.StreamReader

Decodes data from the stream and returns the resulting object.

The codecs.StreamReader takes a file-like object as an input argument. In Python, this means the object should have a read() method. The botocore.response.StreamingBody does have the read() method: https://botocore.amazonaws.com/v1/documentation/api/latest/reference/response.html#botocore.response.StreamingBody.read

Since the codecs.StreamReader also supports the iterator protocol, we can pass the object of this instance into the csv.DictReader: https://github.com/python/cpython/blob/1370d9dd9fbd71e9d3c250c8e6644e0ee6534fca/Lib/codecs.py#L642-L651

The final piece of the puzzle is: How do we create the codecs.StreamReader? That's where the codecs.getreader() function comes in play. We pass the codec of our choice (in this case, utf-8) into the codecs.getreader(), which creates thecodecs.StreamReader. This allows us to read the CSV file row-by-row into dictionary by passing the codec.StreamReader into csv.DictReader:

Thank you for following this long and detailed (maybe too exhausting) explanation of such a short program. I hope you find it useful. Thank your listening ❤️.