从 S3 读取 ORC 文件到 Pandas

Question

I'm trying to read an orc file from s3 into a Pandas dataframe. In my version of pandas there is no pd.read_orc(...).

I tried to do this:

session = boto3.Session()
s3_client = session.client('s3')

s3_key = "my_object_key"


data = s3_client.get_object(
    Bucket='my_bucket',
    Key=s3_key
)

orc_bytes = data['Body'].read()

Which reads the object as bytes.

Now I try to do this:

orc_data = pyorc.Reader(orc_bytes)

But it fails because:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-11-deaabe8232ce> in <module>
----> 1 data = pyorc.Reader(orc_data)

/anaconda3/envs/linear_opt_3.7/lib/python3.7/site-packages/pyorc/reader.py in __init__(self, fileo, batch_size, column_indices, column_names, struct_repr, converters)
     65             conv = converters
     66         super().__init__(
---> 67             fileo, batch_size, column_indices, column_names, struct_repr, conv
     68         )
     69 

TypeError: Parameter must be a file-like object, but `<class 'bytes'>` was provided

Eventually I would like to land it as.csv or something I can read into pandas. Is there a better way to do this?

Answer 1

Try wrapping the S3 data in an io.BytesIO :

import io

orc_bytes = io.BytesIO(data['Body'].read())
orc_data = pyorc.Reader(orc_bytes)

Answer 2

Here's the function that solves the problem end to end:

import boto3
import pyorc
import io
import pandas as pd

session = boto3.Session()
s3_client = session.client('s3')

def load_s3_orc_to_local_df(key, bucket):
    data = s3_client.get_object(Bucket=bucket, Key=key)
    orc_bytes = io.BytesIO(data['Body'].read())   
    reader = pyorc.Reader(orc_bytes)
    schema = reader.schema
    columns = [item for item in schema.fields]
    rows = [row for row in reader]   
    df = pd.DataFrame(data=rows, columns=columns)
    return df