Read & Write files from HDFS

Introduction

This is an introduction on how to interact with HDFS. You will find in this article an explanation on how to connect, read and write on HDFS. Please note, that this manipulation will natively work with a python program executed inside Saagie. To connect to Saagie's HDFS outside Saagie platform, you'll need a specific configuration.

For further information, you can find the full documentation of the InsecureClient package with this link:

Full documentation: https://hdfscli.readthedocs.io/en/latest/api.html

Connecting

Connecting with Insecure Client

import pandas as pd
from hdfs import InsecureClient
import os

To connect to HDFS, you need an URL with this format:

http://hdfs_ip:hdfs_port

The WEBHDFS port is by default 50070. You only need to replace the IP address with the HDFS_IP of your platform.

# Connecting to Webhdfs by providing hdfs host ip and webhdfs port (50070 by default)
client_hdfs = InsecureClient('http://hdfs_ip:50070')

We advise to specify a specific user when connecting to HDFS

client_hdfs = InsecureClient('http://hdfs_ip:50070', user='my_user')

Connecting with Kerberos

Kinit

Before connecting to HDFS, you must obtain a Kerberos ticket through a kinit command. In order to do so, you can launch :

• a bash command inside a Terminal in Jupyter which will prompt for your password

kinit myusername

• a bash command inside your Saagie Python job, directly in the command line

echo $MY_USER_PASSWORD | kinit myusername
python {file} arg1 arg2

• directly in your Python code

import os
import subprocess

password = subprocess.Popen(('echo', os.environ['MY_USER_PASSWORD']), stdout=subprocess.PIPE)
subprocess.call(('kinit', os.environ['MY_USER_LOGIN']), stdin=password.stdout)

Connecting to your kerberized cluster

import pandas as pd
from hdfs.ext.kerberos import KerberosClient
import requests

session = requests.Session()
session.verify = False
client = KerberosClient('https://'+os.environ['HDFS_HOSTNAME']+':50470',mutual_auth="REQUIRED",session=session)

Notice the port to access secure HDFS is 50470.

Writing a file on HDFS

# Creating a simple Pandas DataFrame
liste_hello = ['hello1','hello2']
liste_world = ['world1','world2']
df = pd.DataFrame(data = {'hello' : liste_hello, 'world': liste_world})
 
# Writing Dataframe to hdfs
with client_hdfs.write('/user/hdfs/wiki/helloworld.csv', encoding = 'utf-8') as writer:
  df.to_csv(writer)

Reading a file from HDFS

# ====== Reading files ======
with client_hdfs.read('/user/hdfs/wiki/helloworld.csv', encoding = 'utf-8') as reader:
  df = pd.read_csv(reader,index_col=0)