API and command line interface for HDFS.
- Python bindings for the WebHDFS API, supporting both secure and insecure clusters.
- Lightweight CLI with aliases for convenient namenode URL caching.
- Additional functionality through optional extensions:
avro
, allowing reading/writing Avro files directly from JSON.dataframe
, enabling fast loading/saving of pandas dataframes from/to HDFS.kerberos
, adding support for Kerberos authenticated clusters.
Using pip:
$ pip install hdfs
By default none of the package requirements for extensions are installed. To do so simply suffix the package name with the desired extensions:
$ pip install hdfs[avro,dataframe,kerberos]
By default the command line entry point will be named hdfs
. If this conflicts with another utility, you can choose another name by specifying the HDFS_ENTRY_POINT
environment variable:
$ HDFS_ENTRY_POINT=hdfscli pip install hdfs
Sample snippet using a python client to create a file on HDFS, rename it, download it locally, and finally delete the remote copy.
from hdfs import KerberosClient
client = KerberosClient('http://namenode:port', root='/user/alice')
client.write('hello.md', 'Hello, world!')
client.rename('hello.md', 'hello.rst')
client.download('hello.rst', 'hello.rst')
client.delete('hello.rst')
Sample commands (see below for how to configure cluster aliases):
$ hdfs --read logs/1987-03-23 >>logs
$ hdfs --write -o data/weights.tsv <weights.tsv
Cf. hdfs --help
for the full list of commands and options.
You can configure which clusters to connect to by writing your own configuration at ~/.hdfsrc
(this will also enable the Client.from_alias
method).
Sample configuration defining two aliases, foo
and `bar`:
[hdfs]
default.alias = foo # Used when no alias is specified at the command line.
[foo_alias]
client = KerberosClient
root = /some/directory
url = http://url.to.namenode:port
[bar_alias]
url = http://url.to.another.namenode:port
All options other than url
can be omitted. client
determines which class to use (defaulting to the generic Client
), and the remaining options are passed as named arguments to the appropriate constructor.
The full documentation can be found here.